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>
42 #include <net/netfilter/nf_nat_core.h>
44 #define NF_CONNTRACK_VERSION "0.5.0"
46 int (*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 if (!test_bit(IPS_DYING_BIT, &ct->status))
185 nf_conntrack_event(IPCT_DESTROY, ct);
186 set_bit(IPS_DYING_BIT, &ct->status);
188 /* To make sure we don't get any weird locking issues here:
189 * destroy_conntrack() MUST NOT be called with a write lock
190 * to nf_conntrack_lock!!! -HW */
192 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
193 if (l4proto && l4proto->destroy)
194 l4proto->destroy(ct);
198 spin_lock_bh(&nf_conntrack_lock);
199 /* Expectations will have been removed in clean_from_lists,
200 * except TFTP can create an expectation on the first packet,
201 * before connection is in the list, so we need to clean here,
203 nf_ct_remove_expectations(ct);
205 /* We overload first tuple to link into unconfirmed list. */
206 if (!nf_ct_is_confirmed(ct)) {
207 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
208 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
211 NF_CT_STAT_INC(net, delete);
212 spin_unlock_bh(&nf_conntrack_lock);
215 nf_ct_put(ct->master);
217 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
218 nf_conntrack_free(ct);
221 static void death_by_timeout(unsigned long ul_conntrack)
223 struct nf_conn *ct = (void *)ul_conntrack;
224 struct net *net = nf_ct_net(ct);
225 struct nf_conn_help *help = nfct_help(ct);
226 struct nf_conntrack_helper *helper;
230 helper = rcu_dereference(help->helper);
231 if (helper && helper->destroy)
236 spin_lock_bh(&nf_conntrack_lock);
237 /* Inside lock so preempt is disabled on module removal path.
238 * Otherwise we can get spurious warnings. */
239 NF_CT_STAT_INC(net, delete_list);
240 clean_from_lists(ct);
241 spin_unlock_bh(&nf_conntrack_lock);
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_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.
256 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
257 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
258 NF_CT_STAT_INC(net, found);
262 NF_CT_STAT_INC(net, searched);
268 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
270 /* Find a connection corresponding to a tuple. */
271 struct nf_conntrack_tuple_hash *
272 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
274 struct nf_conntrack_tuple_hash *h;
278 h = __nf_conntrack_find(net, tuple);
280 ct = nf_ct_tuplehash_to_ctrack(h);
281 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
288 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
290 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
292 unsigned int repl_hash)
294 struct net *net = nf_ct_net(ct);
296 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
297 &net->ct.hash[hash]);
298 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
299 &net->ct.hash[repl_hash]);
302 void nf_conntrack_hash_insert(struct nf_conn *ct)
304 unsigned int hash, repl_hash;
306 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
307 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
309 spin_lock_bh(&nf_conntrack_lock);
310 __nf_conntrack_hash_insert(ct, hash, repl_hash);
311 spin_unlock_bh(&nf_conntrack_lock);
313 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
315 /* Confirm a connection given skb; places it in hash table */
317 __nf_conntrack_confirm(struct sk_buff *skb)
319 unsigned int hash, repl_hash;
320 struct nf_conntrack_tuple_hash *h;
322 struct nf_conn_help *help;
323 struct hlist_node *n;
324 enum ip_conntrack_info ctinfo;
327 ct = nf_ct_get(skb, &ctinfo);
330 /* ipt_REJECT uses nf_conntrack_attach to attach related
331 ICMP/TCP RST packets in other direction. Actual packet
332 which created connection will be IP_CT_NEW or for an
333 expected connection, IP_CT_RELATED. */
334 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
337 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
338 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
340 /* We're not in hash table, and we refuse to set up related
341 connections for unconfirmed conns. But packet copies and
342 REJECT will give spurious warnings here. */
343 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
345 /* No external references means noone else could have
347 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
348 pr_debug("Confirming conntrack %p\n", ct);
350 spin_lock_bh(&nf_conntrack_lock);
352 /* See if there's one in the list already, including reverse:
353 NAT could have grabbed it without realizing, since we're
354 not in the hash. If there is, we lost race. */
355 hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode)
356 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
359 hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode)
360 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
364 /* Remove from unconfirmed list */
365 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
367 __nf_conntrack_hash_insert(ct, hash, repl_hash);
368 /* Timer relative to confirmation time, not original
369 setting time, otherwise we'd get timer wrap in
370 weird delay cases. */
371 ct->timeout.expires += jiffies;
372 add_timer(&ct->timeout);
373 atomic_inc(&ct->ct_general.use);
374 set_bit(IPS_CONFIRMED_BIT, &ct->status);
375 NF_CT_STAT_INC(net, insert);
376 spin_unlock_bh(&nf_conntrack_lock);
377 help = nfct_help(ct);
378 if (help && help->helper)
379 nf_conntrack_event_cache(IPCT_HELPER, ct);
380 #ifdef CONFIG_NF_NAT_NEEDED
381 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
382 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
383 nf_conntrack_event_cache(IPCT_NATINFO, ct);
385 nf_conntrack_event_cache(master_ct(ct) ?
