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>
41 #define NF_CONNTRACK_VERSION "0.5.0"
43 DEFINE_RWLOCK(nf_conntrack_lock);
44 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
46 /* nf_conntrack_standalone needs this */
47 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
48 EXPORT_SYMBOL_GPL(nf_conntrack_count);
50 unsigned int nf_conntrack_htable_size __read_mostly;
51 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
53 int nf_conntrack_max __read_mostly;
54 EXPORT_SYMBOL_GPL(nf_conntrack_max);
56 struct hlist_head *nf_conntrack_hash __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
59 struct nf_conn nf_conntrack_untracked __read_mostly;
60 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
62 unsigned int nf_ct_log_invalid __read_mostly;
63 HLIST_HEAD(unconfirmed);
64 static int nf_conntrack_vmalloc __read_mostly;
65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
67 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
68 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
70 static int nf_conntrack_hash_rnd_initted;
71 static unsigned int nf_conntrack_hash_rnd;
73 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
74 unsigned int size, unsigned int rnd)
78 a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
79 (tuple->src.l3num << 16) | tuple->dst.protonum);
80 b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
81 ((__force __u16)tuple->src.u.all << 16) |
82 (__force __u16)tuple->dst.u.all);
84 return ((u64)jhash_2words(a, b, rnd) * size) >> 32;
87 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
89 return __hash_conntrack(tuple, nf_conntrack_htable_size,
90 nf_conntrack_hash_rnd);
94 nf_ct_get_tuple(const struct sk_buff *skb,
99 struct nf_conntrack_tuple *tuple,
100 const struct nf_conntrack_l3proto *l3proto,
101 const struct nf_conntrack_l4proto *l4proto)
103 NF_CT_TUPLE_U_BLANK(tuple);
105 tuple->src.l3num = l3num;
106 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
109 tuple->dst.protonum = protonum;
110 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
112 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
114 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
116 int nf_ct_get_tuplepr(const struct sk_buff *skb,
119 struct nf_conntrack_tuple *tuple)
121 struct nf_conntrack_l3proto *l3proto;
122 struct nf_conntrack_l4proto *l4proto;
123 unsigned int protoff;
129 l3proto = __nf_ct_l3proto_find(l3num);
130 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
131 if (ret != NF_ACCEPT) {
136 l4proto = __nf_ct_l4proto_find(l3num, protonum);
138 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
144 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
147 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
148 const struct nf_conntrack_tuple *orig,
149 const struct nf_conntrack_l3proto *l3proto,
150 const struct nf_conntrack_l4proto *l4proto)
152 NF_CT_TUPLE_U_BLANK(inverse);
154 inverse->src.l3num = orig->src.l3num;
155 if (l3proto->invert_tuple(inverse, orig) == 0)
158 inverse->dst.dir = !orig->dst.dir;
160 inverse->dst.protonum = orig->dst.protonum;
161 return l4proto->invert_tuple(inverse, orig);
163 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
166 clean_from_lists(struct nf_conn *ct)
168 pr_debug("clean_from_lists(%p)\n", ct);
169 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
170 hlist_del(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
172 /* Destroy all pending expectations */
173 nf_ct_remove_expectations(ct);
177 destroy_conntrack(struct nf_conntrack *nfct)
179 struct nf_conn *ct = (struct nf_conn *)nfct;
180 struct nf_conntrack_l4proto *l4proto;
182 pr_debug("destroy_conntrack(%p)\n", ct);
183 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
184 NF_CT_ASSERT(!timer_pending(&ct->timeout));
186 nf_conntrack_event(IPCT_DESTROY, ct);
187 set_bit(IPS_DYING_BIT, &ct->status);
189 /* To make sure we don't get any weird locking issues here:
190 * destroy_conntrack() MUST NOT be called with a write lock
191 * to nf_conntrack_lock!!! -HW */
193 l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
194 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
195 if (l4proto && l4proto->destroy)
196 l4proto->destroy(ct);
198 nf_ct_ext_destroy(ct);
202 write_lock_bh(&nf_conntrack_lock);
203 /* Expectations will have been removed in clean_from_lists,
204 * except TFTP can create an expectation on the first packet,
205 * before connection is in the list, so we need to clean here,
207 nf_ct_remove_expectations(ct);
209 /* We overload first tuple to link into unconfirmed list. */
210 if (!nf_ct_is_confirmed(ct)) {
211 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
212 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
215 NF_CT_STAT_INC(delete);
216 write_unlock_bh(&nf_conntrack_lock);
219 nf_ct_put(ct->master);
221 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
222 nf_conntrack_free(ct);
225 static void death_by_timeout(unsigned long ul_conntrack)
227 struct nf_conn *ct = (void *)ul_conntrack;
228 struct nf_conn_help *help = nfct_help(ct);
229 struct nf_conntrack_helper *helper;
233 helper = rcu_dereference(help->helper);
234 if (helper && helper->destroy)
239 write_lock_bh(&nf_conntrack_lock);
240 /* Inside lock so preempt is disabled on module removal path.
