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 <linux/moduleparam.h>
24 #include <linux/export.h>
25 #include <net/net_namespace.h>
27 #include <net/netfilter/nf_conntrack.h>
28 #include <net/netfilter/nf_conntrack_core.h>
29 #include <net/netfilter/nf_conntrack_expect.h>
30 #include <net/netfilter/nf_conntrack_helper.h>
31 #include <net/netfilter/nf_conntrack_tuple.h>
32 #include <net/netfilter/nf_conntrack_zones.h>
34 unsigned int nf_ct_expect_hsize __read_mostly;
35 EXPORT_SYMBOL_GPL(nf_ct_expect_hsize);
37 unsigned int nf_ct_expect_max __read_mostly;
39 static struct kmem_cache *nf_ct_expect_cachep __read_mostly;
41 static HLIST_HEAD(nf_ct_userspace_expect_list);
43 /* nf_conntrack_expect helper functions */
44 void nf_ct_unlink_expect_report(struct nf_conntrack_expect *exp,
47 struct nf_conn_help *master_help = nfct_help(exp->master);
48 struct net *net = nf_ct_exp_net(exp);
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 if (!(exp->flags & NF_CT_EXPECT_USERSPACE))
57 master_help->expecting[exp->class]--;
59 nf_ct_expect_event_report(IPEXP_DESTROY, exp, pid, report);
60 nf_ct_expect_put(exp);
62 NF_CT_STAT_INC(net, expect_delete);
64 EXPORT_SYMBOL_GPL(nf_ct_unlink_expect_report);
66 static void nf_ct_expectation_timed_out(unsigned long ul_expect)
68 struct nf_conntrack_expect *exp = (void *)ul_expect;
70 spin_lock_bh(&nf_conntrack_lock);
71 nf_ct_unlink_expect(exp);
72 spin_unlock_bh(&nf_conntrack_lock);
73 nf_ct_expect_put(exp);
76 static unsigned int nf_ct_expect_dst_hash(const struct nf_conntrack_tuple *tuple)
80 if (unlikely(!nf_conntrack_hash_rnd)) {
81 init_nf_conntrack_hash_rnd();
84 hash = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
85 (((tuple->dst.protonum ^ tuple->src.l3num) << 16) |
86 (__force __u16)tuple->dst.u.all) ^ nf_conntrack_hash_rnd);
87 return ((u64)hash * nf_ct_expect_hsize) >> 32;
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;
98 if (!net->ct.expect_count)
101 h = nf_ct_expect_dst_hash(tuple);
102 hlist_for_each_entry_rcu(i, n, &net->ct.expect_hash[h], hnode) {
103 if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
104 nf_ct_zone(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, u16 zone,
114 const struct nf_conntrack_tuple *tuple)
116 struct nf_conntrack_expect *i;
119 i = __nf_ct_expect_find(net, zone, tuple);
120 if (i && !atomic_inc_not_zero(&i->use))
126 EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
128 /* If an expectation for this connection is found, it gets delete from
129 * global list then returned. */
130 struct nf_conntrack_expect *
131 nf_ct_find_expectation(struct net *net, u16 zone,
132 const struct nf_conntrack_tuple *tuple)
134 struct nf_conntrack_expect *i, *exp = NULL;
135 struct hlist_node *n;
138 if (!net->ct.expect_count)
141 h = nf_ct_expect_dst_hash(tuple);
142 hlist_for_each_entry(i, n, &net->ct.expect_hash[h], hnode) {
143 if (!(i->flags & NF_CT_EXPECT_INACTIVE) &&
144 nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
145 nf_ct_zone(i->master) == zone) {
153 /* If master is not in hash table yet (ie. packet hasn't left
154 this machine yet), how can other end know about expected?
