1 /* Connection state tracking for netfilter. This is separated from,
2 but required by, the NAT layer; it can also be used by an iptables
5 /* (C) 1999-2001 Paul `Rusty' Russell
6 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
7 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/proc_fs.h>
19 #include <linux/vmalloc.h>
20 #include <linux/stddef.h>
21 #include <linux/slab.h>
22 #include <linux/random.h>
23 #include <linux/jhash.h>
24 #include <linux/err.h>
25 #include <linux/percpu.h>
26 #include <linux/moduleparam.h>
27 #include <linux/notifier.h>
28 #include <linux/kernel.h>
29 #include <linux/netdevice.h>
30 #include <linux/socket.h>
32 #include <linux/rculist_nulls.h>
34 #include <net/netfilter/nf_conntrack.h>
35 #include <net/netfilter/nf_conntrack_l3proto.h>
36 #include <net/netfilter/nf_conntrack_l4proto.h>
37 #include <net/netfilter/nf_conntrack_expect.h>
38 #include <net/netfilter/nf_conntrack_helper.h>
39 #include <net/netfilter/nf_conntrack_core.h>
40 #include <net/netfilter/nf_conntrack_extend.h>
41 #include <net/netfilter/nf_conntrack_acct.h>
42 #include <net/netfilter/nf_conntrack_ecache.h>
43 #include <net/netfilter/nf_nat.h>
44 #include <net/netfilter/nf_nat_core.h>
46 #define NF_CONNTRACK_VERSION "0.5.0"
48 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
49 enum nf_nat_manip_type manip,
50 struct nlattr *attr) __read_mostly;
51 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
53 DEFINE_SPINLOCK(nf_conntrack_lock);
54 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
56 unsigned int nf_conntrack_htable_size __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
59 unsigned int nf_conntrack_max __read_mostly;
60 EXPORT_SYMBOL_GPL(nf_conntrack_max);
62 struct nf_conn nf_conntrack_untracked __read_mostly;
63 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
67 static int nf_conntrack_hash_rnd_initted;
68 static unsigned int nf_conntrack_hash_rnd;
70 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
71 unsigned int size, unsigned int rnd)
76 /* The direction must be ignored, so we hash everything up to the
77 * destination ports (which is a multiple of 4) and treat the last
78 * three bytes manually.
80 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
81 h = jhash2((u32 *)tuple, n,
82 rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
83 tuple->dst.protonum));
85 return ((u64)h * size) >> 32;
88 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
90 return __hash_conntrack(tuple, nf_conntrack_htable_size,
91 nf_conntrack_hash_rnd);
95 nf_ct_get_tuple(const struct sk_buff *skb,
100 struct nf_conntrack_tuple *tuple,
101 const struct nf_conntrack_l3proto *l3proto,
102 const struct nf_conntrack_l4proto *l4proto)
104 memset(tuple, 0, sizeof(*tuple));
106 tuple->src.l3num = l3num;
107 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
110 tuple->dst.protonum = protonum;
111 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
113 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
115 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
117 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
118 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
120 struct nf_conntrack_l3proto *l3proto;
121 struct nf_conntrack_l4proto *l4proto;
122 unsigned int protoff;
128 l3proto = __nf_ct_l3proto_find(l3num);
129 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
130 if (ret != NF_ACCEPT) {
135 l4proto = __nf_ct_l4proto_find(l3num, protonum);
137 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
143 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
146 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
147 const struct nf_conntrack_tuple *orig,
148 const struct nf_conntrack_l3proto *l3proto,
149 const struct nf_conntrack_l4proto *l4proto)
151 memset(inverse, 0, sizeof(*inverse));
153 inverse->src.l3num = orig->src.l3num;
154 if (l3proto->invert_tuple(inverse, orig) == 0)
157 inverse->dst.dir = !orig->dst.dir;
159 inverse->dst.protonum = orig->dst.protonum;
160 return l4proto->invert_tuple(inverse, orig);
162 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
165 clean_from_lists(struct nf_conn *ct)
167 pr_debug("clean_from_lists(%p)\n", ct);
168 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
169 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
171 /* Destroy all pending expectations */
172 nf_ct_remove_expectations(ct);
176 destroy_conntrack(struct nf_conntrack *nfct)
178 struct nf_conn *ct = (struct nf_conn *)nfct;
179 struct net *net = nf_ct_net(ct);
180 struct nf_conntrack_l4proto *l4proto;
182 pr_debug("destroy_conntrack(%p)\n", ct);
183 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
184 NF_CT_ASSERT(!timer_pending(&ct->timeout));
186 if (!test_bit(IPS_DYING_BIT, &ct->status))
187 nf_conntrack_event(IPCT_DESTROY, ct);
188 set_bit(IPS_DYING_BIT, &ct->status);
190 /* To make sure we don't get any weird locking issues here:
191 * destroy_conntrack() MUST NOT be called with a write lock
192 * to nf_conntrack_lock!!! -HW */
194 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
195 if (l4proto && l4proto->destroy)
196 l4proto->destroy(ct);
200 spin_lock_bh(&nf_conntrack_lock);
201 /* Expectations will have been removed in clean_from_lists,
