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>
27 #include <net/netns/hash.h>
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/netfilter/nf_conntrack_core.h>
31 #include <net/netfilter/nf_conntrack_expect.h>
32 #include <net/netfilter/nf_conntrack_helper.h>
33 #include <net/netfilter/nf_conntrack_tuple.h>
34 #include <net/netfilter/nf_conntrack_zones.h>
36 unsigned int nf_ct_expect_hsize __read_mostly;
37 EXPORT_SYMBOL_GPL(nf_ct_expect_hsize);
39 struct hlist_head *nf_ct_expect_hash __read_mostly;
40 EXPORT_SYMBOL_GPL(nf_ct_expect_hash);
42 unsigned int nf_ct_expect_max __read_mostly;
44 static struct kmem_cache *nf_ct_expect_cachep __read_mostly;
45 static unsigned int nf_ct_expect_hashrnd __read_mostly;
47 /* nf_conntrack_expect helper functions */
48 void nf_ct_unlink_expect_report(struct nf_conntrack_expect *exp,
49 u32 portid, int report)
51 struct nf_conn_help *master_help = nfct_help(exp->master);
52 struct net *net = nf_ct_exp_net(exp);
54 WARN_ON(!master_help);
55 WARN_ON(timer_pending(&exp->timeout));
57 hlist_del_rcu(&exp->hnode);
58 net->ct.expect_count--;
60 hlist_del_rcu(&exp->lnode);
61 master_help->expecting[exp->class]--;
63 nf_ct_expect_event_report(IPEXP_DESTROY, exp, portid, report);
64 nf_ct_expect_put(exp);
66 NF_CT_STAT_INC(net, expect_delete);
68 EXPORT_SYMBOL_GPL(nf_ct_unlink_expect_report);
70 static void nf_ct_expectation_timed_out(unsigned long ul_expect)
72 struct nf_conntrack_expect *exp = (void *)ul_expect;
74 spin_lock_bh(&nf_conntrack_expect_lock);
75 nf_ct_unlink_expect(exp);
76 spin_unlock_bh(&nf_conntrack_expect_lock);
77 nf_ct_expect_put(exp);
80 static unsigned int nf_ct_expect_dst_hash(const struct net *n, const struct nf_conntrack_tuple *tuple)
82 unsigned int hash, seed;
84 get_random_once(&nf_ct_expect_hashrnd, sizeof(nf_ct_expect_hashrnd));
86 seed = nf_ct_expect_hashrnd ^ net_hash_mix(n);
88 hash = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
89 (((tuple->dst.protonum ^ tuple->src.l3num) << 16) |
90 (__force __u16)tuple->dst.u.all) ^ seed);
92 return reciprocal_scale(hash, nf_ct_expect_hsize);
96 nf_ct_exp_equal(const struct nf_conntrack_tuple *tuple,
97 const struct nf_conntrack_expect *i,
98 const struct nf_conntrack_zone *zone,
99 const struct net *net)
101 return nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
102 net_eq(net, nf_ct_net(i->master)) &&
103 nf_ct_zone_equal_any(i->master, zone);
106 bool nf_ct_remove_expect(struct nf_conntrack_expect *exp)
108 if (del_timer(&exp->timeout)) {
109 nf_ct_unlink_expect(exp);
110 nf_ct_expect_put(exp);
115 EXPORT_SYMBOL_GPL(nf_ct_remove_expect);
117 struct nf_conntrack_expect *
118 __nf_ct_expect_find(struct net *net,
119 const struct nf_conntrack_zone *zone,
120 const struct nf_conntrack_tuple *tuple)
122 struct nf_conntrack_expect *i;
125 if (!net->ct.expect_count)
128 h = nf_ct_expect_dst_hash(net, tuple);
129 hlist_for_each_entry_rcu(i, &nf_ct_expect_hash[h], hnode) {
130 if (nf_ct_exp_equal(tuple, i, zone, net))
135 EXPORT_SYMBOL_GPL(__nf_ct_expect_find);
137 /* Just find a expectation corresponding to a tuple. */
138 struct nf_conntrack_expect *
139 nf_ct_expect_find_get(struct net *net,
140 const struct nf_conntrack_zone *zone,
141 const struct nf_conntrack_tuple *tuple)
143 struct nf_conntrack_expect *i;
146 i = __nf_ct_expect_find(net, zone, tuple);
147 if (i && !refcount_inc_not_zero(&i->use))
153 EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
155 /* If an expectation for this connection is found, it gets delete from
156 * global list then returned. */
157 struct nf_conntrack_expect *
158 nf_ct_find_expectation(struct net *net,
159 const struct nf_conntrack_zone *zone,
160 const struct nf_conntrack_tuple *tuple)
162 struct nf_conntrack_expect *i, *exp = NULL;
165 if (!net->ct.expect_count)
168 h = nf_ct_expect_dst_hash(net, tuple);
169 hlist_for_each_entry(i, &nf_ct_expect_hash[h], hnode) {
170 if (!(i->flags & NF_CT_EXPECT_INACTIVE) &&
171 nf_ct_exp_equal(tuple, i, zone, net)) {
179 /* If master is not in hash table yet (ie. packet hasn't left
180 this machine yet), how can other end know about expected?
