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>
33 #include <net/netfilter/nf_conntrack.h>
34 #include <net/netfilter/nf_conntrack_l3proto.h>
35 #include <net/netfilter/nf_conntrack_l4proto.h>
36 #include <net/netfilter/nf_conntrack_expect.h>
37 #include <net/netfilter/nf_conntrack_helper.h>
38 #include <net/netfilter/nf_conntrack_core.h>
39 #include <net/netfilter/nf_conntrack_extend.h>
40 #include <net/netfilter/nf_conntrack_acct.h>
41 #include <net/netfilter/nf_nat.h>
43 #define NF_CONNTRACK_VERSION "0.5.0"
46 (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
47 enum nf_nat_manip_type manip,
48 struct nlattr *attr) __read_mostly;
49 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
51 DEFINE_SPINLOCK(nf_conntrack_lock);
52 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
54 unsigned int nf_conntrack_htable_size __read_mostly;
55 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
57 int nf_conntrack_max __read_mostly;
58 EXPORT_SYMBOL_GPL(nf_conntrack_max);
60 struct nf_conn nf_conntrack_untracked __read_mostly;
61 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
63 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
65 static int nf_conntrack_hash_rnd_initted;
66 static unsigned int nf_conntrack_hash_rnd;
68 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
69 unsigned int size, unsigned int rnd)
74 /* The direction must be ignored, so we hash everything up to the
75 * destination ports (which is a multiple of 4) and treat the last
76 * three bytes manually.
78 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
79 h = jhash2((u32 *)tuple, n,
80 rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
81 tuple->dst.protonum));
83 return ((u64)h * size) >> 32;
86 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
88 return __hash_conntrack(tuple, nf_conntrack_htable_size,
89 nf_conntrack_hash_rnd);
93 nf_ct_get_tuple(const struct sk_buff *skb,
98 struct nf_conntrack_tuple *tuple,
99 const struct nf_conntrack_l3proto *l3proto,
100 const struct nf_conntrack_l4proto *l4proto)
102 memset(tuple, 0, sizeof(*tuple));
104 tuple->src.l3num = l3num;
105 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
108 tuple->dst.protonum = protonum;
109 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
111 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
113 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
115 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
116 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
118 struct nf_conntrack_l3proto *l3proto;
119 struct nf_conntrack_l4proto *l4proto;
120 unsigned int protoff;
126 l3proto = __nf_ct_l3proto_find(l3num);
127 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
128 if (ret != NF_ACCEPT) {
133 l4proto = __nf_ct_l4proto_find(l3num, protonum);
135 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
141 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
144 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
145 const struct nf_conntrack_tuple *orig,
146 const struct nf_conntrack_l3proto *l3proto,
147 const struct nf_conntrack_l4proto *l4proto)
149 memset(inverse, 0, sizeof(*inverse));
151 inverse->src.l3num = orig->src.l3num;
152 if (l3proto->invert_tuple(inverse, orig) == 0)
155 inverse->dst.dir = !orig->dst.dir;
157 inverse->dst.protonum = orig->dst.protonum;
158 return l4proto->invert_tuple(inverse, orig);
160 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
163 clean_from_lists(struct nf_conn *ct)
165 pr_debug("clean_from_lists(%p)\n", ct);
166 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
167 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
169 /* Destroy all pending expectations */
170 nf_ct_remove_expectations(ct);
174 destroy_conntrack(struct nf_conntrack *nfct)
176 struct nf_conn *ct = (struct nf_conn *)nfct;
177 struct net *net = nf_ct_net(ct);
178 struct nf_conntrack_l4proto *l4proto;
180 pr_debug("destroy_conntrack(%p)\n", ct);
181 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
182 NF_CT_ASSERT(!timer_pending(&ct->timeout));
184 nf_conntrack_event(IPCT_DESTROY, ct);
185 set_bit(IPS_DYING_BIT, &ct->status);
187 /* To make sure we don't get any weird locking issues here:
188 * destroy_conntrack() MUST NOT be called with a write lock
189 * to nf_conntrack_lock!!! -HW */
191 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
192 if (l4proto && l4proto->destroy)
193 l4proto->destroy(ct);
197 spin_lock_bh(&nf_conntrack_lock);
198 /* Expectations will have been removed in clean_from_lists,
199 * except TFTP can create an expectation on the first packet,
200 * before connection is in the list, so we need to clean here,
202 nf_ct_remove_expectations(ct);
204 /* We overload first tuple to link into unconfirmed list. */
205 if (!nf_ct_is_confirmed(ct)) {
206 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
207 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
210 NF_CT_STAT_INC(net, delete);
211 spin_unlock_bh(&nf_conntrack_lock);
214 nf_ct_put(ct->master);
216 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
217 nf_conntrack_free(ct);
220 static void death_by_timeout(unsigned long ul_conntrack)
222 struct nf_conn *ct = (void *)ul_conntrack;
223 struct net *net = nf_ct_net(ct);
224 struct nf_conn_help *help = nfct_help(ct);
225 struct nf_conntrack_helper *helper;
229 helper = rcu_dereference(help->helper);
230 if (helper && helper->destroy)
235 spin_lock_bh(&nf_conntrack_lock);
236 /* Inside lock so preempt is disabled on module removal path.