202 * except TFTP can create an expectation on the first packet,
203 * before connection is in the list, so we need to clean here,
205 nf_ct_remove_expectations(ct);
207 /* We overload first tuple to link into unconfirmed list. */
208 if (!nf_ct_is_confirmed(ct)) {
209 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
210 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
213 NF_CT_STAT_INC(net, delete);
214 spin_unlock_bh(&nf_conntrack_lock);
217 nf_ct_put(ct->master);
219 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
220 nf_conntrack_free(ct);
223 static void death_by_timeout(unsigned long ul_conntrack)
225 struct nf_conn *ct = (void *)ul_conntrack;
226 struct net *net = nf_ct_net(ct);
228 nf_ct_helper_destroy(ct);
229 spin_lock_bh(&nf_conntrack_lock);
230 /* Inside lock so preempt is disabled on module removal path.
231 * Otherwise we can get spurious warnings. */
232 NF_CT_STAT_INC(net, delete_list);
233 clean_from_lists(ct);
234 spin_unlock_bh(&nf_conntrack_lock);
240 * - Caller must take a reference on returned object
241 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
243 * - Caller must lock nf_conntrack_lock before calling this function
245 struct nf_conntrack_tuple_hash *
246 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
248 struct nf_conntrack_tuple_hash *h;
249 struct hlist_nulls_node *n;
250 unsigned int hash = hash_conntrack(tuple);
252 /* Disable BHs the entire time since we normally need to disable them
253 * at least once for the stats anyway.
257 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
258 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
259 NF_CT_STAT_INC(net, found);
263 NF_CT_STAT_INC(net, searched);
266 * if the nulls value we got at the end of this lookup is
267 * not the expected one, we must restart lookup.
268 * We probably met an item that was moved to another chain.
270 if (get_nulls_value(n) != hash)
276 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
278 /* Find a connection corresponding to a tuple. */
279 struct nf_conntrack_tuple_hash *
280 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
282 struct nf_conntrack_tuple_hash *h;
287 h = __nf_conntrack_find(net, tuple);
289 ct = nf_ct_tuplehash_to_ctrack(h);
290 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
293 if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
303 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
305 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
307 unsigned int repl_hash)
309 struct net *net = nf_ct_net(ct);
311 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
312 &net->ct.hash[hash]);
313 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
314 &net->ct.hash[repl_hash]);
317 void nf_conntrack_hash_insert(struct nf_conn *ct)
319 unsigned int hash, repl_hash;
321 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
322 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
324 __nf_conntrack_hash_insert(ct, hash, repl_hash);
326 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
328 /* Confirm a connection given skb; places it in hash table */
330 __nf_conntrack_confirm(struct sk_buff *skb)
332 unsigned int hash, repl_hash;
333 struct nf_conntrack_tuple_hash *h;
335 struct nf_conn_help *help;
336 struct hlist_nulls_node *n;
337 enum ip_conntrack_info ctinfo;
340 ct = nf_ct_get(skb, &ctinfo);
343 /* ipt_REJECT uses nf_conntrack_attach to attach related
344 ICMP/TCP RST packets in other direction. Actual packet
345 which created connection will be IP_CT_NEW or for an
346 expected connection, IP_CT_RELATED. */
347 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
350 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
351 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
353 /* We're not in hash table, and we refuse to set up related
354 connections for unconfirmed conns. But packet copies and
355 REJECT will give spurious warnings here. */
356 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
358 /* No external references means noone else could have
360 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
361 pr_debug("Confirming conntrack %p\n", ct);
363 spin_lock_bh(&nf_conntrack_lock);
365 /* See if there's one in the list already, including reverse:
366 NAT could have grabbed it without realizing, since we're
367 not in the hash. If there is, we lost race. */
368 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
369 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
372 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
373 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
377 /* Remove from unconfirmed list */
378 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
380 __nf_conntrack_hash_insert(ct, hash, repl_hash);
381 /* Timer relative to confirmation time, not original
382 setting time, otherwise we'd get timer wrap in
383 weird delay cases. */
384 ct->timeout.expires += jiffies;
385 add_timer(&ct->timeout);
386 atomic_inc(&ct->ct_general.use);
387 set_bit(IPS_CONFIRMED_BIT, &ct->status);
388 NF_CT_STAT_INC(net, insert);
389 spin_unlock_bh(&nf_conntrack_lock);
390 help = nfct_help(ct);
391 if (help && help->helper)
392 nf_conntrack_event_cache(IPCT_HELPER, ct);
394 nf_conntrack_event_cache(master_ct(ct) ?