241 * Otherwise we can get spurious warnings. */
242 NF_CT_STAT_INC(delete_list);
243 clean_from_lists(ct);
244 write_unlock_bh(&nf_conntrack_lock);
248 struct nf_conntrack_tuple_hash *
249 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
250 const struct nf_conn *ignored_conntrack)
252 struct nf_conntrack_tuple_hash *h;
253 struct hlist_node *n;
254 unsigned int hash = hash_conntrack(tuple);
256 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
257 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
258 nf_ct_tuple_equal(tuple, &h->tuple)) {
259 NF_CT_STAT_INC(found);
262 NF_CT_STAT_INC(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(const struct nf_conntrack_tuple *tuple)
273 struct nf_conntrack_tuple_hash *h;
275 read_lock_bh(&nf_conntrack_lock);
276 h = __nf_conntrack_find(tuple, NULL);
278 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
279 read_unlock_bh(&nf_conntrack_lock);
283 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
285 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
287 unsigned int repl_hash)
289 hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
290 &nf_conntrack_hash[hash]);
291 hlist_add_head(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
292 &nf_conntrack_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 write_lock_bh(&nf_conntrack_lock);
303 __nf_conntrack_hash_insert(ct, hash, repl_hash);
304 write_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;
319 ct = nf_ct_get(skb, &ctinfo);
321 /* ipt_REJECT uses nf_conntrack_attach to attach related
322 ICMP/TCP RST packets in other direction. Actual packet
323 which created connection will be IP_CT_NEW or for an
324 expected connection, IP_CT_RELATED. */
325 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
328 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
329 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
331 /* We're not in hash table, and we refuse to set up related
332 connections for unconfirmed conns. But packet copies and
333 REJECT will give spurious warnings here. */
334 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
336 /* No external references means noone else could have
338 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
339 pr_debug("Confirming conntrack %p\n", ct);
341 write_lock_bh(&nf_conntrack_lock);
343 /* See if there's one in the list already, including reverse:
344 NAT could have grabbed it without realizing, since we're
345 not in the hash. If there is, we lost race. */
346 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
347 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
350 hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
351 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
355 /* Remove from unconfirmed list */
356 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
358 __nf_conntrack_hash_insert(ct, hash, repl_hash);
359 /* Timer relative to confirmation time, not original
360 setting time, otherwise we'd get timer wrap in
361 weird delay cases. */
362 ct->timeout.expires += jiffies;
363 add_timer(&ct->timeout);
364 atomic_inc(&ct->ct_general.use);
365 set_bit(IPS_CONFIRMED_BIT, &ct->status);
366 NF_CT_STAT_INC(insert);
367 write_unlock_bh(&nf_conntrack_lock);
368 help = nfct_help(ct);
369 if (help && help->helper)
370 nf_conntrack_event_cache(IPCT_HELPER, skb);
371 #ifdef CONFIG_NF_NAT_NEEDED
372 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
373 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
374 nf_conntrack_event_cache(IPCT_NATINFO, skb);
376 nf_conntrack_event_cache(master_ct(ct) ?