237 * Otherwise we can get spurious warnings. */
238 NF_CT_STAT_INC(net, delete_list);
239 clean_from_lists(ct);
240 spin_unlock_bh(&nf_conntrack_lock);
244 struct nf_conntrack_tuple_hash *
245 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
247 struct nf_conntrack_tuple_hash *h;
248 struct hlist_node *n;
249 unsigned int hash = hash_conntrack(tuple);
251 /* Disable BHs the entire time since we normally need to disable them
252 * at least once for the stats anyway.
255 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
256 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
257 NF_CT_STAT_INC(net, found);
261 NF_CT_STAT_INC(net, searched);
267 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
269 /* Find a connection corresponding to a tuple. */
270 struct nf_conntrack_tuple_hash *
271 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
273 struct nf_conntrack_tuple_hash *h;
277 h = __nf_conntrack_find(net, tuple);
279 ct = nf_ct_tuplehash_to_ctrack(h);
280 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
287 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
289 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
291 unsigned int repl_hash)
293 struct net *net = nf_ct_net(ct);
295 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
296 &net->ct.hash[hash]);
297 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
298 &net->ct.hash[repl_hash]);
301 void nf_conntrack_hash_insert(struct nf_conn *ct)
303 unsigned int hash, repl_hash;
305 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
306 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
308 __nf_conntrack_hash_insert(ct, hash, repl_hash);
310 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
312 /* Confirm a connection given skb; places it in hash table */
314 __nf_conntrack_confirm(struct sk_buff *skb)
316 unsigned int hash, repl_hash;
317 struct nf_conntrack_tuple_hash *h;
319 struct nf_conn_help *help;
320 struct hlist_node *n;
321 enum ip_conntrack_info ctinfo;
324 ct = nf_ct_get(skb, &ctinfo);
327 /* ipt_REJECT uses nf_conntrack_attach to attach related
328 ICMP/TCP RST packets in other direction. Actual packet
329 which created connection will be IP_CT_NEW or for an
330 expected connection, IP_CT_RELATED. */
331 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
334 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
335 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
337 /* We're not in hash table, and we refuse to set up related
338 connections for unconfirmed conns. But packet copies and
339 REJECT will give spurious warnings here. */
340 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
342 /* No external references means noone else could have
344 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
345 pr_debug("Confirming conntrack %p\n", ct);
347 spin_lock_bh(&nf_conntrack_lock);
349 /* See if there's one in the list already, including reverse:
350 NAT could have grabbed it without realizing, since we're
351 not in the hash. If there is, we lost race. */
352 hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode)
353 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
356 hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode)
357 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
361 /* Remove from unconfirmed list */
362 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
364 __nf_conntrack_hash_insert(ct, hash, repl_hash);
365 /* Timer relative to confirmation time, not original
366 setting time, otherwise we'd get timer wrap in
367 weird delay cases. */
368 ct->timeout.expires += jiffies;
369 add_timer(&ct->timeout);
370 atomic_inc(&ct->ct_general.use);
371 set_bit(IPS_CONFIRMED_BIT, &ct->status);
372 NF_CT_STAT_INC(net, insert);
373 spin_unlock_bh(&nf_conntrack_lock);
374 help = nfct_help(ct);
375 if (help && help->helper)
376 nf_conntrack_event_cache(IPCT_HELPER, ct);
377 #ifdef CONFIG_NF_NAT_NEEDED
378 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
379 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
380 nf_conntrack_event_cache(IPCT_NATINFO, ct);
382 nf_conntrack_event_cache(master_ct(ct) ?