395 IPCT_RELATED : IPCT_NEW, ct);
399 NF_CT_STAT_INC(net, insert_failed);
400 spin_unlock_bh(&nf_conntrack_lock);
403 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
405 /* Returns true if a connection correspondings to the tuple (required
408 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
409 const struct nf_conn *ignored_conntrack)
411 struct net *net = nf_ct_net(ignored_conntrack);
412 struct nf_conntrack_tuple_hash *h;
413 struct hlist_nulls_node *n;
414 unsigned int hash = hash_conntrack(tuple);
416 /* Disable BHs the entire time since we need to disable them at
417 * least once for the stats anyway.
420 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
421 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
422 nf_ct_tuple_equal(tuple, &h->tuple)) {
423 NF_CT_STAT_INC(net, found);
424 rcu_read_unlock_bh();
427 NF_CT_STAT_INC(net, searched);
429 rcu_read_unlock_bh();
433 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
435 #define NF_CT_EVICTION_RANGE 8
437 /* There's a small race here where we may free a just-assured
438 connection. Too bad: we're in trouble anyway. */
439 static noinline int early_drop(struct net *net, unsigned int hash)
441 /* Use oldest entry, which is roughly LRU */
442 struct nf_conntrack_tuple_hash *h;
443 struct nf_conn *ct = NULL, *tmp;
444 struct hlist_nulls_node *n;
445 unsigned int i, cnt = 0;
449 for (i = 0; i < nf_conntrack_htable_size; i++) {
450 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
452 tmp = nf_ct_tuplehash_to_ctrack(h);
453 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
458 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
460 if (ct || cnt >= NF_CT_EVICTION_RANGE)
462 hash = (hash + 1) % nf_conntrack_htable_size;
469 if (del_timer(&ct->timeout)) {
470 death_by_timeout((unsigned long)ct);
472 NF_CT_STAT_INC_ATOMIC(net, early_drop);
478 struct nf_conn *nf_conntrack_alloc(struct net *net,
479 const struct nf_conntrack_tuple *orig,
480 const struct nf_conntrack_tuple *repl,
485 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
486 get_random_bytes(&nf_conntrack_hash_rnd,
487 sizeof(nf_conntrack_hash_rnd));
488 nf_conntrack_hash_rnd_initted = 1;
491 /* We don't want any race condition at early drop stage */
492 atomic_inc(&net->ct.count);
494 if (nf_conntrack_max &&
495 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
496 unsigned int hash = hash_conntrack(orig);
497 if (!early_drop(net, hash)) {
498 atomic_dec(&net->ct.count);
501 "nf_conntrack: table full, dropping"
503 return ERR_PTR(-ENOMEM);
507 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
509 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
510 atomic_dec(&net->ct.count);
511 return ERR_PTR(-ENOMEM);
514 spin_lock_init(&ct->lock);
515 atomic_set(&ct->ct_general.use, 1);
516 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
517 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
518 /* Don't set timer yet: wait for confirmation */
519 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
526 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
528 void nf_conntrack_free(struct nf_conn *ct)
530 struct net *net = nf_ct_net(ct);
532 nf_ct_ext_destroy(ct);
533 atomic_dec(&net->ct.count);
535 kmem_cache_free(nf_conntrack_cachep, ct);
537 EXPORT_SYMBOL_GPL(nf_conntrack_free);
539 /* Allocate a new conntrack: we return -ENOMEM if classification
540 failed due to stress. Otherwise it really is unclassifiable. */
541 static struct nf_conntrack_tuple_hash *
542 init_conntrack(struct net *net,
543 const struct nf_conntrack_tuple *tuple,
544 struct nf_conntrack_l3proto *l3proto,
545 struct nf_conntrack_l4proto *l4proto,
547 unsigned int dataoff)