377 IPCT_RELATED : IPCT_NEW, skb);
381 NF_CT_STAT_INC(insert_failed);
382 write_unlock_bh(&nf_conntrack_lock);
385 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
387 /* Returns true if a connection correspondings to the tuple (required
390 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
391 const struct nf_conn *ignored_conntrack)
393 struct nf_conntrack_tuple_hash *h;
395 read_lock_bh(&nf_conntrack_lock);
396 h = __nf_conntrack_find(tuple, ignored_conntrack);
397 read_unlock_bh(&nf_conntrack_lock);
401 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
403 #define NF_CT_EVICTION_RANGE 8
405 /* There's a small race here where we may free a just-assured
406 connection. Too bad: we're in trouble anyway. */
407 static int early_drop(unsigned int hash)
409 /* Use oldest entry, which is roughly LRU */
410 struct nf_conntrack_tuple_hash *h;
411 struct nf_conn *ct = NULL, *tmp;
412 struct hlist_node *n;
413 unsigned int i, cnt = 0;
416 read_lock_bh(&nf_conntrack_lock);
417 for (i = 0; i < nf_conntrack_htable_size; i++) {
418 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
419 tmp = nf_ct_tuplehash_to_ctrack(h);
420 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
424 if (ct || cnt >= NF_CT_EVICTION_RANGE)
426 hash = (hash + 1) % nf_conntrack_htable_size;
429 atomic_inc(&ct->ct_general.use);
430 read_unlock_bh(&nf_conntrack_lock);
435 if (del_timer(&ct->timeout)) {
436 death_by_timeout((unsigned long)ct);
438 NF_CT_STAT_INC_ATOMIC(early_drop);
444 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
445 const struct nf_conntrack_tuple *repl)
447 struct nf_conn *conntrack = NULL;
449 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
450 get_random_bytes(&nf_conntrack_hash_rnd, 4);
451 nf_conntrack_hash_rnd_initted = 1;
454 /* We don't want any race condition at early drop stage */
455 atomic_inc(&nf_conntrack_count);
458 && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
459 unsigned int hash = hash_conntrack(orig);
460 if (!early_drop(hash)) {
461 atomic_dec(&nf_conntrack_count);
464 "nf_conntrack: table full, dropping"
466 return ERR_PTR(-ENOMEM);
470 conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
471 if (conntrack == NULL) {
472 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
473 atomic_dec(&nf_conntrack_count);
474 return ERR_PTR(-ENOMEM);
477 atomic_set(&conntrack->ct_general.use, 1);
478 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
479 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
480 /* Don't set timer yet: wait for confirmation */
481 setup_timer(&conntrack->timeout, death_by_timeout,
482 (unsigned long)conntrack);
486 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
488 void nf_conntrack_free(struct nf_conn *conntrack)
490 nf_ct_ext_free(conntrack);
491 kmem_cache_free(nf_conntrack_cachep, conntrack);
492 atomic_dec(&nf_conntrack_count);
494 EXPORT_SYMBOL_GPL(nf_conntrack_free);
496 /* Allocate a new conntrack: we return -ENOMEM if classification
497 failed due to stress. Otherwise it really is unclassifiable. */
498 static struct nf_conntrack_tuple_hash *
499 init_conntrack(const struct nf_conntrack_tuple *tuple,
500 struct nf_conntrack_l3proto *l3proto,
501 struct nf_conntrack_l4proto *l4proto,
503 unsigned int dataoff)
505 struct nf_conn *conntrack;
506 struct nf_conn_help *help;
507 struct nf_conntrack_tuple repl_tuple;
508 struct nf_conntrack_expect *exp;
510 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
511 pr_debug("Can't invert tuple.\n");
515 conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
516 if (conntrack == NULL || IS_ERR(conntrack)) {
517 pr_debug("Can't allocate conntrack.\n");
518 return (struct nf_conntrack_tuple_hash *)conntrack;
521 if (!l4proto->new(conntrack, skb, dataoff)) {
522 nf_conntrack_free(conntrack);
523 pr_debug("init conntrack: can't track with proto module\n");
527 write_lock_bh(&nf_conntrack_lock);
528 exp = nf_ct_find_expectation(tuple);
530 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
532 /* Welcome, Mr. Bond. We've been expecting you... */
533 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
534 conntrack->master = exp->master;
536 help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
538 rcu_assign_pointer(help->helper, exp->helper);
541 #ifdef CONFIG_NF_CONNTRACK_MARK
542 conntrack->mark = exp->master->mark;
544 #ifdef CONFIG_NF_CONNTRACK_SECMARK
545 conntrack->secmark = exp->master->secmark;
547 nf_conntrack_get(&conntrack->master->ct_general);
548 NF_CT_STAT_INC(expect_new);
550 struct nf_conntrack_helper *helper;
552 helper = __nf_ct_helper_find(&repl_tuple);
554 help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
556 rcu_assign_pointer(help->helper, helper);
561 /* Overload tuple linked list to put us in unconfirmed list. */