155 Hence these are not the droids you are looking for (if
156 master ct never got confirmed, we'd hold a reference to it
157 and weird things would happen to future packets). */
158 if (!nf_ct_is_confirmed(exp->master))
161 if (exp->flags & NF_CT_EXPECT_PERMANENT) {
162 atomic_inc(&exp->use);
164 } else if (del_timer(&exp->timeout)) {
165 nf_ct_unlink_expect(exp);
172 /* delete all expectations for this conntrack */
173 void nf_ct_remove_expectations(struct nf_conn *ct)
175 struct nf_conn_help *help = nfct_help(ct);
176 struct nf_conntrack_expect *exp;
177 struct hlist_node *n, *next;
179 /* Optimization: most connection never expect any others. */
183 hlist_for_each_entry_safe(exp, n, next, &help->expectations, lnode) {
184 if (del_timer(&exp->timeout)) {
185 nf_ct_unlink_expect(exp);
186 nf_ct_expect_put(exp);
190 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
192 /* Would two expected things clash? */
193 static inline int expect_clash(const struct nf_conntrack_expect *a,
194 const struct nf_conntrack_expect *b)
196 /* Part covered by intersection of masks must be unequal,
197 otherwise they clash */
198 struct nf_conntrack_tuple_mask intersect_mask;
201 intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
203 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
204 intersect_mask.src.u3.all[count] =
205 a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
208 return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask);
211 static inline int expect_matches(const struct nf_conntrack_expect *a,
212 const struct nf_conntrack_expect *b)
214 return a->master == b->master && a->class == b->class &&
215 nf_ct_tuple_equal(&a->tuple, &b->tuple) &&
216 nf_ct_tuple_mask_equal(&a->mask, &b->mask) &&
217 nf_ct_zone(a->master) == nf_ct_zone(b->master);
220 /* Generally a bad idea to call this: could have matched already. */
221 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
223 spin_lock_bh(&nf_conntrack_lock);
224 if (del_timer(&exp->timeout)) {
225 nf_ct_unlink_expect(exp);
226 nf_ct_expect_put(exp);
228 spin_unlock_bh(&nf_conntrack_lock);
230 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
232 /* We don't increase the master conntrack refcount for non-fulfilled
233 * conntracks. During the conntrack destruction, the expectations are
234 * always killed before the conntrack itself */
235 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
237 struct nf_conntrack_expect *new;
239 new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
244 atomic_set(&new->use, 1);
247 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
249 void nf_ct_expect_init(struct nf_conntrack_expect *exp, unsigned int class,
251 const union nf_inet_addr *saddr,
252 const union nf_inet_addr *daddr,
253 u_int8_t proto, const __be16 *src, const __be16 *dst)
257 if (family == AF_INET)
264 exp->expectfn = NULL;
266 exp->tuple.src.l3num = family;
267 exp->tuple.dst.protonum = proto;
270 memcpy(&exp->tuple.src.u3, saddr, len);
271 if (sizeof(exp->tuple.src.u3) > len)
272 /* address needs to be cleared for nf_ct_tuple_equal */
273 memset((void *)&exp->tuple.src.u3 + len, 0x00,
274 sizeof(exp->tuple.src.u3) - len);
275 memset(&exp->mask.src.u3, 0xFF, len);
276 if (sizeof(exp->mask.src.u3) > len)
277 memset((void *)&exp->mask.src.u3 + len, 0x00,
278 sizeof(exp->mask.src.u3) - len);
280 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
281 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
285 exp->tuple.src.u.all = *src;
286 exp->mask.src.u.all = htons(0xFFFF);
288 exp->tuple.src.u.all = 0;
289 exp->mask.src.u.all = 0;
292 memcpy(&exp->tuple.dst.u3, daddr, len);
293 if (sizeof(exp->tuple.dst.u3) > len)
294 /* address needs to be cleared for nf_ct_tuple_equal */
295 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
296 sizeof(exp->tuple.dst.u3) - len);
298 exp->tuple.dst.u.all = *dst;
300 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