550 struct nf_conn_help *help;
551 struct nf_conntrack_tuple repl_tuple;
552 struct nf_conntrack_expect *exp;
554 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
555 pr_debug("Can't invert tuple.\n");
559 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
561 pr_debug("Can't allocate conntrack.\n");
562 return (struct nf_conntrack_tuple_hash *)ct;
565 if (!l4proto->new(ct, skb, dataoff)) {
566 nf_conntrack_free(ct);
567 pr_debug("init conntrack: can't track with proto module\n");
571 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
572 nf_ct_ecache_ext_add(ct, GFP_ATOMIC);
574 spin_lock_bh(&nf_conntrack_lock);
575 exp = nf_ct_find_expectation(net, tuple);
577 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
579 /* Welcome, Mr. Bond. We've been expecting you... */
580 __set_bit(IPS_EXPECTED_BIT, &ct->status);
581 ct->master = exp->master;
583 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
585 rcu_assign_pointer(help->helper, exp->helper);
588 #ifdef CONFIG_NF_CONNTRACK_MARK
589 ct->mark = exp->master->mark;
591 #ifdef CONFIG_NF_CONNTRACK_SECMARK
592 ct->secmark = exp->master->secmark;
594 nf_conntrack_get(&ct->master->ct_general);
595 NF_CT_STAT_INC(net, expect_new);
597 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
598 NF_CT_STAT_INC(net, new);
601 /* Overload tuple linked list to put us in unconfirmed list. */
602 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
603 &net->ct.unconfirmed);
605 spin_unlock_bh(&nf_conntrack_lock);
609 exp->expectfn(ct, exp);
610 nf_ct_expect_put(exp);
613 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
616 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
617 static inline struct nf_conn *
618 resolve_normal_ct(struct net *net,
620 unsigned int dataoff,
623 struct nf_conntrack_l3proto *l3proto,
624 struct nf_conntrack_l4proto *l4proto,
626 enum ip_conntrack_info *ctinfo)
628 struct nf_conntrack_tuple tuple;
629 struct nf_conntrack_tuple_hash *h;
632 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
633 dataoff, l3num, protonum, &tuple, l3proto,
635 pr_debug("resolve_normal_ct: Can't get tuple\n");
639 /* look for tuple match */
640 h = nf_conntrack_find_get(net, &tuple);
642 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
648 ct = nf_ct_tuplehash_to_ctrack(h);
650 /* It exists; we have (non-exclusive) reference. */
651 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
652 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
653 /* Please set reply bit if this packet OK */
656 /* Once we've had two way comms, always ESTABLISHED. */
657 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
658 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
659 *ctinfo = IP_CT_ESTABLISHED;
660 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
661 pr_debug("nf_conntrack_in: related packet for %p\n",
663 *ctinfo = IP_CT_RELATED;
665 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
670 skb->nfct = &ct->ct_general;
671 skb->nfctinfo = *ctinfo;
676 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
680 enum ip_conntrack_info ctinfo;
681 struct nf_conntrack_l3proto *l3proto;
682 struct nf_conntrack_l4proto *l4proto;
683 unsigned int dataoff;
688 /* Previously seen (loopback or untracked)? Ignore. */
690 NF_CT_STAT_INC_ATOMIC(net, ignore);
694 /* rcu_read_lock()ed by nf_hook_slow */
695 l3proto = __nf_ct_l3proto_find(pf);
696 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
697 &dataoff, &protonum);
699 pr_debug("not prepared to track yet or error occured\n");
700 NF_CT_STAT_INC_ATOMIC(net, error);
701 NF_CT_STAT_INC_ATOMIC(net, invalid);
705 l4proto = __nf_ct_l4proto_find(pf, protonum);
707 /* It may be an special packet, error, unclean...