386 IPCT_RELATED : IPCT_NEW, ct);
390 NF_CT_STAT_INC(net, insert_failed);
391 spin_unlock_bh(&nf_conntrack_lock);
394 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
396 /* Returns true if a connection correspondings to the tuple (required
399 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
400 const struct nf_conn *ignored_conntrack)
402 struct net *net = nf_ct_net(ignored_conntrack);
403 struct nf_conntrack_tuple_hash *h;
404 struct hlist_node *n;
405 unsigned int hash = hash_conntrack(tuple);
407 /* Disable BHs the entire time since we need to disable them at
408 * least once for the stats anyway.
411 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
412 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
413 nf_ct_tuple_equal(tuple, &h->tuple)) {
414 NF_CT_STAT_INC(net, found);
415 rcu_read_unlock_bh();
418 NF_CT_STAT_INC(net, searched);
420 rcu_read_unlock_bh();
424 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
426 #define NF_CT_EVICTION_RANGE 8
428 /* There's a small race here where we may free a just-assured
429 connection. Too bad: we're in trouble anyway. */
430 static noinline int early_drop(struct net *net, unsigned int hash)
432 /* Use oldest entry, which is roughly LRU */
433 struct nf_conntrack_tuple_hash *h;
434 struct nf_conn *ct = NULL, *tmp;
435 struct hlist_node *n;
436 unsigned int i, cnt = 0;
440 for (i = 0; i < nf_conntrack_htable_size; i++) {
441 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash],
443 tmp = nf_ct_tuplehash_to_ctrack(h);
444 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
449 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
451 if (ct || cnt >= NF_CT_EVICTION_RANGE)
453 hash = (hash + 1) % nf_conntrack_htable_size;
460 if (del_timer(&ct->timeout)) {
461 death_by_timeout((unsigned long)ct);
463 NF_CT_STAT_INC_ATOMIC(net, early_drop);
469 struct nf_conn *nf_conntrack_alloc(struct net *net,
470 const struct nf_conntrack_tuple *orig,
471 const struct nf_conntrack_tuple *repl,
474 struct nf_conn *ct = NULL;
476 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
477 get_random_bytes(&nf_conntrack_hash_rnd, 4);
478 nf_conntrack_hash_rnd_initted = 1;
481 /* We don't want any race condition at early drop stage */
482 atomic_inc(&net->ct.count);
484 if (nf_conntrack_max &&
485 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
486 unsigned int hash = hash_conntrack(orig);
487 if (!early_drop(net, hash)) {
488 atomic_dec(&net->ct.count);
491 "nf_conntrack: table full, dropping"
493 return ERR_PTR(-ENOMEM);
497 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
499 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
500 atomic_dec(&net->ct.count);
501 return ERR_PTR(-ENOMEM);
504 atomic_set(&ct->ct_general.use, 1);
505 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
506 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
507 /* Don't set timer yet: wait for confirmation */
508 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
512 INIT_RCU_HEAD(&ct->rcu);
516 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
518 static void nf_conntrack_free_rcu(struct rcu_head *head)
520 struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
521 struct net *net = nf_ct_net(ct);
524 kmem_cache_free(nf_conntrack_cachep, ct);
525 atomic_dec(&net->ct.count);
528 void nf_conntrack_free(struct nf_conn *ct)
530 nf_ct_ext_destroy(ct);
531 call_rcu(&ct->rcu, nf_conntrack_free_rcu);
533 EXPORT_SYMBOL_GPL(nf_conntrack_free);
535 /* Allocate a new conntrack: we return -ENOMEM if classification
536 failed due to stress. Otherwise it really is unclassifiable. */
537 static struct nf_conntrack_tuple_hash *
538 init_conntrack(struct net *net,
539 const struct nf_conntrack_tuple *tuple,
540 struct nf_conntrack_l3proto *l3proto,
