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>
42 #define NF_CONNTRACK_VERSION "0.5.0"
44 DEFINE_SPINLOCK(nf_conntrack_lock);
45 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
47 unsigned int nf_conntrack_htable_size __read_mostly;
48 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
50 int nf_conntrack_max __read_mostly;
51 EXPORT_SYMBOL_GPL(nf_conntrack_max);
53 struct nf_conn nf_conntrack_untracked __read_mostly;
54 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
56 unsigned int nf_ct_log_invalid __read_mostly;
57 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
59 static int nf_conntrack_hash_rnd_initted;
60 static unsigned int nf_conntrack_hash_rnd;
62 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
63 unsigned int size, unsigned int rnd)
68 /* The direction must be ignored, so we hash everything up to the
69 * destination ports (which is a multiple of 4) and treat the last
70 * three bytes manually.
72 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
73 h = jhash2((u32 *)tuple, n,
74 rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
75 tuple->dst.protonum));
77 return ((u64)h * size) >> 32;
80 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
82 return __hash_conntrack(tuple, nf_conntrack_htable_size,
83 nf_conntrack_hash_rnd);
87 nf_ct_get_tuple(const struct sk_buff *skb,
92 struct nf_conntrack_tuple *tuple,
93 const struct nf_conntrack_l3proto *l3proto,
94 const struct nf_conntrack_l4proto *l4proto)
96 memset(tuple, 0, sizeof(*tuple));
98 tuple->src.l3num = l3num;
99 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
102 tuple->dst.protonum = protonum;
103 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
105 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
107 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
109 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
110 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
112 struct nf_conntrack_l3proto *l3proto;
113 struct nf_conntrack_l4proto *l4proto;
114 unsigned int protoff;
120 l3proto = __nf_ct_l3proto_find(l3num);
121 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
122 if (ret != NF_ACCEPT) {
127 l4proto = __nf_ct_l4proto_find(l3num, protonum);
129 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
135 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
138 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
139 const struct nf_conntrack_tuple *orig,
140 const struct nf_conntrack_l3proto *l3proto,
141 const struct nf_conntrack_l4proto *l4proto)
143 memset(inverse, 0, sizeof(*inverse));
145 inverse->src.l3num = orig->src.l3num;
146 if (l3proto->invert_tuple(inverse, orig) == 0)
149 inverse->dst.dir = !orig->dst.dir;
151 inverse->dst.protonum = orig->dst.protonum;
152 return l4proto->invert_tuple(inverse, orig);
154 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
157 clean_from_lists(struct nf_conn *ct)
159 pr_debug("clean_from_lists(%p)\n", ct);
160 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
161 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
163 /* Destroy all pending expectations */
164 nf_ct_remove_expectations(ct);
168 destroy_conntrack(struct nf_conntrack *nfct)
170 struct nf_conn *ct = (struct nf_conn *)nfct;
171 struct net *net = nf_ct_net(ct);
172 struct nf_conntrack_l4proto *l4proto;
174 pr_debug("destroy_conntrack(%p)\n", ct);
175 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
176 NF_CT_ASSERT(!timer_pending(&ct->timeout));
178 nf_conntrack_event(IPCT_DESTROY, ct);
179 set_bit(IPS_DYING_BIT, &ct->status);
181 /* To make sure we don't get any weird locking issues here:
182 * destroy_conntrack() MUST NOT be called with a write lock
183 * to nf_conntrack_lock!!! -HW */
185 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
186 if (l4proto && l4proto->destroy)
187 l4proto->destroy(ct);
191 spin_lock_bh(&nf_conntrack_lock);
192 /* Expectations will have been removed in clean_from_lists,
193 * except TFTP can create an expectation on the first packet,
194 * before connection is in the list, so we need to clean here,
196 nf_ct_remove_expectations(ct);
198 /* We overload first tuple to link into unconfirmed list. */
199 if (!nf_ct_is_confirmed(ct)) {
200 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
201 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
204 NF_CT_STAT_INC(net, delete);
205 spin_unlock_bh(&nf_conntrack_lock);
208 nf_ct_put(ct->master);
210 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
211 nf_conntrack_free(ct);
214 static void death_by_timeout(unsigned long ul_conntrack)
216 struct nf_conn *ct = (void *)ul_conntrack;
217 struct net *net = nf_ct_net(ct);
218 struct nf_conn_help *help = nfct_help(ct);
219 struct nf_conntrack_helper *helper;
223 helper = rcu_dereference(help->helper);
224 if (helper && helper->destroy)
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);
238 struct nf_conntrack_tuple_hash *
239 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
241 struct nf_conntrack_tuple_hash *h;
242 struct hlist_node *n;
243 unsigned int hash = hash_conntrack(tuple);
245 /* Disable BHs the entire time since we normally need to disable them
246 * at least once for the stats anyway.