302 static void nf_ct_expect_free_rcu(struct rcu_head *head)
304 struct nf_conntrack_expect *exp;
306 exp = container_of(head, struct nf_conntrack_expect, rcu);
307 kmem_cache_free(nf_ct_expect_cachep, exp);
310 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
312 if (atomic_dec_and_test(&exp->use))
313 call_rcu(&exp->rcu, nf_ct_expect_free_rcu);
315 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
317 static void nf_ct_expect_insert(struct nf_conntrack_expect *exp)
319 struct nf_conn_help *master_help = nfct_help(exp->master);
320 struct net *net = nf_ct_exp_net(exp);
321 const struct nf_conntrack_expect_policy *p;
322 unsigned int h = nf_ct_expect_dst_hash(&exp->tuple);
324 /* two references : one for hash insert, one for the timer */
325 atomic_add(2, &exp->use);
328 hlist_add_head(&exp->lnode, &master_help->expectations);
329 master_help->expecting[exp->class]++;
330 } else if (exp->flags & NF_CT_EXPECT_USERSPACE)
331 hlist_add_head(&exp->lnode, &nf_ct_userspace_expect_list);
333 hlist_add_head_rcu(&exp->hnode, &net->ct.expect_hash[h]);
334 net->ct.expect_count++;
336 setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
339 p = &rcu_dereference_protected(
341 lockdep_is_held(&nf_conntrack_lock)
342 )->expect_policy[exp->class];
343 exp->timeout.expires = jiffies + p->timeout * HZ;
345 add_timer(&exp->timeout);
347 NF_CT_STAT_INC(net, expect_create);
350 /* Race with expectations being used means we could have none to find; OK. */
351 static void evict_oldest_expect(struct nf_conn *master,
352 struct nf_conntrack_expect *new)
354 struct nf_conn_help *master_help = nfct_help(master);
355 struct nf_conntrack_expect *exp, *last = NULL;
356 struct hlist_node *n;
358 hlist_for_each_entry(exp, n, &master_help->expectations, lnode) {
359 if (exp->class == new->class)
363 if (last && del_timer(&last->timeout)) {
364 nf_ct_unlink_expect(last);
365 nf_ct_expect_put(last);
369 static inline int __nf_ct_expect_check(struct nf_conntrack_expect *expect)
371 const struct nf_conntrack_expect_policy *p;
372 struct nf_conntrack_expect *i;
373 struct nf_conn *master = expect->master;
374 struct nf_conn_help *master_help = nfct_help(master);
375 struct net *net = nf_ct_exp_net(expect);
376 struct hlist_node *n, *next;
380 /* Don't allow expectations created from kernel-space with no helper */
381 if (!(expect->flags & NF_CT_EXPECT_USERSPACE) &&
382 (!master_help || (master_help && !master_help->helper))) {
386 h = nf_ct_expect_dst_hash(&expect->tuple);
387 hlist_for_each_entry_safe(i, n, next, &net->ct.expect_hash[h], hnode) {
388 if (expect_matches(i, expect)) {
389 if (del_timer(&i->timeout)) {
390 nf_ct_unlink_expect(i);
394 } else if (expect_clash(i, expect)) {
399 /* Will be over limit? */
401 p = &rcu_dereference_protected(
403 lockdep_is_held(&nf_conntrack_lock)
404 )->expect_policy[expect->class];
405 if (p->max_expected &&
406 master_help->expecting[expect->class] >= p->max_expected) {
407 evict_oldest_expect(master, expect);
408 if (master_help->expecting[expect->class]
409 >= p->max_expected) {
416 if (net->ct.expect_count >= nf_ct_expect_max) {
419 "nf_conntrack: expectation table full\n");
426 int nf_ct_expect_related_report(struct nf_conntrack_expect *expect,
431 spin_lock_bh(&nf_conntrack_lock);
432 ret = __nf_ct_expect_check(expect);
437 nf_ct_expect_insert(expect);
438 spin_unlock_bh(&nf_conntrack_lock);
439 nf_ct_expect_event_report(IPEXP_NEW, expect, pid, report);
442 spin_unlock_bh(&nf_conntrack_lock);
445 EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
447 void nf_ct_remove_userspace_expectations(void)
449 struct nf_conntrack_expect *exp;
450 struct hlist_node *n, *next;
452 hlist_for_each_entry_safe(exp, n, next,
453 &nf_ct_userspace_expect_list, lnode) {
454 if (del_timer(&exp->timeout)) {
455 nf_ct_unlink_expect(exp);
456 nf_ct_expect_put(exp);
460 EXPORT_SYMBOL_GPL(nf_ct_remove_userspace_expectations);