181 Hence these are not the droids you are looking for (if
182 master ct never got confirmed, we'd hold a reference to it
183 and weird things would happen to future packets). */
184 if (!nf_ct_is_confirmed(exp->master))
187 /* Avoid race with other CPUs, that for exp->master ct, is
188 * about to invoke ->destroy(), or nf_ct_delete() via timeout
191 * The atomic_inc_not_zero() check tells: If that fails, we
192 * know that the ct is being destroyed. If it succeeds, we
193 * can be sure the ct cannot disappear underneath.
195 if (unlikely(nf_ct_is_dying(exp->master) ||
196 !atomic_inc_not_zero(&exp->master->ct_general.use)))
199 if (exp->flags & NF_CT_EXPECT_PERMANENT) {
200 refcount_inc(&exp->use);
202 } else if (del_timer(&exp->timeout)) {
203 nf_ct_unlink_expect(exp);
206 /* Undo exp->master refcnt increase, if del_timer() failed */
207 nf_ct_put(exp->master);
212 /* delete all expectations for this conntrack */
213 void nf_ct_remove_expectations(struct nf_conn *ct)
215 struct nf_conn_help *help = nfct_help(ct);
216 struct nf_conntrack_expect *exp;
217 struct hlist_node *next;
219 /* Optimization: most connection never expect any others. */
223 spin_lock_bh(&nf_conntrack_expect_lock);
224 hlist_for_each_entry_safe(exp, next, &help->expectations, lnode) {
225 nf_ct_remove_expect(exp);
227 spin_unlock_bh(&nf_conntrack_expect_lock);
229 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
231 /* Would two expected things clash? */
232 static inline int expect_clash(const struct nf_conntrack_expect *a,
233 const struct nf_conntrack_expect *b)
235 /* Part covered by intersection of masks must be unequal,
236 otherwise they clash */
237 struct nf_conntrack_tuple_mask intersect_mask;
240 intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
242 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
243 intersect_mask.src.u3.all[count] =
244 a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
247 return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask) &&
248 net_eq(nf_ct_net(a->master), nf_ct_net(b->master)) &&
249 nf_ct_zone_equal_any(a->master, nf_ct_zone(b->master));
252 static inline int expect_matches(const struct nf_conntrack_expect *a,
253 const struct nf_conntrack_expect *b)
255 return a->master == b->master && a->class == b->class &&
256 nf_ct_tuple_equal(&a->tuple, &b->tuple) &&
257 nf_ct_tuple_mask_equal(&a->mask, &b->mask) &&
258 net_eq(nf_ct_net(a->master), nf_ct_net(b->master)) &&
259 nf_ct_zone_equal_any(a->master, nf_ct_zone(b->master));
262 /* Generally a bad idea to call this: could have matched already. */
263 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
265 spin_lock_bh(&nf_conntrack_expect_lock);
266 nf_ct_remove_expect(exp);
267 spin_unlock_bh(&nf_conntrack_expect_lock);
269 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
271 /* We don't increase the master conntrack refcount for non-fulfilled
272 * conntracks. During the conntrack destruction, the expectations are
273 * always killed before the conntrack itself */
274 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
276 struct nf_conntrack_expect *new;
278 new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
283 refcount_set(&new->use, 1);
286 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
288 void nf_ct_expect_init(struct nf_conntrack_expect *exp, unsigned int class,
290 const union nf_inet_addr *saddr,
291 const union nf_inet_addr *daddr,