562 hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
565 write_unlock_bh(&nf_conntrack_lock);
569 exp->expectfn(conntrack, exp);
570 nf_ct_expect_put(exp);
573 return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
576 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
577 static inline struct nf_conn *
578 resolve_normal_ct(struct sk_buff *skb,
579 unsigned int dataoff,
582 struct nf_conntrack_l3proto *l3proto,
583 struct nf_conntrack_l4proto *l4proto,
585 enum ip_conntrack_info *ctinfo)
587 struct nf_conntrack_tuple tuple;
588 struct nf_conntrack_tuple_hash *h;
591 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
592 dataoff, l3num, protonum, &tuple, l3proto,
594 pr_debug("resolve_normal_ct: Can't get tuple\n");
598 /* look for tuple match */
599 h = nf_conntrack_find_get(&tuple);
601 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
607 ct = nf_ct_tuplehash_to_ctrack(h);
609 /* It exists; we have (non-exclusive) reference. */
610 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
611 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
612 /* Please set reply bit if this packet OK */
615 /* Once we've had two way comms, always ESTABLISHED. */
616 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
617 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
618 *ctinfo = IP_CT_ESTABLISHED;
619 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
620 pr_debug("nf_conntrack_in: related packet for %p\n",
622 *ctinfo = IP_CT_RELATED;
624 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
629 skb->nfct = &ct->ct_general;
630 skb->nfctinfo = *ctinfo;
635 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb)
638 enum ip_conntrack_info ctinfo;
639 struct nf_conntrack_l3proto *l3proto;
640 struct nf_conntrack_l4proto *l4proto;
641 unsigned int dataoff;
646 /* Previously seen (loopback or untracked)? Ignore. */
648 NF_CT_STAT_INC_ATOMIC(ignore);
652 /* rcu_read_lock()ed by nf_hook_slow */
653 l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
654 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
655 &dataoff, &protonum);
657 pr_debug("not prepared to track yet or error occured\n");
658 NF_CT_STAT_INC_ATOMIC(error);
659 NF_CT_STAT_INC_ATOMIC(invalid);
663 l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
665 /* It may be an special packet, error, unclean...
666 * inverse of the return code tells to the netfilter
667 * core what to do with the packet. */
668 if (l4proto->error != NULL &&
669 (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
670 NF_CT_STAT_INC_ATOMIC(error);
671 NF_CT_STAT_INC_ATOMIC(invalid);
675 ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
676 &set_reply, &ctinfo);
678 /* Not valid part of a connection */
679 NF_CT_STAT_INC_ATOMIC(invalid);
684 /* Too stressed to deal. */
685 NF_CT_STAT_INC_ATOMIC(drop);
689 NF_CT_ASSERT(skb->nfct);
691 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
693 /* Invalid: inverse of the return code tells
694 * the netfilter core what to do */
695 pr_debug("nf_conntrack_in: Can't track with proto module\n");
696 nf_conntrack_put(skb->nfct);
698 NF_CT_STAT_INC_ATOMIC(invalid);
702 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
703 nf_conntrack_event_cache(IPCT_STATUS, skb);
707 EXPORT_SYMBOL_GPL(nf_conntrack_in);
709 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
710 const struct nf_conntrack_tuple *orig)
715 ret = nf_ct_invert_tuple(inverse, orig,
716 __nf_ct_l3proto_find(orig->src.l3num),
717 __nf_ct_l4proto_find(orig->src.l3num,
718 orig->dst.protonum));
722 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
724 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
725 implicitly racy: see __nf_conntrack_confirm */
726 void nf_conntrack_alter_reply(struct nf_conn *ct,
727 const struct nf_conntrack_tuple *newreply)
729 struct nf_conn_help *help = nfct_help(ct);
730 struct nf_conntrack_helper *helper;
732 write_lock_bh(&nf_conntrack_lock);
733 /* Should be unconfirmed, so not in hash table yet */
734 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
736 pr_debug("Altering reply tuple of %p to ", ct);
737 NF_CT_DUMP_TUPLE(newreply);
739 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
740 if (ct->master || (help && help->expecting != 0))
743 helper = __nf_ct_helper_find(newreply);
744 if (helper == NULL) {
746 rcu_assign_pointer(help->helper, NULL);
751 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
755 memset(&help->help, 0, sizeof(help->help));
758 rcu_assign_pointer(help->helper, helper);
760 write_unlock_bh(&nf_conntrack_lock);
762 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
764 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
765 void __nf_ct_refresh_acct(struct nf_conn *ct,
766 enum ip_conntrack_info ctinfo,