249 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
250 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
251 NF_CT_STAT_INC(net, found);
255 NF_CT_STAT_INC(net, searched);
261 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
263 /* Find a connection corresponding to a tuple. */
264 struct nf_conntrack_tuple_hash *
265 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
267 struct nf_conntrack_tuple_hash *h;
271 h = __nf_conntrack_find(net, tuple);
273 ct = nf_ct_tuplehash_to_ctrack(h);
274 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
281 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
283 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
285 unsigned int repl_hash)
287 struct net *net = nf_ct_net(ct);
289 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
290 &net->ct.hash[hash]);
291 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
292 &net->ct.hash[repl_hash]);
295 void nf_conntrack_hash_insert(struct nf_conn *ct)
297 unsigned int hash, repl_hash;
299 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
300 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
302 spin_lock_bh(&nf_conntrack_lock);
303 __nf_conntrack_hash_insert(ct, hash, repl_hash);
304 spin_unlock_bh(&nf_conntrack_lock);
306 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
308 /* Confirm a connection given skb; places it in hash table */
310 __nf_conntrack_confirm(struct sk_buff *skb)
312 unsigned int hash, repl_hash;
313 struct nf_conntrack_tuple_hash *h;
315 struct nf_conn_help *help;
316 struct hlist_node *n;
317 enum ip_conntrack_info ctinfo;
320 ct = nf_ct_get(skb, &ctinfo);
323 /* ipt_REJECT uses nf_conntrack_attach to attach related
324 ICMP/TCP RST packets in other direction. Actual packet
325 which created connection will be IP_CT_NEW or for an
326 expected connection, IP_CT_RELATED. */
327 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
330 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
331 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
333 /* We're not in hash table, and we refuse to set up related
334 connections for unconfirmed conns. But packet copies and
335 REJECT will give spurious warnings here. */
336 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
338 /* No external references means noone else could have
340 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
341 pr_debug("Confirming conntrack %p\n", ct);
343 spin_lock_bh(&nf_conntrack_lock);
345 /* See if there's one in the list already, including reverse:
346 NAT could have grabbed it without realizing, since we're
347 not in the hash. If there is, we lost race. */
348 hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode)
349 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
352 hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode)
353 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
357 /* Remove from unconfirmed list */
358 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
360 __nf_conntrack_hash_insert(ct, hash, repl_hash);
361 /* Timer relative to confirmation time, not original
362 setting time, otherwise we'd get timer wrap in
363 weird delay cases. */
364 ct->timeout.expires += jiffies;
365 add_timer(&ct->timeout);
366 atomic_inc(&ct->ct_general.use);
367 set_bit(IPS_CONFIRMED_BIT, &ct->status);
368 NF_CT_STAT_INC(net, insert);
369 spin_unlock_bh(&nf_conntrack_lock);
370 help = nfct_help(ct);
371 if (help && help->helper)
372 nf_conntrack_event_cache(IPCT_HELPER, ct);
373 #ifdef CONFIG_NF_NAT_NEEDED
374 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
375 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
376 nf_conntrack_event_cache(IPCT_NATINFO, ct);
378 nf_conntrack_event_cache(master_ct(ct) ?