292 u_int8_t proto, const __be16 *src, const __be16 *dst)
296 if (family == AF_INET)
303 exp->expectfn = NULL;
305 exp->tuple.src.l3num = family;
306 exp->tuple.dst.protonum = proto;
309 memcpy(&exp->tuple.src.u3, saddr, len);
310 if (sizeof(exp->tuple.src.u3) > len)
311 /* address needs to be cleared for nf_ct_tuple_equal */
312 memset((void *)&exp->tuple.src.u3 + len, 0x00,
313 sizeof(exp->tuple.src.u3) - len);
314 memset(&exp->mask.src.u3, 0xFF, len);
315 if (sizeof(exp->mask.src.u3) > len)
316 memset((void *)&exp->mask.src.u3 + len, 0x00,
317 sizeof(exp->mask.src.u3) - len);
319 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
320 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
324 exp->tuple.src.u.all = *src;
325 exp->mask.src.u.all = htons(0xFFFF);
327 exp->tuple.src.u.all = 0;
328 exp->mask.src.u.all = 0;
331 memcpy(&exp->tuple.dst.u3, daddr, len);
332 if (sizeof(exp->tuple.dst.u3) > len)
333 /* address needs to be cleared for nf_ct_tuple_equal */
334 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
335 sizeof(exp->tuple.dst.u3) - len);
337 exp->tuple.dst.u.all = *dst;
339 #ifdef CONFIG_NF_NAT_NEEDED
340 memset(&exp->saved_addr, 0, sizeof(exp->saved_addr));
341 memset(&exp->saved_proto, 0, sizeof(exp->saved_proto));
344 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
346 static void nf_ct_expect_free_rcu(struct rcu_head *head)
348 struct nf_conntrack_expect *exp;
350 exp = container_of(head, struct nf_conntrack_expect, rcu);
351 kmem_cache_free(nf_ct_expect_cachep, exp);
354 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
356 if (refcount_dec_and_test(&exp->use))
357 call_rcu(&exp->rcu, nf_ct_expect_free_rcu);
359 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
361 static void nf_ct_expect_insert(struct nf_conntrack_expect *exp)
363 struct nf_conn_help *master_help = nfct_help(exp->master);
364 struct nf_conntrack_helper *helper;
365 struct net *net = nf_ct_exp_net(exp);
366 unsigned int h = nf_ct_expect_dst_hash(net, &exp->tuple);
368 /* two references : one for hash insert, one for the timer */
369 refcount_add(2, &exp->use);
371 setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
373 helper = rcu_dereference_protected(master_help->helper,
374 lockdep_is_held(&nf_conntrack_expect_lock));
376 exp->timeout.expires = jiffies +
377 helper->expect_policy[exp->class].timeout * HZ;
379 add_timer(&exp->timeout);
381 hlist_add_head_rcu(&exp->lnode, &master_help->expectations);
382 master_help->expecting[exp->class]++;
384 hlist_add_head_rcu(&exp->hnode, &nf_ct_expect_hash[h]);
385 net->ct.expect_count++;
387 NF_CT_STAT_INC(net, expect_create);
390 /* Race with expectations being used means we could have none to find; OK. */
391 static void evict_oldest_expect(struct nf_conn *master,
392 struct nf_conntrack_expect *new)
394 struct nf_conn_help *master_help = nfct_help(master);
395 struct nf_conntrack_expect *exp, *last = NULL;
397 hlist_for_each_entry(exp, &master_help->expectations, lnode) {
398 if (exp->class == new->class)
403 nf_ct_remove_expect(last);
406 static inline int __nf_ct_expect_check(struct nf_conntrack_expect *expect)
408 const struct nf_conntrack_expect_policy *p;
409 struct nf_conntrack_expect *i;
410 struct nf_conn *master = expect->master;
411 struct nf_conn_help *master_help = nfct_help(master);
412 struct nf_conntrack_helper *helper;
413 struct net *net = nf_ct_exp_net(expect);
