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 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
60 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
62 static int nf_conntrack_hash_rnd_initted;
63 static unsigned int nf_conntrack_hash_rnd;
65 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
66 unsigned int size, unsigned int rnd)
71 /* The direction must be ignored, so we hash everything up to the
72 * destination ports (which is a multiple of 4) and treat the last
73 * three bytes manually.
75 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
76 h = jhash2((u32 *)tuple, n,
77 rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
78 tuple->dst.protonum));
80 return ((u64)h * size) >> 32;
83 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
85 return __hash_conntrack(tuple, nf_conntrack_htable_size,
86 nf_conntrack_hash_rnd);
90 nf_ct_get_tuple(const struct sk_buff *skb,
95 struct nf_conntrack_tuple *tuple,
96 const struct nf_conntrack_l3proto *l3proto,
97 const struct nf_conntrack_l4proto *l4proto)
99 memset(tuple, 0, sizeof(*tuple));
101 tuple->src.l3num = l3num;
102 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
105 tuple->dst.protonum = protonum;
106 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
108 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
110 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
112 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
113 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
115 struct nf_conntrack_l3proto *l3proto;
116 struct nf_conntrack_l4proto *l4proto;
117 unsigned int protoff;
123 l3proto = __nf_ct_l3proto_find(l3num);
124 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
125 if (ret != NF_ACCEPT) {
130 l4proto = __nf_ct_l4proto_find(l3num, protonum);
132 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
138 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
141 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
142 const struct nf_conntrack_tuple *orig,
143 const struct nf_conntrack_l3proto *l3proto,
144 const struct nf_conntrack_l4proto *l4proto)
146 memset(inverse, 0, sizeof(*inverse));
148 inverse->src.l3num = orig->src.l3num;
149 if (l3proto->invert_tuple(inverse, orig) == 0)
152 inverse->dst.dir = !orig->dst.dir;
154 inverse->dst.protonum = orig->dst.protonum;
155 return l4proto->invert_tuple(inverse, orig);
157 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
160 clean_from_lists(struct nf_conn *ct)
162 pr_debug("clean_from_lists(%p)\n", ct);
163 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
164 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
166 /* Destroy all pending expectations */
167 nf_ct_remove_expectations(ct);
171 destroy_conntrack(struct nf_conntrack *nfct)
173 struct nf_conn *ct = (struct nf_conn *)nfct;
174 struct nf_conntrack_l4proto *l4proto;
176 pr_debug("destroy_conntrack(%p)\n", ct);
177 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
178 NF_CT_ASSERT(!timer_pending(&ct->timeout));
180 nf_conntrack_event(IPCT_DESTROY, ct);
181 set_bit(IPS_DYING_BIT, &ct->status);
183 /* To make sure we don't get any weird locking issues here:
184 * destroy_conntrack() MUST NOT be called with a write lock
185 * to nf_conntrack_lock!!! -HW */
187 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
188 if (l4proto && l4proto->destroy)
189 l4proto->destroy(ct);
193 spin_lock_bh(&nf_conntrack_lock);
194 /* Expectations will have been removed in clean_from_lists,
195 * except TFTP can create an expectation on the first packet,
196 * before connection is in the list, so we need to clean here,
198 nf_ct_remove_expectations(ct);
200 /* We overload first tuple to link into unconfirmed list. */
201 if (!nf_ct_is_confirmed(ct)) {
202 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
203 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
206 NF_CT_STAT_INC(delete);
207 spin_unlock_bh(&nf_conntrack_lock);
210 nf_ct_put(ct->master);
212 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
213 nf_conntrack_free(ct);
216 static void death_by_timeout(unsigned long ul_conntrack)
218 struct nf_conn *ct = (void *)ul_conntrack;
219 struct nf_conn_help *help = nfct_help(ct);
220 struct nf_conntrack_helper *helper;
224 helper = rcu_dereference(help->helper);
225 if (helper && helper->destroy)
230 spin_lock_bh(&nf_conntrack_lock);
231 /* Inside lock so preempt is disabled on module removal path.