383 IPCT_RELATED : IPCT_NEW, ct);
387 NF_CT_STAT_INC(net, insert_failed);
388 spin_unlock_bh(&nf_conntrack_lock);
391 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
393 /* Returns true if a connection correspondings to the tuple (required
396 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
397 const struct nf_conn *ignored_conntrack)
399 struct net *net = nf_ct_net(ignored_conntrack);
400 struct nf_conntrack_tuple_hash *h;
401 struct hlist_node *n;
402 unsigned int hash = hash_conntrack(tuple);
404 /* Disable BHs the entire time since we need to disable them at
405 * least once for the stats anyway.
408 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
409 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
410 nf_ct_tuple_equal(tuple, &h->tuple)) {
411 NF_CT_STAT_INC(net, found);
412 rcu_read_unlock_bh();
415 NF_CT_STAT_INC(net, searched);
417 rcu_read_unlock_bh();
421 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
423 #define NF_CT_EVICTION_RANGE 8
425 /* There's a small race here where we may free a just-assured
426 connection. Too bad: we're in trouble anyway. */
427 static noinline int early_drop(struct net *net, unsigned int hash)
429 /* Use oldest entry, which is roughly LRU */
430 struct nf_conntrack_tuple_hash *h;
431 struct nf_conn *ct = NULL, *tmp;
432 struct hlist_node *n;
433 unsigned int i, cnt = 0;
437 for (i = 0; i < nf_conntrack_htable_size; i++) {
438 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash],
440 tmp = nf_ct_tuplehash_to_ctrack(h);
441 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
446 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
448 if (ct || cnt >= NF_CT_EVICTION_RANGE)
450 hash = (hash + 1) % nf_conntrack_htable_size;
457 if (del_timer(&ct->timeout)) {
458 death_by_timeout((unsigned long)ct);
460 NF_CT_STAT_INC_ATOMIC(net, early_drop);
466 struct nf_conn *nf_conntrack_alloc(struct net *net,
467 const struct nf_conntrack_tuple *orig,
468 const struct nf_conntrack_tuple *repl,
471 struct nf_conn *ct = NULL;
473 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
474 get_random_bytes(&nf_conntrack_hash_rnd, 4);
475 nf_conntrack_hash_rnd_initted = 1;
478 /* We don't want any race condition at early drop stage */
479 atomic_inc(&net->ct.count);
481 if (nf_conntrack_max &&
482 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
483 unsigned int hash = hash_conntrack(orig);
484 if (!early_drop(net, hash)) {
485 atomic_dec(&net->ct.count);
488 "nf_conntrack: table full, dropping"
490 return ERR_PTR(-ENOMEM);
494 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
496 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
497 atomic_dec(&net->ct.count);
498 return ERR_PTR(-ENOMEM);
501 atomic_set(&ct->ct_general.use, 1);
502 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
503 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
504 /* Don't set timer yet: wait for confirmation */
505 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
509 INIT_RCU_HEAD(&ct->rcu);
513 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
515 static void nf_conntrack_free_rcu(struct rcu_head *head)
517 struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
518 struct net *net = nf_ct_net(ct);
521 kmem_cache_free(nf_conntrack_cachep, ct);
522 atomic_dec(&net->ct.count);
525 void nf_conntrack_free(struct nf_conn *ct)
527 nf_ct_ext_destroy(ct);
528 call_rcu(&ct->rcu, nf_conntrack_free_rcu);
530 EXPORT_SYMBOL_GPL(nf_conntrack_free);
532 /* Allocate a new conntrack: we return -ENOMEM if classification
533 failed due to stress. Otherwise it really is unclassifiable. */
534 static struct nf_conntrack_tuple_hash *
535 init_conntrack(struct net *net,
536 const struct nf_conntrack_tuple *tuple,
537 struct nf_conntrack_l3proto *l3proto,
538 struct nf_conntrack_l4proto *l4proto,
540 unsigned int dataoff)