708 * inverse of the return code tells to the netfilter
709 * core what to do with the packet. */
710 if (l4proto->error != NULL) {
711 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
713 NF_CT_STAT_INC_ATOMIC(net, error);
714 NF_CT_STAT_INC_ATOMIC(net, invalid);
719 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
720 l3proto, l4proto, &set_reply, &ctinfo);
722 /* Not valid part of a connection */
723 NF_CT_STAT_INC_ATOMIC(net, invalid);
728 /* Too stressed to deal. */
729 NF_CT_STAT_INC_ATOMIC(net, drop);
733 NF_CT_ASSERT(skb->nfct);
735 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
737 /* Invalid: inverse of the return code tells
738 * the netfilter core what to do */
739 pr_debug("nf_conntrack_in: Can't track with proto module\n");
740 nf_conntrack_put(skb->nfct);
742 NF_CT_STAT_INC_ATOMIC(net, invalid);
744 NF_CT_STAT_INC_ATOMIC(net, drop);
748 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
749 nf_conntrack_event_cache(IPCT_STATUS, ct);
753 EXPORT_SYMBOL_GPL(nf_conntrack_in);
755 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
756 const struct nf_conntrack_tuple *orig)
761 ret = nf_ct_invert_tuple(inverse, orig,
762 __nf_ct_l3proto_find(orig->src.l3num),
763 __nf_ct_l4proto_find(orig->src.l3num,
764 orig->dst.protonum));
768 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
770 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
771 implicitly racy: see __nf_conntrack_confirm */
772 void nf_conntrack_alter_reply(struct nf_conn *ct,
773 const struct nf_conntrack_tuple *newreply)
775 struct nf_conn_help *help = nfct_help(ct);
777 /* Should be unconfirmed, so not in hash table yet */
778 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
780 pr_debug("Altering reply tuple of %p to ", ct);
781 nf_ct_dump_tuple(newreply);
783 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
784 if (ct->master || (help && !hlist_empty(&help->expectations)))
788 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
791 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
793 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
794 void __nf_ct_refresh_acct(struct nf_conn *ct,
795 enum ip_conntrack_info ctinfo,
796 const struct sk_buff *skb,
797 unsigned long extra_jiffies,
800 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
803 /* Only update if this is not a fixed timeout */
804 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
807 /* If not in hash table, timer will not be active yet */
808 if (!nf_ct_is_confirmed(ct)) {
809 ct->timeout.expires = extra_jiffies;
811 unsigned long newtime = jiffies + extra_jiffies;
813 /* Only update the timeout if the new timeout is at least
814 HZ jiffies from the old timeout. Need del_timer for race
815 avoidance (may already be dying). */
816 if (newtime - ct->timeout.expires >= HZ)
817 mod_timer_pending(&ct->timeout, newtime);
822 struct nf_conn_counter *acct;
824 acct = nf_conn_acct_find(ct);
826 spin_lock_bh(&ct->lock);
827 acct[CTINFO2DIR(ctinfo)].packets++;
828 acct[CTINFO2DIR(ctinfo)].bytes +=
829 skb->len - skb_network_offset(skb);
830 spin_unlock_bh(&ct->lock);
834 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
836 bool __nf_ct_kill_acct(struct nf_conn *ct,
837 enum ip_conntrack_info ctinfo,
838 const struct sk_buff *skb,
842 struct nf_conn_counter *acct;
844 acct = nf_conn_acct_find(ct);
846 spin_lock_bh(&ct->lock);
847 acct[CTINFO2DIR(ctinfo)].packets++;
848 acct[CTINFO2DIR(ctinfo)].bytes +=
849 skb->len - skb_network_offset(skb);
850 spin_unlock_bh(&ct->lock);
854 if (del_timer(&ct->timeout)) {
855 ct->timeout.function((unsigned long)ct);
860 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
862 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
864 #include <linux/netfilter/nfnetlink.h>
865 #include <linux/netfilter/nfnetlink_conntrack.h>
866 #include <linux/mutex.h>
868 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
869 * in ip_conntrack_core, since we don't want the protocols to autoload
870 * or depend on ctnetlink */
871 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
872 const struct nf_conntrack_tuple *tuple)
874 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
875 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
881 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
883 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
884 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
885 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
887 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
889 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
890 struct nf_conntrack_tuple *t)
892 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
895 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
896 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
900 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
902 int nf_ct_port_nlattr_tuple_size(void)
904 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
906 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