232 * Otherwise we can get spurious warnings. */
233 NF_CT_STAT_INC(delete_list);
234 clean_from_lists(ct);
235 spin_unlock_bh(&nf_conntrack_lock);
239 struct nf_conntrack_tuple_hash *
240 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
242 struct nf_conntrack_tuple_hash *h;
243 struct hlist_node *n;
244 unsigned int hash = hash_conntrack(tuple);
246 /* Disable BHs the entire time since we normally need to disable them
247 * at least once for the stats anyway.
250 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
251 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
252 NF_CT_STAT_INC(found);
256 NF_CT_STAT_INC(searched);
262 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
264 /* Find a connection corresponding to a tuple. */
265 struct nf_conntrack_tuple_hash *
266 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
268 struct nf_conntrack_tuple_hash *h;
272 h = __nf_conntrack_find(net, tuple);
274 ct = nf_ct_tuplehash_to_ctrack(h);
275 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
282 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
284 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
286 unsigned int repl_hash)
288 struct net *net = nf_ct_net(ct);
290 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
291 &net->ct.hash[hash]);
292 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
293 &net->ct.hash[repl_hash]);
296 void nf_conntrack_hash_insert(struct nf_conn *ct)
298 unsigned int hash, repl_hash;
300 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
301 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
303 spin_lock_bh(&nf_conntrack_lock);
304 __nf_conntrack_hash_insert(ct, hash, repl_hash);
305 spin_unlock_bh(&nf_conntrack_lock);
307 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
309 /* Confirm a connection given skb; places it in hash table */
311 __nf_conntrack_confirm(struct sk_buff *skb)
313 unsigned int hash, repl_hash;
314 struct nf_conntrack_tuple_hash *h;
316 struct nf_conn_help *help;
317 struct hlist_node *n;
318 enum ip_conntrack_info ctinfo;
321 ct = nf_ct_get(skb, &ctinfo);
324 /* ipt_REJECT uses nf_conntrack_attach to attach related
325 ICMP/TCP RST packets in other direction. Actual packet
326 which created connection will be IP_CT_NEW or for an
327 expected connection, IP_CT_RELATED. */
328 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
331 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
332 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
334 /* We're not in hash table, and we refuse to set up related
335 connections for unconfirmed conns. But packet copies and
336 REJECT will give spurious warnings here. */
337 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
339 /* No external references means noone else could have
341 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
342 pr_debug("Confirming conntrack %p\n", ct);
344 spin_lock_bh(&nf_conntrack_lock);
346 /* See if there's one in the list already, including reverse:
347 NAT could have grabbed it without realizing, since we're
348 not in the hash. If there is, we lost race. */
349 hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode)
350 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
353 hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode)
354 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
358 /* Remove from unconfirmed list */
359 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
361 __nf_conntrack_hash_insert(ct, hash, repl_hash);
362 /* Timer relative to confirmation time, not original
363 setting time, otherwise we'd get timer wrap in
364 weird delay cases. */
365 ct->timeout.expires += jiffies;
366 add_timer(&ct->timeout);
367 atomic_inc(&ct->ct_general.use);
368 set_bit(IPS_CONFIRMED_BIT, &ct->status);
369 NF_CT_STAT_INC(insert);
370 spin_unlock_bh(&nf_conntrack_lock);
371 help = nfct_help(ct);
372 if (help && help->helper)
373 nf_conntrack_event_cache(IPCT_HELPER, skb);
374 #ifdef CONFIG_NF_NAT_NEEDED
375 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
376 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
377 nf_conntrack_event_cache(IPCT_NATINFO, skb);
379 nf_conntrack_event_cache(master_ct(ct) ?