379 IPCT_RELATED : IPCT_NEW, ct);
383 NF_CT_STAT_INC(net, insert_failed);
384 spin_unlock_bh(&nf_conntrack_lock);
387 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
389 /* Returns true if a connection correspondings to the tuple (required
392 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
393 const struct nf_conn *ignored_conntrack)
395 struct net *net = nf_ct_net(ignored_conntrack);
396 struct nf_conntrack_tuple_hash *h;
397 struct hlist_node *n;
398 unsigned int hash = hash_conntrack(tuple);
400 /* Disable BHs the entire time since we need to disable them at
401 * least once for the stats anyway.
404 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
405 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
406 nf_ct_tuple_equal(tuple, &h->tuple)) {
407 NF_CT_STAT_INC(net, found);
408 rcu_read_unlock_bh();
411 NF_CT_STAT_INC(net, searched);
413 rcu_read_unlock_bh();
417 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
419 #define NF_CT_EVICTION_RANGE 8
421 /* There's a small race here where we may free a just-assured
422 connection. Too bad: we're in trouble anyway. */
423 static noinline int early_drop(struct net *net, unsigned int hash)
425 /* Use oldest entry, which is roughly LRU */
426 struct nf_conntrack_tuple_hash *h;
427 struct nf_conn *ct = NULL, *tmp;
428 struct hlist_node *n;
429 unsigned int i, cnt = 0;
433 for (i = 0; i < nf_conntrack_htable_size; i++) {
434 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash],
436 tmp = nf_ct_tuplehash_to_ctrack(h);
437 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
442 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
444 if (ct || cnt >= NF_CT_EVICTION_RANGE)
446 hash = (hash + 1) % nf_conntrack_htable_size;
453 if (del_timer(&ct->timeout)) {
454 death_by_timeout((unsigned long)ct);
456 NF_CT_STAT_INC_ATOMIC(net, early_drop);
462 struct nf_conn *nf_conntrack_alloc(struct net *net,
463 const struct nf_conntrack_tuple *orig,
464 const struct nf_conntrack_tuple *repl,
467 struct nf_conn *ct = NULL;
469 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
470 get_random_bytes(&nf_conntrack_hash_rnd, 4);
471 nf_conntrack_hash_rnd_initted = 1;
474 /* We don't want any race condition at early drop stage */
475 atomic_inc(&net->ct.count);
477 if (nf_conntrack_max &&
478 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
479 unsigned int hash = hash_conntrack(orig);
480 if (!early_drop(net, hash)) {
481 atomic_dec(&net->ct.count);
484 "nf_conntrack: table full, dropping"
486 return ERR_PTR(-ENOMEM);
490 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
492 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
493 atomic_dec(&net->ct.count);
494 return ERR_PTR(-ENOMEM);
497 atomic_set(&ct->ct_general.use, 1);
498 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
499 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
500 /* Don't set timer yet: wait for confirmation */
501 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
505 INIT_RCU_HEAD(&ct->rcu);
509 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
511 static void nf_conntrack_free_rcu(struct rcu_head *head)
513 struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
514 struct net *net = nf_ct_net(ct);
517 kmem_cache_free(nf_conntrack_cachep, ct);
518 atomic_dec(&net->ct.count);
521 void nf_conntrack_free(struct nf_conn *ct)
523 nf_ct_ext_destroy(ct);
524 call_rcu(&ct->rcu, nf_conntrack_free_rcu);
526 EXPORT_SYMBOL_GPL(nf_conntrack_free);
528 /* Allocate a new conntrack: we return -ENOMEM if classification
529 failed due to stress. Otherwise it really is unclassifiable. */
530 static struct nf_conntrack_tuple_hash *
531 init_conntrack(struct net *net,
532 const struct nf_conntrack_tuple *tuple,
533 struct nf_conntrack_l3proto *l3proto,
534 struct nf_conntrack_l4proto *l4proto,
536 unsigned int dataoff)
539 struct nf_conn_help *help;
540 struct nf_conntrack_tuple repl_tuple;
541 struct nf_conntrack_expect *exp;
543 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
544 pr_debug("Can't invert tuple.\n");
548 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
549 if (ct == NULL || IS_ERR(ct)) {
550 pr_debug("Can't allocate conntrack.\n");
551 return (struct nf_conntrack_tuple_hash *)ct;
554 if (!l4proto->new(ct, skb, dataoff)) {
555 nf_conntrack_free(ct);
556 pr_debug("init conntrack: can't track with proto module\n");