414 struct hlist_node *next;
422 h = nf_ct_expect_dst_hash(net, &expect->tuple);
423 hlist_for_each_entry_safe(i, next, &nf_ct_expect_hash[h], hnode) {
424 if (expect_matches(i, expect)) {
425 if (nf_ct_remove_expect(i))
427 } else if (expect_clash(i, expect)) {
432 /* Will be over limit? */
433 helper = rcu_dereference_protected(master_help->helper,
434 lockdep_is_held(&nf_conntrack_expect_lock));
436 p = &helper->expect_policy[expect->class];
437 if (p->max_expected &&
438 master_help->expecting[expect->class] >= p->max_expected) {
439 evict_oldest_expect(master, expect);
440 if (master_help->expecting[expect->class]
441 >= p->max_expected) {
448 if (net->ct.expect_count >= nf_ct_expect_max) {
449 net_warn_ratelimited("nf_conntrack: expectation table full\n");
456 int nf_ct_expect_related_report(struct nf_conntrack_expect *expect,
457 u32 portid, int report)
461 spin_lock_bh(&nf_conntrack_expect_lock);
462 ret = __nf_ct_expect_check(expect);
466 nf_ct_expect_insert(expect);
468 spin_unlock_bh(&nf_conntrack_expect_lock);
469 nf_ct_expect_event_report(IPEXP_NEW, expect, portid, report);
472 spin_unlock_bh(&nf_conntrack_expect_lock);
475 EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
477 void nf_ct_expect_iterate_destroy(bool (*iter)(struct nf_conntrack_expect *e, void *data),
480 struct nf_conntrack_expect *exp;
481 const struct hlist_node *next;
484 spin_lock_bh(&nf_conntrack_expect_lock);
486 for (i = 0; i < nf_ct_expect_hsize; i++) {
487 hlist_for_each_entry_safe(exp, next,
488 &nf_ct_expect_hash[i],
490 if (iter(exp, data) && del_timer(&exp->timeout)) {
491 nf_ct_unlink_expect(exp);
492 nf_ct_expect_put(exp);
497 spin_unlock_bh(&nf_conntrack_expect_lock);
499 EXPORT_SYMBOL_GPL(nf_ct_expect_iterate_destroy);
501 void nf_ct_expect_iterate_net(struct net *net,
502 bool (*iter)(struct nf_conntrack_expect *e, void *data),
504 u32 portid, int report)
506 struct nf_conntrack_expect *exp;
507 const struct hlist_node *next;
510 spin_lock_bh(&nf_conntrack_expect_lock);
512 for (i = 0; i < nf_ct_expect_hsize; i++) {
513 hlist_for_each_entry_safe(exp, next,
514 &nf_ct_expect_hash[i],
517 if (!net_eq(nf_ct_exp_net(exp), net))
520 if (iter(exp, data) && del_timer(&exp->timeout)) {
521 nf_ct_unlink_expect_report(exp, portid, report);
522 nf_ct_expect_put(exp);
527 spin_unlock_bh(&nf_conntrack_expect_lock);
529 EXPORT_SYMBOL_GPL(nf_ct_expect_iterate_net);
531 #ifdef CONFIG_NF_CONNTRACK_PROCFS
532 struct ct_expect_iter_state {
533 struct seq_net_private p;
537 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
539 struct ct_expect_iter_state *st = seq->private;
540 struct hlist_node *n;
542 for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
543 n = rcu_dereference(hlist_first_rcu(&nf_ct_expect_hash[st->bucket]));
550 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
551 struct hlist_node *head)
553 struct ct_expect_iter_state *st = seq->private;
555 head = rcu_dereference(hlist_next_rcu(head));
556 while (head == NULL) {
557 if (++st->bucket >= nf_ct_expect_hsize)
559 head = rcu_dereference(hlist_first_rcu(&nf_ct_expect_hash[st->bucket]));
564 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
566 struct hlist_node *head = ct_expect_get_first(seq);
569 while (pos && (head = ct_expect_get_next(seq, head)))
571 return pos ? NULL : head;
574 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
578 return ct_expect_get_idx(seq, *pos);