380 IPCT_RELATED : IPCT_NEW, skb);
384 NF_CT_STAT_INC(insert_failed);
385 spin_unlock_bh(&nf_conntrack_lock);
388 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
390 /* Returns true if a connection correspondings to the tuple (required
393 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
394 const struct nf_conn *ignored_conntrack)
396 struct net *net = nf_ct_net(ignored_conntrack);
397 struct nf_conntrack_tuple_hash *h;
398 struct hlist_node *n;
399 unsigned int hash = hash_conntrack(tuple);
401 /* Disable BHs the entire time since we need to disable them at
402 * least once for the stats anyway.
405 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
406 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
407 nf_ct_tuple_equal(tuple, &h->tuple)) {
408 NF_CT_STAT_INC(found);
409 rcu_read_unlock_bh();
412 NF_CT_STAT_INC(searched);
414 rcu_read_unlock_bh();
418 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
420 #define NF_CT_EVICTION_RANGE 8
422 /* There's a small race here where we may free a just-assured
423 connection. Too bad: we're in trouble anyway. */
424 static noinline int early_drop(struct net *net, unsigned int hash)
426 /* Use oldest entry, which is roughly LRU */
427 struct nf_conntrack_tuple_hash *h;
428 struct nf_conn *ct = NULL, *tmp;
429 struct hlist_node *n;
430 unsigned int i, cnt = 0;
434 for (i = 0; i < nf_conntrack_htable_size; i++) {
435 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash],
437 tmp = nf_ct_tuplehash_to_ctrack(h);
438 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
443 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
445 if (ct || cnt >= NF_CT_EVICTION_RANGE)
447 hash = (hash + 1) % nf_conntrack_htable_size;
454 if (del_timer(&ct->timeout)) {
455 death_by_timeout((unsigned long)ct);
457 NF_CT_STAT_INC_ATOMIC(early_drop);
463 struct nf_conn *nf_conntrack_alloc(struct net *net,
464 const struct nf_conntrack_tuple *orig,
465 const struct nf_conntrack_tuple *repl,
468 struct nf_conn *ct = NULL;
470 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
471 get_random_bytes(&nf_conntrack_hash_rnd, 4);
472 nf_conntrack_hash_rnd_initted = 1;
475 /* We don't want any race condition at early drop stage */
476 atomic_inc(&net->ct.count);
478 if (nf_conntrack_max &&
479 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
480 unsigned int hash = hash_conntrack(orig);
481 if (!early_drop(net, hash)) {
482 atomic_dec(&net->ct.count);
485 "nf_conntrack: table full, dropping"
487 return ERR_PTR(-ENOMEM);
491 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
493 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
494 atomic_dec(&net->ct.count);
495 return ERR_PTR(-ENOMEM);
498 atomic_set(&ct->ct_general.use, 1);
499 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
500 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
501 /* Don't set timer yet: wait for confirmation */
502 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
506 INIT_RCU_HEAD(&ct->rcu);
510 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
512 static void nf_conntrack_free_rcu(struct rcu_head *head)
514 struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
515 struct net *net = nf_ct_net(ct);
518 kmem_cache_free(nf_conntrack_cachep, ct);
519 atomic_dec(&net->ct.count);
522 void nf_conntrack_free(struct nf_conn *ct)
524 nf_ct_ext_destroy(ct);
525 call_rcu(&ct->rcu, nf_conntrack_free_rcu);
527 EXPORT_SYMBOL_GPL(nf_conntrack_free);
529 /* Allocate a new conntrack: we return -ENOMEM if classification
530 failed due to stress. Otherwise it really is unclassifiable. */
531 static struct nf_conntrack_tuple_hash *
532 init_conntrack(struct net *net,
533 const struct nf_conntrack_tuple *tuple,
534 struct nf_conntrack_l3proto *l3proto,
535 struct nf_conntrack_l4proto *l4proto,