462 #ifdef CONFIG_PROC_FS
463 struct ct_expect_iter_state {
464 struct seq_net_private p;
468 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
470 struct net *net = seq_file_net(seq);
471 struct ct_expect_iter_state *st = seq->private;
472 struct hlist_node *n;
474 for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
475 n = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
482 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
483 struct hlist_node *head)
485 struct net *net = seq_file_net(seq);
486 struct ct_expect_iter_state *st = seq->private;
488 head = rcu_dereference(hlist_next_rcu(head));
489 while (head == NULL) {
490 if (++st->bucket >= nf_ct_expect_hsize)
492 head = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
497 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
499 struct hlist_node *head = ct_expect_get_first(seq);
502 while (pos && (head = ct_expect_get_next(seq, head)))
504 return pos ? NULL : head;
507 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
511 return ct_expect_get_idx(seq, *pos);
514 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
517 return ct_expect_get_next(seq, v);
520 static void exp_seq_stop(struct seq_file *seq, void *v)
526 static int exp_seq_show(struct seq_file *s, void *v)
528 struct nf_conntrack_expect *expect;
529 struct nf_conntrack_helper *helper;
530 struct hlist_node *n = v;
533 expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
535 if (expect->timeout.function)
536 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
537 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
540 seq_printf(s, "l3proto = %u proto=%u ",
541 expect->tuple.src.l3num,
542 expect->tuple.dst.protonum);
543 print_tuple(s, &expect->tuple,
544 __nf_ct_l3proto_find(expect->tuple.src.l3num),
545 __nf_ct_l4proto_find(expect->tuple.src.l3num,
546 expect->tuple.dst.protonum));
548 if (expect->flags & NF_CT_EXPECT_PERMANENT) {
549 seq_printf(s, "PERMANENT");
552 if (expect->flags & NF_CT_EXPECT_INACTIVE) {
553 seq_printf(s, "%sINACTIVE", delim);
556 if (expect->flags & NF_CT_EXPECT_USERSPACE)
557 seq_printf(s, "%sUSERSPACE", delim);
559 helper = rcu_dereference(nfct_help(expect->master)->helper);
561 seq_printf(s, "%s%s", expect->flags ? " " : "", helper->name);
562 if (helper->expect_policy[expect->class].name)
564 helper->expect_policy[expect->class].name);
567 return seq_putc(s, '\n');
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_PROC_FS */
592 static int exp_proc_init(struct net *net)
594 #ifdef CONFIG_PROC_FS
595 struct proc_dir_entry *proc;
597 proc = proc_net_fops_create(net, "nf_conntrack_expect", 0440, &exp_file_ops);
600 #endif /* CONFIG_PROC_FS */
604 static void exp_proc_remove(struct net *net)
606 #ifdef CONFIG_PROC_FS
607 proc_net_remove(net, "nf_conntrack_expect");
608 #endif /* CONFIG_PROC_FS */
611 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
613 int nf_conntrack_expect_init(struct net *net)
617 if (net_eq(net, &init_net)) {
618 if (!nf_ct_expect_hsize) {
619 nf_ct_expect_hsize = net->ct.htable_size / 256;
620 if (!nf_ct_expect_hsize)
621 nf_ct_expect_hsize = 1;
623 nf_ct_expect_max = nf_ct_expect_hsize * 4;
626 net->ct.expect_count = 0;
627 net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize, 0);
628 if (net->ct.expect_hash == NULL)
631 if (net_eq(net, &init_net)) {
632 nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
633 sizeof(struct nf_conntrack_expect),
635 if (!nf_ct_expect_cachep)
639 err = exp_proc_init(net);
646 if (net_eq(net, &init_net))
647 kmem_cache_destroy(nf_ct_expect_cachep);
649 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
654 void nf_conntrack_expect_fini(struct net *net)
656 exp_proc_remove(net);
657 if (net_eq(net, &init_net)) {
658 rcu_barrier(); /* Wait for call_rcu() before destroy */
659 kmem_cache_destroy(nf_ct_expect_cachep);
661 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);