541 struct nf_conntrack_l4proto *l4proto,
543 unsigned int dataoff)
546 struct nf_conn_help *help;
547 struct nf_conntrack_tuple repl_tuple;
548 struct nf_conntrack_expect *exp;
550 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
551 pr_debug("Can't invert tuple.\n");
555 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
556 if (ct == NULL || IS_ERR(ct)) {
557 pr_debug("Can't allocate conntrack.\n");
558 return (struct nf_conntrack_tuple_hash *)ct;
561 if (!l4proto->new(ct, skb, dataoff)) {
562 nf_conntrack_free(ct);
563 pr_debug("init conntrack: can't track with proto module\n");
567 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
569 spin_lock_bh(&nf_conntrack_lock);
570 exp = nf_ct_find_expectation(net, tuple);
572 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
574 /* Welcome, Mr. Bond. We've been expecting you... */
575 __set_bit(IPS_EXPECTED_BIT, &ct->status);
576 ct->master = exp->master;
578 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
580 rcu_assign_pointer(help->helper, exp->helper);
583 #ifdef CONFIG_NF_CONNTRACK_MARK
584 ct->mark = exp->master->mark;
586 #ifdef CONFIG_NF_CONNTRACK_SECMARK
587 ct->secmark = exp->master->secmark;
589 nf_conntrack_get(&ct->master->ct_general);
590 NF_CT_STAT_INC(net, expect_new);
592 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
593 NF_CT_STAT_INC(net, new);
596 /* Overload tuple linked list to put us in unconfirmed list. */
597 hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
598 &net->ct.unconfirmed);
600 spin_unlock_bh(&nf_conntrack_lock);
604 exp->expectfn(ct, exp);
605 nf_ct_expect_put(exp);
608 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
611 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
612 static inline struct nf_conn *
613 resolve_normal_ct(struct net *net,
615 unsigned int dataoff,
618 struct nf_conntrack_l3proto *l3proto,
619 struct nf_conntrack_l4proto *l4proto,
621 enum ip_conntrack_info *ctinfo)
623 struct nf_conntrack_tuple tuple;
624 struct nf_conntrack_tuple_hash *h;
627 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
628 dataoff, l3num, protonum, &tuple, l3proto,
630 pr_debug("resolve_normal_ct: Can't get tuple\n");
634 /* look for tuple match */
635 h = nf_conntrack_find_get(net, &tuple);
637 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
643 ct = nf_ct_tuplehash_to_ctrack(h);
645 /* It exists; we have (non-exclusive) reference. */
646 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
647 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
648 /* Please set reply bit if this packet OK */
651 /* Once we've had two way comms, always ESTABLISHED. */
652 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
653 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
654 *ctinfo = IP_CT_ESTABLISHED;
655 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
656 pr_debug("nf_conntrack_in: related packet for %p\n",
658 *ctinfo = IP_CT_RELATED;
660 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
665 skb->nfct = &ct->ct_general;
666 skb->nfctinfo = *ctinfo;
671 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
675 enum ip_conntrack_info ctinfo;
676 struct nf_conntrack_l3proto *l3proto;
677 struct nf_conntrack_l4proto *l4proto;
678 unsigned int dataoff;
683 /* Previously seen (loopback or untracked)? Ignore. */
685 NF_CT_STAT_INC_ATOMIC(net, ignore);
689 /* rcu_read_lock()ed by nf_hook_slow */
690 l3proto = __nf_ct_l3proto_find(pf);
691 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
692 &dataoff, &protonum);
694 pr_debug("not prepared to track yet or error occured\n");
695 NF_CT_STAT_INC_ATOMIC(net, error);
696 NF_CT_STAT_INC_ATOMIC(net, invalid);
700 l4proto = __nf_ct_l4proto_find(pf, protonum);
702 /* It may be an special packet, error, unclean...