537 unsigned int dataoff)
540 struct nf_conn_help *help;
541 struct nf_conntrack_tuple repl_tuple;
542 struct nf_conntrack_expect *exp;
544 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
545 pr_debug("Can't invert tuple.\n");
549 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
550 if (ct == NULL || IS_ERR(ct)) {
551 pr_debug("Can't allocate conntrack.\n");
552 return (struct nf_conntrack_tuple_hash *)ct;
555 if (!l4proto->new(ct, skb, dataoff)) {
556 nf_conntrack_free(ct);
557 pr_debug("init conntrack: can't track with proto module\n");
561 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
563 spin_lock_bh(&nf_conntrack_lock);
564 exp = nf_ct_find_expectation(net, tuple);
566 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
568 /* Welcome, Mr. Bond. We've been expecting you... */
569 __set_bit(IPS_EXPECTED_BIT, &ct->status);
570 ct->master = exp->master;
572 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
574 rcu_assign_pointer(help->helper, exp->helper);
577 #ifdef CONFIG_NF_CONNTRACK_MARK
578 ct->mark = exp->master->mark;
580 #ifdef CONFIG_NF_CONNTRACK_SECMARK
581 ct->secmark = exp->master->secmark;
583 nf_conntrack_get(&ct->master->ct_general);
584 NF_CT_STAT_INC(expect_new);
586 struct nf_conntrack_helper *helper;
588 helper = __nf_ct_helper_find(&repl_tuple);
590 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
592 rcu_assign_pointer(help->helper, helper);
597 /* Overload tuple linked list to put us in unconfirmed list. */
598 hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
599 &net->ct.unconfirmed);
601 spin_unlock_bh(&nf_conntrack_lock);
605 exp->expectfn(ct, exp);
606 nf_ct_expect_put(exp);
609 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
612 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
613 static inline struct nf_conn *
614 resolve_normal_ct(struct net *net,
616 unsigned int dataoff,
619 struct nf_conntrack_l3proto *l3proto,
620 struct nf_conntrack_l4proto *l4proto,
622 enum ip_conntrack_info *ctinfo)
624 struct nf_conntrack_tuple tuple;
625 struct nf_conntrack_tuple_hash *h;
628 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
629 dataoff, l3num, protonum, &tuple, l3proto,
631 pr_debug("resolve_normal_ct: Can't get tuple\n");
635 /* look for tuple match */
636 h = nf_conntrack_find_get(net, &tuple);
638 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
644 ct = nf_ct_tuplehash_to_ctrack(h);
646 /* It exists; we have (non-exclusive) reference. */
647 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
648 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
649 /* Please set reply bit if this packet OK */
652 /* Once we've had two way comms, always ESTABLISHED. */
653 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
654 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
655 *ctinfo = IP_CT_ESTABLISHED;
656 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
657 pr_debug("nf_conntrack_in: related packet for %p\n",
659 *ctinfo = IP_CT_RELATED;
661 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
666 skb->nfct = &ct->ct_general;
667 skb->nfctinfo = *ctinfo;
672 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
676 enum ip_conntrack_info ctinfo;
677 struct nf_conntrack_l3proto *l3proto;
678 struct nf_conntrack_l4proto *l4proto;
679 unsigned int dataoff;
684 /* Previously seen (loopback or untracked)? Ignore. */
686 NF_CT_STAT_INC_ATOMIC(ignore);
690 /* rcu_read_lock()ed by nf_hook_slow */
691 l3proto = __nf_ct_l3proto_find(pf);
692 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
693 &dataoff, &protonum);
695 pr_debug("not prepared to track yet or error occured\n");
696 NF_CT_STAT_INC_ATOMIC(error);
697 NF_CT_STAT_INC_ATOMIC(invalid);
701 l4proto = __nf_ct_l4proto_find(pf, protonum);
703 /* It may be an special packet, error, unclean...