909 /* Used by ipt_REJECT and ip6t_REJECT. */
910 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
913 enum ip_conntrack_info ctinfo;
915 /* This ICMP is in reverse direction to the packet which caused it */
916 ct = nf_ct_get(skb, &ctinfo);
917 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
918 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
920 ctinfo = IP_CT_RELATED;
922 /* Attach to new skbuff, and increment count */
923 nskb->nfct = &ct->ct_general;
924 nskb->nfctinfo = ctinfo;
925 nf_conntrack_get(nskb->nfct);
928 /* Bring out ya dead! */
929 static struct nf_conn *
930 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
931 void *data, unsigned int *bucket)
933 struct nf_conntrack_tuple_hash *h;
935 struct hlist_nulls_node *n;
937 spin_lock_bh(&nf_conntrack_lock);
938 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
939 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
940 ct = nf_ct_tuplehash_to_ctrack(h);
945 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
946 ct = nf_ct_tuplehash_to_ctrack(h);
948 set_bit(IPS_DYING_BIT, &ct->status);
950 spin_unlock_bh(&nf_conntrack_lock);
953 atomic_inc(&ct->ct_general.use);
954 spin_unlock_bh(&nf_conntrack_lock);
958 void nf_ct_iterate_cleanup(struct net *net,
959 int (*iter)(struct nf_conn *i, void *data),
963 unsigned int bucket = 0;
965 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
966 /* Time to push up daises... */
967 if (del_timer(&ct->timeout))
968 death_by_timeout((unsigned long)ct);
969 /* ... else the timer will get him soon. */
974 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
976 struct __nf_ct_flush_report {
981 static int kill_report(struct nf_conn *i, void *data)
983 struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
985 /* get_next_corpse sets the dying bit for us */
986 nf_conntrack_event_report(IPCT_DESTROY,
993 static int kill_all(struct nf_conn *i, void *data)
998 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1003 free_pages((unsigned long)hash,
1004 get_order(sizeof(struct hlist_head) * size));
1006 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1008 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1010 struct __nf_ct_flush_report fr = {
1014 nf_ct_iterate_cleanup(net, kill_report, &fr);
1016 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1018 static void nf_conntrack_cleanup_init_net(void)
1020 nf_conntrack_helper_fini();
1021 nf_conntrack_proto_fini();
1022 kmem_cache_destroy(nf_conntrack_cachep);
1025 static void nf_conntrack_cleanup_net(struct net *net)
1028 nf_ct_iterate_cleanup(net, kill_all, NULL);
1029 if (atomic_read(&net->ct.count) != 0) {
1031 goto i_see_dead_people;
1033 /* wait until all references to nf_conntrack_untracked are dropped */
1034 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1037 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1038 nf_conntrack_htable_size);
1039 nf_conntrack_ecache_fini(net);
1040 nf_conntrack_acct_fini(net);
1041 nf_conntrack_expect_fini(net);
1042 free_percpu(net->ct.stat);
1045 /* Mishearing the voices in his head, our hero wonders how he's
1046 supposed to kill the mall. */
1047 void nf_conntrack_cleanup(struct net *net)
1049 if (net_eq(net, &init_net))
1050 rcu_assign_pointer(ip_ct_attach, NULL);
1052 /* This makes sure all current packets have passed through
1053 netfilter framework. Roll on, two-stage module
1057 nf_conntrack_cleanup_net(net);
1059 if (net_eq(net, &init_net)) {
1060 rcu_assign_pointer(nf_ct_destroy, NULL);
1061 nf_conntrack_cleanup_init_net();
1065 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1067 struct hlist_nulls_head *hash;
1068 unsigned int nr_slots, i;
1073 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1074 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1075 sz = nr_slots * sizeof(struct hlist_nulls_head);
1076 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1080 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1081 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1085 for (i = 0; i < nr_slots; i++)
1086 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1090 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1092 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1094 int i, bucket, vmalloced, old_vmalloced;
1095 unsigned int hashsize, old_size;
1097 struct hlist_nulls_head *hash, *old_hash;
1098 struct nf_conntrack_tuple_hash *h;
1100 /* On boot, we can set this without any fancy locking. */
1101 if (!nf_conntrack_htable_size)
1102 return param_set_uint(val, kp);
1104 hashsize = simple_strtoul(val, NULL, 0);
1108 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1112 /* We have to rehahs for the new table anyway, so we also can
1113 * use a newrandom seed */
1114 get_random_bytes(&rnd, sizeof(rnd));
1116 /* Lookups in the old hash might happen in parallel, which means we
1117 * might get false negatives during connection lookup. New connections
1118 * created because of a false negative won't make it into the hash
1119 * though since that required taking the lock.