543 struct nf_conn_help *help;
544 struct nf_conntrack_tuple repl_tuple;
545 struct nf_conntrack_expect *exp;
547 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
548 pr_debug("Can't invert tuple.\n");
552 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
553 if (ct == NULL || IS_ERR(ct)) {
554 pr_debug("Can't allocate conntrack.\n");
555 return (struct nf_conntrack_tuple_hash *)ct;
558 if (!l4proto->new(ct, skb, dataoff)) {
559 nf_conntrack_free(ct);
560 pr_debug("init conntrack: can't track with proto module\n");
564 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
566 spin_lock_bh(&nf_conntrack_lock);
567 exp = nf_ct_find_expectation(net, tuple);
569 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
571 /* Welcome, Mr. Bond. We've been expecting you... */
572 __set_bit(IPS_EXPECTED_BIT, &ct->status);
573 ct->master = exp->master;
575 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
577 rcu_assign_pointer(help->helper, exp->helper);
580 #ifdef CONFIG_NF_CONNTRACK_MARK
581 ct->mark = exp->master->mark;
583 #ifdef CONFIG_NF_CONNTRACK_SECMARK
584 ct->secmark = exp->master->secmark;
586 nf_conntrack_get(&ct->master->ct_general);
587 NF_CT_STAT_INC(net, expect_new);
589 struct nf_conntrack_helper *helper;
591 helper = __nf_ct_helper_find(&repl_tuple);
593 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
595 rcu_assign_pointer(help->helper, helper);
597 NF_CT_STAT_INC(net, new);
600 /* Overload tuple linked list to put us in unconfirmed list. */
601 hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
602 &net->ct.unconfirmed);
604 spin_unlock_bh(&nf_conntrack_lock);
608 exp->expectfn(ct, exp);
609 nf_ct_expect_put(exp);
612 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
615 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
616 static inline struct nf_conn *
617 resolve_normal_ct(struct net *net,
619 unsigned int dataoff,
622 struct nf_conntrack_l3proto *l3proto,
623 struct nf_conntrack_l4proto *l4proto,
625 enum ip_conntrack_info *ctinfo)
627 struct nf_conntrack_tuple tuple;
628 struct nf_conntrack_tuple_hash *h;
631 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
632 dataoff, l3num, protonum, &tuple, l3proto,
634 pr_debug("resolve_normal_ct: Can't get tuple\n");
638 /* look for tuple match */
639 h = nf_conntrack_find_get(net, &tuple);
641 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
647 ct = nf_ct_tuplehash_to_ctrack(h);
649 /* It exists; we have (non-exclusive) reference. */
650 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
651 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
652 /* Please set reply bit if this packet OK */
655 /* Once we've had two way comms, always ESTABLISHED. */
656 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
657 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
658 *ctinfo = IP_CT_ESTABLISHED;
659 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
660 pr_debug("nf_conntrack_in: related packet for %p\n",
662 *ctinfo = IP_CT_RELATED;
664 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
669 skb->nfct = &ct->ct_general;
670 skb->nfctinfo = *ctinfo;
675 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
679 enum ip_conntrack_info ctinfo;
680 struct nf_conntrack_l3proto *l3proto;
681 struct nf_conntrack_l4proto *l4proto;
682 unsigned int dataoff;
687 /* Previously seen (loopback or untracked)? Ignore. */
689 NF_CT_STAT_INC_ATOMIC(net, ignore);
693 /* rcu_read_lock()ed by nf_hook_slow */
694 l3proto = __nf_ct_l3proto_find(pf);
695 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
696 &dataoff, &protonum);
698 pr_debug("not prepared to track yet or error occured\n");
699 NF_CT_STAT_INC_ATOMIC(net, error);
700 NF_CT_STAT_INC_ATOMIC(net, invalid);
704 l4proto = __nf_ct_l4proto_find(pf, protonum);
706 /* It may be an special packet, error, unclean...