704 * inverse of the return code tells to the netfilter
705 * core what to do with the packet. */
706 if (l4proto->error != NULL &&
707 (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
708 NF_CT_STAT_INC_ATOMIC(error);
709 NF_CT_STAT_INC_ATOMIC(invalid);
713 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
714 l3proto, l4proto, &set_reply, &ctinfo);
716 /* Not valid part of a connection */
717 NF_CT_STAT_INC_ATOMIC(invalid);
722 /* Too stressed to deal. */
723 NF_CT_STAT_INC_ATOMIC(drop);
727 NF_CT_ASSERT(skb->nfct);
729 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
731 /* Invalid: inverse of the return code tells
732 * the netfilter core what to do */
733 pr_debug("nf_conntrack_in: Can't track with proto module\n");
734 nf_conntrack_put(skb->nfct);
736 NF_CT_STAT_INC_ATOMIC(invalid);
740 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
741 nf_conntrack_event_cache(IPCT_STATUS, skb);
745 EXPORT_SYMBOL_GPL(nf_conntrack_in);
747 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
748 const struct nf_conntrack_tuple *orig)
753 ret = nf_ct_invert_tuple(inverse, orig,
754 __nf_ct_l3proto_find(orig->src.l3num),
755 __nf_ct_l4proto_find(orig->src.l3num,
756 orig->dst.protonum));
760 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
762 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
763 implicitly racy: see __nf_conntrack_confirm */
764 void nf_conntrack_alter_reply(struct nf_conn *ct,
765 const struct nf_conntrack_tuple *newreply)
767 struct nf_conn_help *help = nfct_help(ct);
768 struct nf_conntrack_helper *helper;
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 helper = __nf_ct_helper_find(newreply);
782 if (helper == NULL) {
784 rcu_assign_pointer(help->helper, NULL);
789 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
793 memset(&help->help, 0, sizeof(help->help));
796 rcu_assign_pointer(help->helper, helper);
800 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
802 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
803 void __nf_ct_refresh_acct(struct nf_conn *ct,
804 enum ip_conntrack_info ctinfo,
805 const struct sk_buff *skb,
806 unsigned long extra_jiffies,
811 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
814 spin_lock_bh(&nf_conntrack_lock);
816 /* Only update if this is not a fixed timeout */
817 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
820 /* If not in hash table, timer will not be active yet */
821 if (!nf_ct_is_confirmed(ct)) {
822 ct->timeout.expires = extra_jiffies;
823 event = IPCT_REFRESH;
825 unsigned long newtime = jiffies + extra_jiffies;
827 /* Only update the timeout if the new timeout is at least
828 HZ jiffies from the old timeout. Need del_timer for race
829 avoidance (may already be dying). */
830 if (newtime - ct->timeout.expires >= HZ
831 && del_timer(&ct->timeout)) {
832 ct->timeout.expires = newtime;
833 add_timer(&ct->timeout);
834 event = IPCT_REFRESH;
840 struct nf_conn_counter *acct;
842 acct = nf_conn_acct_find(ct);
844 acct[CTINFO2DIR(ctinfo)].packets++;
845 acct[CTINFO2DIR(ctinfo)].bytes +=
846 skb->len - skb_network_offset(skb);
850 spin_unlock_bh(&nf_conntrack_lock);
852 /* must be unlocked when calling event cache */
854 nf_conntrack_event_cache(event, skb);
856 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
858 bool __nf_ct_kill_acct(struct nf_conn *ct,
859 enum ip_conntrack_info ctinfo,
860 const struct sk_buff *skb,
864 struct nf_conn_counter *acct;
866 spin_lock_bh(&nf_conntrack_lock);
867 acct = nf_conn_acct_find(ct);
869 acct[CTINFO2DIR(ctinfo)].packets++;
870 acct[CTINFO2DIR(ctinfo)].bytes +=
871 skb->len - skb_network_offset(skb);
873 spin_unlock_bh(&nf_conntrack_lock);
876 if (del_timer(&ct->timeout)) {
877 ct->timeout.function((unsigned long)ct);
882 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
884 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
886 #include <linux/netfilter/nfnetlink.h>
887 #include <linux/netfilter/nfnetlink_conntrack.h>
888 #include <linux/mutex.h>
890 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
891 * in ip_conntrack_core, since we don't want the protocols to autoload
892 * or depend on ctnetlink */
893 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
894 const struct nf_conntrack_tuple *tuple)
896 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
897 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
903 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
905 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
906 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
907 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
909 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