1121 spin_lock_bh(&nf_conntrack_lock);
1122 for (i = 0; i < nf_conntrack_htable_size; i++) {
1123 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1124 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1125 struct nf_conntrack_tuple_hash, hnnode);
1126 hlist_nulls_del_rcu(&h->hnnode);
1127 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1128 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1131 old_size = nf_conntrack_htable_size;
1132 old_vmalloced = init_net.ct.hash_vmalloc;
1133 old_hash = init_net.ct.hash;
1135 nf_conntrack_htable_size = hashsize;
1136 init_net.ct.hash_vmalloc = vmalloced;
1137 init_net.ct.hash = hash;
1138 nf_conntrack_hash_rnd = rnd;
1139 spin_unlock_bh(&nf_conntrack_lock);
1141 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1144 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1146 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1147 &nf_conntrack_htable_size, 0600);
1149 static int nf_conntrack_init_init_net(void)
1154 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1155 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1156 if (!nf_conntrack_htable_size) {
1157 nf_conntrack_htable_size
1158 = (((num_physpages << PAGE_SHIFT) / 16384)
1159 / sizeof(struct hlist_head));
1160 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1161 nf_conntrack_htable_size = 16384;
1162 if (nf_conntrack_htable_size < 32)
1163 nf_conntrack_htable_size = 32;
1165 /* Use a max. factor of four by default to get the same max as
1166 * with the old struct list_heads. When a table size is given
1167 * we use the old value of 8 to avoid reducing the max.
1171 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1173 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1174 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1177 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1178 sizeof(struct nf_conn),
1179 0, SLAB_DESTROY_BY_RCU, NULL);
1180 if (!nf_conntrack_cachep) {
1181 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1186 ret = nf_conntrack_proto_init();
1190 ret = nf_conntrack_helper_init();
1197 nf_conntrack_proto_fini();
1199 kmem_cache_destroy(nf_conntrack_cachep);
1204 static int nf_conntrack_init_net(struct net *net)
1208 atomic_set(&net->ct.count, 0);
1209 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, 0);
1210 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1211 if (!net->ct.stat) {
1215 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1216 &net->ct.hash_vmalloc, 1);
1217 if (!net->ct.hash) {
1219 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1222 ret = nf_conntrack_expect_init(net);
1225 ret = nf_conntrack_acct_init(net);
1228 ret = nf_conntrack_ecache_init(net);
1232 /* Set up fake conntrack:
1233 - to never be deleted, not in any hashes */
1234 #ifdef CONFIG_NET_NS
1235 nf_conntrack_untracked.ct_net = &init_net;
1237 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1238 /* - and look it like as a confirmed connection */
1239 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1244 nf_conntrack_acct_fini(net);
1246 nf_conntrack_expect_fini(net);
1248 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1249 nf_conntrack_htable_size);
1251 free_percpu(net->ct.stat);
1256 int nf_conntrack_init(struct net *net)
1260 if (net_eq(net, &init_net)) {
1261 ret = nf_conntrack_init_init_net();
1265 ret = nf_conntrack_init_net(net);
1269 if (net_eq(net, &init_net)) {
1270 /* For use by REJECT target */
1271 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1272 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1277 if (net_eq(net, &init_net))
1278 nf_conntrack_cleanup_init_net();