560 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
562 spin_lock_bh(&nf_conntrack_lock);
563 exp = nf_ct_find_expectation(net, tuple);
565 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
567 /* Welcome, Mr. Bond. We've been expecting you... */
568 __set_bit(IPS_EXPECTED_BIT, &ct->status);
569 ct->master = exp->master;
571 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
573 rcu_assign_pointer(help->helper, exp->helper);
576 #ifdef CONFIG_NF_CONNTRACK_MARK
577 ct->mark = exp->master->mark;
579 #ifdef CONFIG_NF_CONNTRACK_SECMARK
580 ct->secmark = exp->master->secmark;
582 nf_conntrack_get(&ct->master->ct_general);
583 NF_CT_STAT_INC(net, expect_new);
585 struct nf_conntrack_helper *helper;
587 helper = __nf_ct_helper_find(&repl_tuple);
589 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
591 rcu_assign_pointer(help->helper, helper);
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);
769 struct nf_conntrack_helper *helper;
771 /* Should be unconfirmed, so not in hash table yet */
772 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
774 pr_debug("Altering reply tuple of %p to ", ct);
775 nf_ct_dump_tuple(newreply);
777 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
778 if (ct->master || (help && !hlist_empty(&help->expectations)))
782 helper = __nf_ct_helper_find(newreply);
783 if (helper == NULL) {
785 rcu_assign_pointer(help->helper, NULL);
790 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
794 memset(&help->help, 0, sizeof(help->help));
797 rcu_assign_pointer(help->helper, helper);
801 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
803 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
804 void __nf_ct_refresh_acct(struct nf_conn *ct,
805 enum ip_conntrack_info ctinfo,
806 const struct sk_buff *skb,
807 unsigned long extra_jiffies,
812 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
815 spin_lock_bh(&nf_conntrack_lock);
817 /* Only update if this is not a fixed timeout */
818 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
821 /* If not in hash table, timer will not be active yet */
822 if (!nf_ct_is_confirmed(ct)) {
823 ct->timeout.expires = extra_jiffies;
824 event = IPCT_REFRESH;
826 unsigned long newtime = jiffies + extra_jiffies;
828 /* Only update the timeout if the new timeout is at least
829 HZ jiffies from the old timeout. Need del_timer for race
830 avoidance (may already be dying). */
831 if (newtime - ct->timeout.expires >= HZ
832 && del_timer(&ct->timeout)) {
833 ct->timeout.expires = newtime;
834 add_timer(&ct->timeout);
835 event = IPCT_REFRESH;
841 struct nf_conn_counter *acct;
843 acct = nf_conn_acct_find(ct);
845 acct[CTINFO2DIR(ctinfo)].packets++;
846 acct[CTINFO2DIR(ctinfo)].bytes +=
847 skb->len - skb_network_offset(skb);
851 spin_unlock_bh(&nf_conntrack_lock);
853 /* must be unlocked when calling event cache */
855 nf_conntrack_event_cache(event, ct);
857 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
859 bool __nf_ct_kill_acct(struct nf_conn *ct,
860 enum ip_conntrack_info ctinfo,
861 const struct sk_buff *skb,
865 struct nf_conn_counter *acct;
867 spin_lock_bh(&nf_conntrack_lock);
868 acct = nf_conn_acct_find(ct);
870 acct[CTINFO2DIR(ctinfo)].packets++;
871 acct[CTINFO2DIR(ctinfo)].bytes +=
872 skb->len - skb_network_offset(skb);
874 spin_unlock_bh(&nf_conntrack_lock);
877 if (del_timer(&ct->timeout)) {
878 ct->timeout.function((unsigned long)ct);
883 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
885 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
887 #include <linux/netfilter/nfnetlink.h>
888 #include <linux/netfilter/nfnetlink_conntrack.h>
889 #include <linux/mutex.h>
891 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
892 * in ip_conntrack_core, since we don't want the protocols to autoload
893 * or depend on ctnetlink */
894 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
895 const struct nf_conntrack_tuple *tuple)
897 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
898 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
904 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
906 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
907 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
908 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
910 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
912 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