707 * inverse of the return code tells to the netfilter
708 * core what to do with the packet. */
709 if (l4proto->error != NULL) {
710 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
712 NF_CT_STAT_INC_ATOMIC(net, error);
713 NF_CT_STAT_INC_ATOMIC(net, invalid);
718 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
719 l3proto, l4proto, &set_reply, &ctinfo);
721 /* Not valid part of a connection */
722 NF_CT_STAT_INC_ATOMIC(net, invalid);
727 /* Too stressed to deal. */
728 NF_CT_STAT_INC_ATOMIC(net, drop);
732 NF_CT_ASSERT(skb->nfct);
734 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
736 /* Invalid: inverse of the return code tells
737 * the netfilter core what to do */
738 pr_debug("nf_conntrack_in: Can't track with proto module\n");
739 nf_conntrack_put(skb->nfct);
741 NF_CT_STAT_INC_ATOMIC(net, invalid);
745 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
746 nf_conntrack_event_cache(IPCT_STATUS, ct);
750 EXPORT_SYMBOL_GPL(nf_conntrack_in);
752 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
753 const struct nf_conntrack_tuple *orig)
758 ret = nf_ct_invert_tuple(inverse, orig,
759 __nf_ct_l3proto_find(orig->src.l3num),
760 __nf_ct_l4proto_find(orig->src.l3num,
761 orig->dst.protonum));
765 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
767 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
768 implicitly racy: see __nf_conntrack_confirm */
769 void nf_conntrack_alter_reply(struct nf_conn *ct,
770 const struct nf_conntrack_tuple *newreply)
772 struct nf_conn_help *help = nfct_help(ct);
773 struct nf_conntrack_helper *helper;
775 /* Should be unconfirmed, so not in hash table yet */
776 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
778 pr_debug("Altering reply tuple of %p to ", ct);
779 nf_ct_dump_tuple(newreply);
781 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
782 if (ct->master || (help && !hlist_empty(&help->expectations)))
786 helper = __nf_ct_helper_find(newreply);
787 if (helper == NULL) {
789 rcu_assign_pointer(help->helper, NULL);
794 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
798 memset(&help->help, 0, sizeof(help->help));
801 rcu_assign_pointer(help->helper, helper);
805 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
807 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
808 void __nf_ct_refresh_acct(struct nf_conn *ct,
809 enum ip_conntrack_info ctinfo,
810 const struct sk_buff *skb,
811 unsigned long extra_jiffies,
816 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
819 spin_lock_bh(&nf_conntrack_lock);
821 /* Only update if this is not a fixed timeout */
822 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
825 /* If not in hash table, timer will not be active yet */
826 if (!nf_ct_is_confirmed(ct)) {
827 ct->timeout.expires = extra_jiffies;
828 event = IPCT_REFRESH;
830 unsigned long newtime = jiffies + extra_jiffies;
832 /* Only update the timeout if the new timeout is at least
833 HZ jiffies from the old timeout. Need del_timer for race
834 avoidance (may already be dying). */
835 if (newtime - ct->timeout.expires >= HZ
836 && del_timer(&ct->timeout)) {
837 ct->timeout.expires = newtime;
838 add_timer(&ct->timeout);
839 event = IPCT_REFRESH;
845 struct nf_conn_counter *acct;
847 acct = nf_conn_acct_find(ct);
849 acct[CTINFO2DIR(ctinfo)].packets++;
850 acct[CTINFO2DIR(ctinfo)].bytes +=
851 skb->len - skb_network_offset(skb);
855 spin_unlock_bh(&nf_conntrack_lock);
857 /* must be unlocked when calling event cache */
859 nf_conntrack_event_cache(event, ct);
861 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
863 bool __nf_ct_kill_acct(struct nf_conn *ct,
864 enum ip_conntrack_info ctinfo,
865 const struct sk_buff *skb,
869 struct nf_conn_counter *acct;
871 spin_lock_bh(&nf_conntrack_lock);
872 acct = nf_conn_acct_find(ct);
874 acct[CTINFO2DIR(ctinfo)].packets++;
875 acct[CTINFO2DIR(ctinfo)].bytes +=
876 skb->len - skb_network_offset(skb);
878 spin_unlock_bh(&nf_conntrack_lock);
881 if (del_timer(&ct->timeout)) {
882 ct->timeout.function((unsigned long)ct);
887 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
889 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
891 #include <linux/netfilter/nfnetlink.h>
892 #include <linux/netfilter/nfnetlink_conntrack.h>
893 #include <linux/mutex.h>
895 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
896 * in ip_conntrack_core, since we don't want the protocols to autoload
897 * or depend on ctnetlink */
898 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
899 const struct nf_conntrack_tuple *tuple)
901 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
902 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
908 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
910 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
911 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