581 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
584 return ct_expect_get_next(seq, v);
587 static void exp_seq_stop(struct seq_file *seq, void *v)
593 static int exp_seq_show(struct seq_file *s, void *v)
595 struct nf_conntrack_expect *expect;
596 struct nf_conntrack_helper *helper;
597 struct hlist_node *n = v;
600 expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
602 if (expect->timeout.function)
603 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
604 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
607 seq_printf(s, "l3proto = %u proto=%u ",
608 expect->tuple.src.l3num,
609 expect->tuple.dst.protonum);
610 print_tuple(s, &expect->tuple,
611 __nf_ct_l3proto_find(expect->tuple.src.l3num),
612 __nf_ct_l4proto_find(expect->tuple.src.l3num,
613 expect->tuple.dst.protonum));
615 if (expect->flags & NF_CT_EXPECT_PERMANENT) {
616 seq_puts(s, "PERMANENT");
619 if (expect->flags & NF_CT_EXPECT_INACTIVE) {
620 seq_printf(s, "%sINACTIVE", delim);
623 if (expect->flags & NF_CT_EXPECT_USERSPACE)
624 seq_printf(s, "%sUSERSPACE", delim);
626 helper = rcu_dereference(nfct_help(expect->master)->helper);
628 seq_printf(s, "%s%s", expect->flags ? " " : "", helper->name);
629 if (helper->expect_policy[expect->class].name[0])
631 helper->expect_policy[expect->class].name);
639 static const struct seq_operations exp_seq_ops = {
640 .start = exp_seq_start,
641 .next = exp_seq_next,
642 .stop = exp_seq_stop,
646 static int exp_open(struct inode *inode, struct file *file)
648 return seq_open_net(inode, file, &exp_seq_ops,
649 sizeof(struct ct_expect_iter_state));
652 static const struct file_operations exp_file_ops = {
653 .owner = THIS_MODULE,
657 .release = seq_release_net,
659 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
661 static int exp_proc_init(struct net *net)
663 #ifdef CONFIG_NF_CONNTRACK_PROCFS
664 struct proc_dir_entry *proc;
668 proc = proc_create("nf_conntrack_expect", 0440, net->proc_net,
673 root_uid = make_kuid(net->user_ns, 0);
674 root_gid = make_kgid(net->user_ns, 0);
675 if (uid_valid(root_uid) && gid_valid(root_gid))
676 proc_set_user(proc, root_uid, root_gid);
677 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
681 static void exp_proc_remove(struct net *net)
683 #ifdef CONFIG_NF_CONNTRACK_PROCFS
684 remove_proc_entry("nf_conntrack_expect", net->proc_net);
685 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
688 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
690 int nf_conntrack_expect_pernet_init(struct net *net)
692 net->ct.expect_count = 0;
693 return exp_proc_init(net);
696 void nf_conntrack_expect_pernet_fini(struct net *net)
698 exp_proc_remove(net);
701 int nf_conntrack_expect_init(void)
703 if (!nf_ct_expect_hsize) {
704 nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
705 if (!nf_ct_expect_hsize)
706 nf_ct_expect_hsize = 1;
708 nf_ct_expect_max = nf_ct_expect_hsize * 4;
709 nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
710 sizeof(struct nf_conntrack_expect),
712 if (!nf_ct_expect_cachep)
715 nf_ct_expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize, 0);
716 if (!nf_ct_expect_hash) {
717 kmem_cache_destroy(nf_ct_expect_cachep);
724 void nf_conntrack_expect_fini(void)
726 rcu_barrier(); /* Wait for call_rcu() before destroy */
727 kmem_cache_destroy(nf_ct_expect_cachep);
728 nf_ct_free_hashtable(nf_ct_expect_hash, nf_ct_expect_hsize);