913 struct nf_conntrack_tuple *t)
915 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
918 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
919 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
923 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
926 /* Used by ipt_REJECT and ip6t_REJECT. */
927 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
930 enum ip_conntrack_info ctinfo;
932 /* This ICMP is in reverse direction to the packet which caused it */
933 ct = nf_ct_get(skb, &ctinfo);
934 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
935 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
937 ctinfo = IP_CT_RELATED;
939 /* Attach to new skbuff, and increment count */
940 nskb->nfct = &ct->ct_general;
941 nskb->nfctinfo = ctinfo;
942 nf_conntrack_get(nskb->nfct);
945 /* Bring out ya dead! */
946 static struct nf_conn *
947 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
948 void *data, unsigned int *bucket)
950 struct nf_conntrack_tuple_hash *h;
952 struct hlist_node *n;
954 spin_lock_bh(&nf_conntrack_lock);
955 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
956 hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) {
957 ct = nf_ct_tuplehash_to_ctrack(h);
962 hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) {
963 ct = nf_ct_tuplehash_to_ctrack(h);
965 set_bit(IPS_DYING_BIT, &ct->status);
967 spin_unlock_bh(&nf_conntrack_lock);
970 atomic_inc(&ct->ct_general.use);
971 spin_unlock_bh(&nf_conntrack_lock);
975 void nf_ct_iterate_cleanup(struct net *net,
976 int (*iter)(struct nf_conn *i, void *data),
980 unsigned int bucket = 0;
982 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
983 /* Time to push up daises... */
984 if (del_timer(&ct->timeout))
985 death_by_timeout((unsigned long)ct);
986 /* ... else the timer will get him soon. */
991 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
993 static int kill_all(struct nf_conn *i, void *data)
998 void nf_ct_free_hashtable(struct hlist_head *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(struct net *net)
1010 nf_ct_iterate_cleanup(net, kill_all, NULL);
1012 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1014 /* Mishearing the voices in his head, our hero wonders how he's
1015 supposed to kill the mall. */
1016 void nf_conntrack_cleanup(struct net *net)
1018 rcu_assign_pointer(ip_ct_attach, NULL);
1020 /* This makes sure all current packets have passed through
1021 netfilter framework. Roll on, two-stage module
1025 nf_ct_event_cache_flush(net);
1026 nf_conntrack_ecache_fini(net);
1028 nf_conntrack_flush(net);
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 rcu_assign_pointer(nf_ct_destroy, NULL);
1039 kmem_cache_destroy(nf_conntrack_cachep);
1040 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1041 nf_conntrack_htable_size);
1043 nf_conntrack_acct_fini();
1044 nf_conntrack_expect_fini(net);
1045 free_percpu(net->ct.stat);
1046 nf_conntrack_helper_fini();
1047 nf_conntrack_proto_fini();
1050 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1052 struct hlist_head *hash;
1053 unsigned int size, i;
1057 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1058 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1059 get_order(sizeof(struct hlist_head)
1063 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1064 hash = vmalloc(sizeof(struct hlist_head) * size);
1068 for (i = 0; i < size; i++)
1069 INIT_HLIST_HEAD(&hash[i]);
1073 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1075 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1077 int i, bucket, vmalloced, old_vmalloced;
1078 unsigned int hashsize, old_size;
1080 struct hlist_head *hash, *old_hash;
1081 struct nf_conntrack_tuple_hash *h;
1083 /* On boot, we can set this without any fancy locking. */
1084 if (!nf_conntrack_htable_size)
1085 return param_set_uint(val, kp);
1087 hashsize = simple_strtoul(val, NULL, 0);
1091 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1095 /* We have to rehahs for the new table anyway, so we also can
1096 * use a newrandom seed */
1097 get_random_bytes(&rnd, 4);
1099 /* Lookups in the old hash might happen in parallel, which means we
1100 * might get false negatives during connection lookup. New connections
1101 * created because of a false negative won't make it into the hash
1102 * though since that required taking the lock.