703 * inverse of the return code tells to the netfilter
704 * core what to do with the packet. */
705 if (l4proto->error != NULL) {
706 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
708 NF_CT_STAT_INC_ATOMIC(net, error);
709 NF_CT_STAT_INC_ATOMIC(net, invalid);
714 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
715 l3proto, l4proto, &set_reply, &ctinfo);
717 /* Not valid part of a connection */
718 NF_CT_STAT_INC_ATOMIC(net, invalid);
723 /* Too stressed to deal. */
724 NF_CT_STAT_INC_ATOMIC(net, drop);
728 NF_CT_ASSERT(skb->nfct);
730 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
732 /* Invalid: inverse of the return code tells
733 * the netfilter core what to do */
734 pr_debug("nf_conntrack_in: Can't track with proto module\n");
735 nf_conntrack_put(skb->nfct);
737 NF_CT_STAT_INC_ATOMIC(net, invalid);
741 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
742 nf_conntrack_event_cache(IPCT_STATUS, ct);
746 EXPORT_SYMBOL_GPL(nf_conntrack_in);
748 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
749 const struct nf_conntrack_tuple *orig)
754 ret = nf_ct_invert_tuple(inverse, orig,
755 __nf_ct_l3proto_find(orig->src.l3num),
756 __nf_ct_l4proto_find(orig->src.l3num,
757 orig->dst.protonum));
761 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
763 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
764 implicitly racy: see __nf_conntrack_confirm */
765 void nf_conntrack_alter_reply(struct nf_conn *ct,
766 const struct nf_conntrack_tuple *newreply)
768 struct nf_conn_help *help = nfct_help(ct);
770 /* Should be unconfirmed, so not in hash table yet */
771 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
773 pr_debug("Altering reply tuple of %p to ", ct);
774 nf_ct_dump_tuple(newreply);
776 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
777 if (ct->master || (help && !hlist_empty(&help->expectations)))
781 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
784 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
786 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
787 void __nf_ct_refresh_acct(struct nf_conn *ct,
788 enum ip_conntrack_info ctinfo,
789 const struct sk_buff *skb,
790 unsigned long extra_jiffies,
795 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
798 spin_lock_bh(&nf_conntrack_lock);
800 /* Only update if this is not a fixed timeout */
801 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
804 /* If not in hash table, timer will not be active yet */
805 if (!nf_ct_is_confirmed(ct)) {
806 ct->timeout.expires = extra_jiffies;
807 event = IPCT_REFRESH;
809 unsigned long newtime = jiffies + extra_jiffies;
811 /* Only update the timeout if the new timeout is at least
812 HZ jiffies from the old timeout. Need del_timer for race
813 avoidance (may already be dying). */
814 if (newtime - ct->timeout.expires >= HZ
815 && del_timer(&ct->timeout)) {
816 ct->timeout.expires = newtime;
817 add_timer(&ct->timeout);
818 event = IPCT_REFRESH;
824 struct nf_conn_counter *acct;
826 acct = nf_conn_acct_find(ct);
828 acct[CTINFO2DIR(ctinfo)].packets++;
829 acct[CTINFO2DIR(ctinfo)].bytes +=
830 skb->len - skb_network_offset(skb);
834 spin_unlock_bh(&nf_conntrack_lock);
836 /* must be unlocked when calling event cache */
838 nf_conntrack_event_cache(event, ct);
840 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
842 bool __nf_ct_kill_acct(struct nf_conn *ct,
843 enum ip_conntrack_info ctinfo,
844 const struct sk_buff *skb,
848 struct nf_conn_counter *acct;
850 spin_lock_bh(&nf_conntrack_lock);
851 acct = nf_conn_acct_find(ct);
853 acct[CTINFO2DIR(ctinfo)].packets++;
854 acct[CTINFO2DIR(ctinfo)].bytes +=
855 skb->len - skb_network_offset(skb);
857 spin_unlock_bh(&nf_conntrack_lock);
860 if (del_timer(&ct->timeout)) {
861 ct->timeout.function((unsigned long)ct);
866 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
868 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
870 #include <linux/netfilter/nfnetlink.h>
871 #include <linux/netfilter/nfnetlink_conntrack.h>
872 #include <linux/mutex.h>
874 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
875 * in ip_conntrack_core, since we don't want the protocols to autoload
876 * or depend on ctnetlink */
877 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