912 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
914 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
916 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
917 struct nf_conntrack_tuple *t)
919 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
922 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
923 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
927 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
930 /* Used by ipt_REJECT and ip6t_REJECT. */
931 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
934 enum ip_conntrack_info ctinfo;
936 /* This ICMP is in reverse direction to the packet which caused it */
937 ct = nf_ct_get(skb, &ctinfo);
938 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
939 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
941 ctinfo = IP_CT_RELATED;
943 /* Attach to new skbuff, and increment count */
944 nskb->nfct = &ct->ct_general;
945 nskb->nfctinfo = ctinfo;
946 nf_conntrack_get(nskb->nfct);
949 /* Bring out ya dead! */
950 static struct nf_conn *
951 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
952 void *data, unsigned int *bucket)
954 struct nf_conntrack_tuple_hash *h;
956 struct hlist_node *n;
958 spin_lock_bh(&nf_conntrack_lock);
959 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
960 hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) {
961 ct = nf_ct_tuplehash_to_ctrack(h);
966 hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) {
967 ct = nf_ct_tuplehash_to_ctrack(h);
969 set_bit(IPS_DYING_BIT, &ct->status);
971 spin_unlock_bh(&nf_conntrack_lock);
974 atomic_inc(&ct->ct_general.use);
975 spin_unlock_bh(&nf_conntrack_lock);
979 void nf_ct_iterate_cleanup(struct net *net,
980 int (*iter)(struct nf_conn *i, void *data),
984 unsigned int bucket = 0;
986 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
987 /* Time to push up daises... */
988 if (del_timer(&ct->timeout))
989 death_by_timeout((unsigned long)ct);
990 /* ... else the timer will get him soon. */
995 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
997 static int kill_all(struct nf_conn *i, void *data)
1002 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
1007 free_pages((unsigned long)hash,
1008 get_order(sizeof(struct hlist_head) * size));
1010 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1012 void nf_conntrack_flush(struct net *net)
1014 nf_ct_iterate_cleanup(net, kill_all, NULL);
1016 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
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)
1027 nf_ct_event_cache_flush(net);
1028 nf_conntrack_ecache_fini(net);
1030 nf_conntrack_flush(net);
1031 if (atomic_read(&net->ct.count) != 0) {
1033 goto i_see_dead_people;
1035 /* wait until all references to nf_conntrack_untracked are dropped */
1036 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1039 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1040 nf_conntrack_htable_size);
1041 nf_conntrack_acct_fini(net);
1042 nf_conntrack_expect_fini(net);
1043 free_percpu(net->ct.stat);
1046 /* Mishearing the voices in his head, our hero wonders how he's
1047 supposed to kill the mall. */
1048 void nf_conntrack_cleanup(struct net *net)
1050 if (net_eq(net, &init_net))
1051 rcu_assign_pointer(ip_ct_attach, NULL);
1053 /* This makes sure all current packets have passed through
1054 netfilter framework. Roll on, two-stage module
1058 nf_conntrack_cleanup_net(net);
1060 if (net_eq(net, &init_net)) {
1061 rcu_assign_pointer(nf_ct_destroy, NULL);
1062 nf_conntrack_cleanup_init_net();
1066 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1068 struct hlist_head *hash;
1069 unsigned int size, i;
1073 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1074 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1075 get_order(sizeof(struct hlist_head)
1079 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1080 hash = vmalloc(sizeof(struct hlist_head) * size);
1084 for (i = 0; i < size; i++)
1085 INIT_HLIST_HEAD(&hash[i]);
1089 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1091 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1093 int i, bucket, vmalloced, old_vmalloced;
1094 unsigned int hashsize, old_size;
1096 struct hlist_head *hash, *old_hash;
1097 struct nf_conntrack_tuple_hash *h;
1099 /* On boot, we can set this without any fancy locking. */
1100 if (!nf_conntrack_htable_size)
1101 return param_set_uint(val, kp);
1103 hashsize = simple_strtoul(val, NULL, 0);
1107 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1111 /* We have to rehahs for the new table anyway, so we also can
1112 * use a newrandom seed */
1113 get_random_bytes(&rnd, 4);
1115 /* Lookups in the old hash might happen in parallel, which means we
1116 * might get false negatives during connection lookup. New connections
1117 * created because of a false negative won't make it into the hash
1118 * though since that required taking the lock.