1104 spin_lock_bh(&nf_conntrack_lock);
1105 for (i = 0; i < nf_conntrack_htable_size; i++) {
1106 while (!hlist_empty(&init_net.ct.hash[i])) {
1107 h = hlist_entry(init_net.ct.hash[i].first,
1108 struct nf_conntrack_tuple_hash, hnode);
1109 hlist_del_rcu(&h->hnode);
1110 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1111 hlist_add_head(&h->hnode, &hash[bucket]);
1114 old_size = nf_conntrack_htable_size;
1115 old_vmalloced = init_net.ct.hash_vmalloc;
1116 old_hash = init_net.ct.hash;
1118 nf_conntrack_htable_size = hashsize;
1119 init_net.ct.hash_vmalloc = vmalloced;
1120 init_net.ct.hash = hash;
1121 nf_conntrack_hash_rnd = rnd;
1122 spin_unlock_bh(&nf_conntrack_lock);
1124 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1127 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1129 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1130 &nf_conntrack_htable_size, 0600);
1132 int nf_conntrack_init(struct net *net)
1137 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1138 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1139 if (!nf_conntrack_htable_size) {
1140 nf_conntrack_htable_size
1141 = (((num_physpages << PAGE_SHIFT) / 16384)
1142 / sizeof(struct hlist_head));
1143 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1144 nf_conntrack_htable_size = 16384;
1145 if (nf_conntrack_htable_size < 32)
1146 nf_conntrack_htable_size = 32;
1148 /* Use a max. factor of four by default to get the same max as
1149 * with the old struct list_heads. When a table size is given
1150 * we use the old value of 8 to avoid reducing the max.
1154 atomic_set(&net->ct.count, 0);
1155 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1158 ret = nf_conntrack_ecache_init(net);
1161 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1162 &net->ct.hash_vmalloc);
1163 if (!net->ct.hash) {
1164 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1167 INIT_HLIST_HEAD(&net->ct.unconfirmed);
1169 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1171 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1172 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1175 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1176 sizeof(struct nf_conn),
1178 if (!nf_conntrack_cachep) {
1179 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1183 ret = nf_conntrack_proto_init();
1185 goto err_free_conntrack_slab;
1187 ret = nf_conntrack_expect_init(net);
1189 goto out_fini_proto;
1191 ret = nf_conntrack_helper_init();
1193 goto out_fini_expect;
1195 ret = nf_conntrack_acct_init();
1197 goto out_fini_helper;
1199 /* For use by REJECT target */
1200 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1201 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1203 /* Set up fake conntrack:
1204 - to never be deleted, not in any hashes */
1205 #ifdef CONFIG_NET_NS
1206 nf_conntrack_untracked.ct_net = &init_net;
1208 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1209 /* - and look it like as a confirmed connection */
1210 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1215 nf_conntrack_helper_fini();
1217 nf_conntrack_expect_fini(net);
1219 nf_conntrack_proto_fini();
1220 err_free_conntrack_slab:
1221 kmem_cache_destroy(nf_conntrack_cachep);
1223 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1224 nf_conntrack_htable_size);
1226 nf_conntrack_ecache_fini(net);
1228 free_percpu(net->ct.stat);