911 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
912 struct nf_conntrack_tuple *t)
914 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
917 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
918 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
922 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
925 /* Used by ipt_REJECT and ip6t_REJECT. */
926 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
929 enum ip_conntrack_info ctinfo;
931 /* This ICMP is in reverse direction to the packet which caused it */
932 ct = nf_ct_get(skb, &ctinfo);
933 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
934 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
936 ctinfo = IP_CT_RELATED;
938 /* Attach to new skbuff, and increment count */
939 nskb->nfct = &ct->ct_general;
940 nskb->nfctinfo = ctinfo;
941 nf_conntrack_get(nskb->nfct);
944 /* Bring out ya dead! */
945 static struct nf_conn *
946 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
947 void *data, unsigned int *bucket)
949 struct nf_conntrack_tuple_hash *h;
951 struct hlist_node *n;
953 spin_lock_bh(&nf_conntrack_lock);
954 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
955 hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) {
956 ct = nf_ct_tuplehash_to_ctrack(h);
961 hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) {
962 ct = nf_ct_tuplehash_to_ctrack(h);
964 set_bit(IPS_DYING_BIT, &ct->status);
966 spin_unlock_bh(&nf_conntrack_lock);
969 atomic_inc(&ct->ct_general.use);
970 spin_unlock_bh(&nf_conntrack_lock);
974 void nf_ct_iterate_cleanup(struct net *net,
975 int (*iter)(struct nf_conn *i, void *data),
979 unsigned int bucket = 0;
981 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
982 /* Time to push up daises... */
983 if (del_timer(&ct->timeout))
984 death_by_timeout((unsigned long)ct);
985 /* ... else the timer will get him soon. */
990 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
992 static int kill_all(struct nf_conn *i, void *data)
997 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
1002 free_pages((unsigned long)hash,
1003 get_order(sizeof(struct hlist_head) * size));
1005 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1007 void nf_conntrack_flush(struct net *net)
1009 nf_ct_iterate_cleanup(net, kill_all, NULL);
1011 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1013 /* Mishearing the voices in his head, our hero wonders how he's
1014 supposed to kill the mall. */
1015 void nf_conntrack_cleanup(struct net *net)
1017 rcu_assign_pointer(ip_ct_attach, NULL);
1019 /* This makes sure all current packets have passed through
1020 netfilter framework. Roll on, two-stage module
1024 nf_ct_event_cache_flush();
1026 nf_conntrack_flush(net);
1027 if (atomic_read(&net->ct.count) != 0) {
1029 goto i_see_dead_people;
1031 /* wait until all references to nf_conntrack_untracked are dropped */
1032 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1035 rcu_assign_pointer(nf_ct_destroy, NULL);
1037 kmem_cache_destroy(nf_conntrack_cachep);
1038 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1039 nf_conntrack_htable_size);
1041 nf_conntrack_acct_fini();
1042 nf_conntrack_expect_fini(net);
1043 nf_conntrack_helper_fini();
1044 nf_conntrack_proto_fini();
1047 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1049 struct hlist_head *hash;
1050 unsigned int size, i;
1054 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1055 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1056 get_order(sizeof(struct hlist_head)
1060 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1061 hash = vmalloc(sizeof(struct hlist_head) * size);
1065 for (i = 0; i < size; i++)
1066 INIT_HLIST_HEAD(&hash[i]);
1070 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1072 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1074 int i, bucket, vmalloced, old_vmalloced;
1075 unsigned int hashsize, old_size;
1077 struct hlist_head *hash, *old_hash;
1078 struct nf_conntrack_tuple_hash *h;
1080 /* On boot, we can set this without any fancy locking. */
1081 if (!nf_conntrack_htable_size)
1082 return param_set_uint(val, kp);
1084 hashsize = simple_strtoul(val, NULL, 0);
1088 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1092 /* We have to rehahs for the new table anyway, so we also can
1093 * use a newrandom seed */
1094 get_random_bytes(&rnd, 4);
1096 /* Lookups in the old hash might happen in parallel, which means we
1097 * might get false negatives during connection lookup. New connections
1098 * created because of a false negative won't make it into the hash
1099 * though since that required taking the lock.