1120 spin_lock_bh(&nf_conntrack_lock);
1121 for (i = 0; i < nf_conntrack_htable_size; i++) {
1122 while (!hlist_empty(&init_net.ct.hash[i])) {
1123 h = hlist_entry(init_net.ct.hash[i].first,
1124 struct nf_conntrack_tuple_hash, hnode);
1125 hlist_del_rcu(&h->hnode);
1126 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1127 hlist_add_head(&h->hnode, &hash[bucket]);
1130 old_size = nf_conntrack_htable_size;
1131 old_vmalloced = init_net.ct.hash_vmalloc;
1132 old_hash = init_net.ct.hash;
1134 nf_conntrack_htable_size = hashsize;
1135 init_net.ct.hash_vmalloc = vmalloced;
1136 init_net.ct.hash = hash;
1137 nf_conntrack_hash_rnd = rnd;
1138 spin_unlock_bh(&nf_conntrack_lock);
1140 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1143 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1145 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1146 &nf_conntrack_htable_size, 0600);
1148 static int nf_conntrack_init_init_net(void)
1153 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1154 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1155 if (!nf_conntrack_htable_size) {
1156 nf_conntrack_htable_size
1157 = (((num_physpages << PAGE_SHIFT) / 16384)
1158 / sizeof(struct hlist_head));
1159 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1160 nf_conntrack_htable_size = 16384;
1161 if (nf_conntrack_htable_size < 32)
1162 nf_conntrack_htable_size = 32;
1164 /* Use a max. factor of four by default to get the same max as
1165 * with the old struct list_heads. When a table size is given
1166 * we use the old value of 8 to avoid reducing the max.
1170 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1172 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1173 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1176 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1177 sizeof(struct nf_conn),
1179 if (!nf_conntrack_cachep) {
1180 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1185 ret = nf_conntrack_proto_init();
1189 ret = nf_conntrack_helper_init();
1196 nf_conntrack_proto_fini();
1198 kmem_cache_destroy(nf_conntrack_cachep);
1203 static int nf_conntrack_init_net(struct net *net)
1207 atomic_set(&net->ct.count, 0);
1208 INIT_HLIST_HEAD(&net->ct.unconfirmed);
1209 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1210 if (!net->ct.stat) {
1214 ret = nf_conntrack_ecache_init(net);
1217 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1218 &net->ct.hash_vmalloc);
1219 if (!net->ct.hash) {
1221 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1224 ret = nf_conntrack_expect_init(net);
1227 ret = nf_conntrack_acct_init(net);
1231 /* Set up fake conntrack:
1232 - to never be deleted, not in any hashes */
1233 #ifdef CONFIG_NET_NS
1234 nf_conntrack_untracked.ct_net = &init_net;
1236 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1237 /* - and look it like as a confirmed connection */
1238 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1243 nf_conntrack_expect_fini(net);
1245 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1246 nf_conntrack_htable_size);
1248 nf_conntrack_ecache_fini(net);
1250 free_percpu(net->ct.stat);
1255 int nf_conntrack_init(struct net *net)
1259 if (net_eq(net, &init_net)) {
1260 ret = nf_conntrack_init_init_net();
1264 ret = nf_conntrack_init_net(net);
1268 if (net_eq(net, &init_net)) {
1269 /* For use by REJECT target */
1270 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1271 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1276 if (net_eq(net, &init_net))
1277 nf_conntrack_cleanup_init_net();