767 const struct sk_buff *skb,
768 unsigned long extra_jiffies,
773 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
776 write_lock_bh(&nf_conntrack_lock);
778 /* Only update if this is not a fixed timeout */
779 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
782 /* If not in hash table, timer will not be active yet */
783 if (!nf_ct_is_confirmed(ct)) {
784 ct->timeout.expires = extra_jiffies;
785 event = IPCT_REFRESH;
787 unsigned long newtime = jiffies + extra_jiffies;
789 /* Only update the timeout if the new timeout is at least
790 HZ jiffies from the old timeout. Need del_timer for race
791 avoidance (may already be dying). */
792 if (newtime - ct->timeout.expires >= HZ
793 && del_timer(&ct->timeout)) {
794 ct->timeout.expires = newtime;
795 add_timer(&ct->timeout);
796 event = IPCT_REFRESH;
801 #ifdef CONFIG_NF_CT_ACCT
803 ct->counters[CTINFO2DIR(ctinfo)].packets++;
804 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
805 skb->len - skb_network_offset(skb);
807 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
808 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
809 event |= IPCT_COUNTER_FILLING;
813 write_unlock_bh(&nf_conntrack_lock);
815 /* must be unlocked when calling event cache */
817 nf_conntrack_event_cache(event, skb);
819 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
821 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
823 #include <linux/netfilter/nfnetlink.h>
824 #include <linux/netfilter/nfnetlink_conntrack.h>
825 #include <linux/mutex.h>
827 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
828 * in ip_conntrack_core, since we don't want the protocols to autoload
829 * or depend on ctnetlink */
830 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
831 const struct nf_conntrack_tuple *tuple)
833 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
834 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
840 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
842 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
843 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
844 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
846 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
848 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
849 struct nf_conntrack_tuple *t)
851 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
854 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
855 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
859 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
862 /* Used by ipt_REJECT and ip6t_REJECT. */
863 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
866 enum ip_conntrack_info ctinfo;
868 /* This ICMP is in reverse direction to the packet which caused it */
869 ct = nf_ct_get(skb, &ctinfo);
870 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
871 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
873 ctinfo = IP_CT_RELATED;
875 /* Attach to new skbuff, and increment count */
876 nskb->nfct = &ct->ct_general;
877 nskb->nfctinfo = ctinfo;
878 nf_conntrack_get(nskb->nfct);
882 do_iter(const struct nf_conntrack_tuple_hash *i,
883 int (*iter)(struct nf_conn *i, void *data),
886 return iter(nf_ct_tuplehash_to_ctrack(i), data);
889 /* Bring out ya dead! */
890 static struct nf_conn *
891 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
892 void *data, unsigned int *bucket)
894 struct nf_conntrack_tuple_hash *h;
896 struct hlist_node *n;
898 write_lock_bh(&nf_conntrack_lock);
899 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
900 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
901 ct = nf_ct_tuplehash_to_ctrack(h);
906 hlist_for_each_entry(h, n, &unconfirmed, hnode) {
907 ct = nf_ct_tuplehash_to_ctrack(h);
909 set_bit(IPS_DYING_BIT, &ct->status);
911 write_unlock_bh(&nf_conntrack_lock);
914 atomic_inc(&ct->ct_general.use);
915 write_unlock_bh(&nf_conntrack_lock);
920 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
923 unsigned int bucket = 0;
925 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
926 /* Time to push up daises... */
927 if (del_timer(&ct->timeout))
928 death_by_timeout((unsigned long)ct);
929 /* ... else the timer will get him soon. */
934 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
936 static int kill_all(struct nf_conn *i, void *data)
941 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
946 free_pages((unsigned long)hash,
947 get_order(sizeof(struct hlist_head) * size));
949 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
951 void nf_conntrack_flush(void)
953 nf_ct_iterate_cleanup(kill_all, NULL);
955 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
957 /* Mishearing the voices in his head, our hero wonders how he's
958 supposed to kill the mall. */