1101 spin_lock_bh(&nf_conntrack_lock);
1102 for (i = 0; i < nf_conntrack_htable_size; i++) {
1103 while (!hlist_empty(&init_net.ct.hash[i])) {
1104 h = hlist_entry(init_net.ct.hash[i].first,
1105 struct nf_conntrack_tuple_hash, hnode);
1106 hlist_del_rcu(&h->hnode);
1107 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1108 hlist_add_head(&h->hnode, &hash[bucket]);
1111 old_size = nf_conntrack_htable_size;
1112 old_vmalloced = init_net.ct.hash_vmalloc;
1113 old_hash = init_net.ct.hash;
1115 nf_conntrack_htable_size = hashsize;
1116 init_net.ct.hash_vmalloc = vmalloced;
1117 init_net.ct.hash = hash;
1118 nf_conntrack_hash_rnd = rnd;
1119 spin_unlock_bh(&nf_conntrack_lock);
1121 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1124 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1126 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1127 &nf_conntrack_htable_size, 0600);
1129 int nf_conntrack_init(struct net *net)
1134 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1135 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1136 if (!nf_conntrack_htable_size) {
1137 nf_conntrack_htable_size
1138 = (((num_physpages << PAGE_SHIFT) / 16384)
1139 / sizeof(struct hlist_head));
1140 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1141 nf_conntrack_htable_size = 16384;
1142 if (nf_conntrack_htable_size < 32)
1143 nf_conntrack_htable_size = 32;
1145 /* Use a max. factor of four by default to get the same max as
1146 * with the old struct list_heads. When a table size is given
1147 * we use the old value of 8 to avoid reducing the max.
1151 atomic_set(&net->ct.count, 0);
1152 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1153 &net->ct.hash_vmalloc);
1154 if (!net->ct.hash) {
1155 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1158 INIT_HLIST_HEAD(&net->ct.unconfirmed);
1160 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1162 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1163 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1166 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1167 sizeof(struct nf_conn),
1169 if (!nf_conntrack_cachep) {
1170 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1174 ret = nf_conntrack_proto_init();
1176 goto err_free_conntrack_slab;
1178 ret = nf_conntrack_expect_init(net);
1180 goto out_fini_proto;
1182 ret = nf_conntrack_helper_init();
1184 goto out_fini_expect;
1186 ret = nf_conntrack_acct_init();
1188 goto out_fini_helper;
1190 /* For use by REJECT target */
1191 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1192 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1194 /* Set up fake conntrack:
1195 - to never be deleted, not in any hashes */
1196 #ifdef CONFIG_NET_NS
1197 nf_conntrack_untracked.ct_net = &init_net;
1199 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1200 /* - and look it like as a confirmed connection */
1201 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1206 nf_conntrack_helper_fini();
1208 nf_conntrack_expect_fini(net);
1210 nf_conntrack_proto_fini();
1211 err_free_conntrack_slab:
1212 kmem_cache_destroy(nf_conntrack_cachep);
1214 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1215 nf_conntrack_htable_size);