878 const struct nf_conntrack_tuple *tuple)
880 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
881 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
887 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
889 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
890 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
891 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
893 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
895 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
896 struct nf_conntrack_tuple *t)
898 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
901 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
902 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
906 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
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_node *n;
937 spin_lock_bh(&nf_conntrack_lock);
938 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
939 hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) {
940 ct = nf_ct_tuplehash_to_ctrack(h);
945 hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) {
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_all(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 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
998 free_pages((unsigned long)hash,
999 get_order(sizeof(struct hlist_head) * size));
1001 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1003 void nf_conntrack_flush(struct net *net, u32 pid, int report)
1005 struct __nf_ct_flush_report fr = {
1009 nf_ct_iterate_cleanup(net, kill_all, &fr);
1011 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1013 static void nf_conntrack_cleanup_init_net(void)
1015 nf_conntrack_helper_fini();
1016 nf_conntrack_proto_fini();
1017 kmem_cache_destroy(nf_conntrack_cachep);
1020 static void nf_conntrack_cleanup_net(struct net *net)
1022 nf_ct_event_cache_flush(net);
1023 nf_conntrack_ecache_fini(net);
1025 nf_conntrack_flush(net, 0, 0);
1026 if (atomic_read(&net->ct.count) != 0) {
1028 goto i_see_dead_people;
1030 /* wait until all references to nf_conntrack_untracked are dropped */
1031 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1034 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1035 nf_conntrack_htable_size);
1036 nf_conntrack_acct_fini(net);
1037 nf_conntrack_expect_fini(net);
1038 free_percpu(net->ct.stat);
1041 /* Mishearing the voices in his head, our hero wonders how he's
1042 supposed to kill the mall. */
1043 void nf_conntrack_cleanup(struct net *net)
1045 if (net_eq(net, &init_net))
1046 rcu_assign_pointer(ip_ct_attach, NULL);
1048 /* This makes sure all current packets have passed through
1049 netfilter framework. Roll on, two-stage module
1053 nf_conntrack_cleanup_net(net);
1055 if (net_eq(net, &init_net)) {
1056 rcu_assign_pointer(nf_ct_destroy, NULL);
1057 nf_conntrack_cleanup_init_net();
1061 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1063 struct hlist_head *hash;
1064 unsigned int size, i;
1068 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1069 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1070 get_order(sizeof(struct hlist_head)
1074 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1075 hash = vmalloc(sizeof(struct hlist_head) * size);
1079 for (i = 0; i < size; i++)
1080 INIT_HLIST_HEAD(&hash[i]);
1084 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1086 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1088 int i, bucket, vmalloced, old_vmalloced;
1089 unsigned int hashsize, old_size;
1091 struct hlist_head *hash, *old_hash;
1092 struct nf_conntrack_tuple_hash *h;
1094 /* On boot, we can set this without any fancy locking. */
1095 if (!nf_conntrack_htable_size)
1096 return param_set_uint(val, kp);
1098 hashsize = simple_strtoul(val, NULL, 0);
1102 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1106 /* We have to rehahs for the new table anyway, so we also can
1107 * use a newrandom seed */
1108 get_random_bytes(&rnd, 4);
1110 /* Lookups in the old hash might happen in parallel, which means we
1111 * might get false negatives during connection lookup. New connections
1112 * created because of a false negative won't make it into the hash
1113 * though since that required taking the lock.