959 void nf_conntrack_cleanup(void)
961 rcu_assign_pointer(ip_ct_attach, NULL);
963 /* This makes sure all current packets have passed through
964 netfilter framework. Roll on, two-stage module
968 nf_ct_event_cache_flush();
970 nf_conntrack_flush();
971 if (atomic_read(&nf_conntrack_count) != 0) {
973 goto i_see_dead_people;
975 /* wait until all references to nf_conntrack_untracked are dropped */
976 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
979 rcu_assign_pointer(nf_ct_destroy, NULL);
981 kmem_cache_destroy(nf_conntrack_cachep);
982 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
983 nf_conntrack_htable_size);
985 nf_conntrack_proto_fini();
986 nf_conntrack_helper_fini();
987 nf_conntrack_expect_fini();
990 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
992 struct hlist_head *hash;
993 unsigned int size, i;
997 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
998 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
999 get_order(sizeof(struct hlist_head)
1003 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1004 hash = vmalloc(sizeof(struct hlist_head) * size);
1008 for (i = 0; i < size; i++)
1009 INIT_HLIST_HEAD(&hash[i]);
1013 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1015 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1017 int i, bucket, vmalloced, old_vmalloced;
1018 unsigned int hashsize, old_size;
1020 struct hlist_head *hash, *old_hash;
1021 struct nf_conntrack_tuple_hash *h;
1023 /* On boot, we can set this without any fancy locking. */
1024 if (!nf_conntrack_htable_size)
1025 return param_set_uint(val, kp);
1027 hashsize = simple_strtoul(val, NULL, 0);
1031 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1035 /* We have to rehahs for the new table anyway, so we also can
1036 * use a newrandom seed */
1037 get_random_bytes(&rnd, 4);
1039 write_lock_bh(&nf_conntrack_lock);
1040 for (i = 0; i < nf_conntrack_htable_size; i++) {
1041 while (!hlist_empty(&nf_conntrack_hash[i])) {
1042 h = hlist_entry(nf_conntrack_hash[i].first,
1043 struct nf_conntrack_tuple_hash, hnode);
1044 hlist_del(&h->hnode);
1045 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1046 hlist_add_head(&h->hnode, &hash[bucket]);
1049 old_size = nf_conntrack_htable_size;
1050 old_vmalloced = nf_conntrack_vmalloc;
1051 old_hash = nf_conntrack_hash;
1053 nf_conntrack_htable_size = hashsize;
1054 nf_conntrack_vmalloc = vmalloced;
1055 nf_conntrack_hash = hash;
1056 nf_conntrack_hash_rnd = rnd;
1057 write_unlock_bh(&nf_conntrack_lock);
1059 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1062 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1064 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1065 &nf_conntrack_htable_size, 0600);
1067 int __init nf_conntrack_init(void)
1072 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1073 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1074 if (!nf_conntrack_htable_size) {
1075 nf_conntrack_htable_size
1076 = (((num_physpages << PAGE_SHIFT) / 16384)
1077 / sizeof(struct hlist_head));
1078 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1079 nf_conntrack_htable_size = 16384;
1080 if (nf_conntrack_htable_size < 32)
1081 nf_conntrack_htable_size = 32;
1083 /* Use a max. factor of four by default to get the same max as
1084 * with the old struct list_heads. When a table size is given
1085 * we use the old value of 8 to avoid reducing the max.
1089 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1090 &nf_conntrack_vmalloc);
1091 if (!nf_conntrack_hash) {
1092 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1096 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1098 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1099 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1102 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1103 sizeof(struct nf_conn),
1105 if (!nf_conntrack_cachep) {
1106 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1110 ret = nf_conntrack_proto_init();
1112 goto err_free_conntrack_slab;
1114 ret = nf_conntrack_expect_init();
1116 goto out_fini_proto;
1118 ret = nf_conntrack_helper_init();
1120 goto out_fini_expect;
1122 /* For use by REJECT target */
1123 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1124 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1126 /* Set up fake conntrack:
1127 - to never be deleted, not in any hashes */
1128 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1129 /* - and look it like as a confirmed connection */
1130 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1135 nf_conntrack_expect_fini();
1137 nf_conntrack_proto_fini();
1138 err_free_conntrack_slab:
1139 kmem_cache_destroy(nf_conntrack_cachep);
1141 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1142 nf_conntrack_htable_size);