1115 spin_lock_bh(&nf_conntrack_lock);
1116 for (i = 0; i < nf_conntrack_htable_size; i++) {
1117 while (!hlist_empty(&init_net.ct.hash[i])) {
1118 h = hlist_entry(init_net.ct.hash[i].first,
1119 struct nf_conntrack_tuple_hash, hnode);
1120 hlist_del_rcu(&h->hnode);
1121 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1122 hlist_add_head(&h->hnode, &hash[bucket]);
1125 old_size = nf_conntrack_htable_size;
1126 old_vmalloced = init_net.ct.hash_vmalloc;
1127 old_hash = init_net.ct.hash;
1129 nf_conntrack_htable_size = hashsize;
1130 init_net.ct.hash_vmalloc = vmalloced;
1131 init_net.ct.hash = hash;
1132 nf_conntrack_hash_rnd = rnd;
1133 spin_unlock_bh(&nf_conntrack_lock);
1135 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1138 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1140 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1141 &nf_conntrack_htable_size, 0600);
1143 static int nf_conntrack_init_init_net(void)
1148 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1149 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1150 if (!nf_conntrack_htable_size) {
1151 nf_conntrack_htable_size
1152 = (((num_physpages << PAGE_SHIFT) / 16384)
1153 / sizeof(struct hlist_head));
1154 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1155 nf_conntrack_htable_size = 16384;
1156 if (nf_conntrack_htable_size < 32)
1157 nf_conntrack_htable_size = 32;
1159 /* Use a max. factor of four by default to get the same max as
1160 * with the old struct list_heads. When a table size is given
1161 * we use the old value of 8 to avoid reducing the max.
1165 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1167 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1168 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1171 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1172 sizeof(struct nf_conn),
1174 if (!nf_conntrack_cachep) {
1175 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1180 ret = nf_conntrack_proto_init();
1184 ret = nf_conntrack_helper_init();
1191 nf_conntrack_proto_fini();
1193 kmem_cache_destroy(nf_conntrack_cachep);
1198 static int nf_conntrack_init_net(struct net *net)
1202 atomic_set(&net->ct.count, 0);
1203 INIT_HLIST_HEAD(&net->ct.unconfirmed);
1204 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1205 if (!net->ct.stat) {
1209 ret = nf_conntrack_ecache_init(net);
1212 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1213 &net->ct.hash_vmalloc);
1214 if (!net->ct.hash) {
1216 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1219 ret = nf_conntrack_expect_init(net);
1222 ret = nf_conntrack_acct_init(net);
1226 /* Set up fake conntrack:
1227 - to never be deleted, not in any hashes */
1228 #ifdef CONFIG_NET_NS
1229 nf_conntrack_untracked.ct_net = &init_net;
1231 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1232 /* - and look it like as a confirmed connection */
1233 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1238 nf_conntrack_expect_fini(net);
1240 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1241 nf_conntrack_htable_size);
1243 nf_conntrack_ecache_fini(net);
1245 free_percpu(net->ct.stat);
1250 int nf_conntrack_init(struct net *net)
1254 if (net_eq(net, &init_net)) {
1255 ret = nf_conntrack_init_init_net();
1259 ret = nf_conntrack_init_net(net);
1263 if (net_eq(net, &init_net)) {
1264 /* For use by REJECT target */
1265 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1266 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1271 if (net_eq(net, &init_net))
1272 nf_conntrack_cleanup_init_net();