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 hlist_head *nf_conntrack_hash __read_mostly;
54 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
56 struct nf_conn nf_conntrack_untracked __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
59 unsigned int nf_ct_log_invalid __read_mostly;
60 HLIST_HEAD(unconfirmed);
61 static int nf_conntrack_vmalloc __read_mostly;
62 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
64 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
65 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
67 static int nf_conntrack_hash_rnd_initted;
68 static unsigned int nf_conntrack_hash_rnd;
70 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
71 unsigned int size, unsigned int rnd)
76 /* The direction must be ignored, so we hash everything up to the
77 * destination ports (which is a multiple of 4) and treat the last
78 * three bytes manually.
80 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
81 h = jhash2((u32 *)tuple, n,
82 rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
83 tuple->dst.protonum));
85 return ((u64)h * size) >> 32;
88 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
90 return __hash_conntrack(tuple, nf_conntrack_htable_size,
91 nf_conntrack_hash_rnd);
95 nf_ct_get_tuple(const struct sk_buff *skb,
100 struct nf_conntrack_tuple *tuple,
101 const struct nf_conntrack_l3proto *l3proto,
102 const struct nf_conntrack_l4proto *l4proto)
104 memset(tuple, 0, sizeof(*tuple));
106 tuple->src.l3num = l3num;
107 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
110 tuple->dst.protonum = protonum;
111 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
113 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
115 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
117 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
118 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
120 struct nf_conntrack_l3proto *l3proto;
121 struct nf_conntrack_l4proto *l4proto;
122 unsigned int protoff;
128 l3proto = __nf_ct_l3proto_find(l3num);
129 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
130 if (ret != NF_ACCEPT) {
135 l4proto = __nf_ct_l4proto_find(l3num, protonum);
137 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
143 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
146 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
147 const struct nf_conntrack_tuple *orig,
148 const struct nf_conntrack_l3proto *l3proto,
149 const struct nf_conntrack_l4proto *l4proto)
151 memset(inverse, 0, sizeof(*inverse));
153 inverse->src.l3num = orig->src.l3num;
154 if (l3proto->invert_tuple(inverse, orig) == 0)
157 inverse->dst.dir = !orig->dst.dir;
159 inverse->dst.protonum = orig->dst.protonum;
160 return l4proto->invert_tuple(inverse, orig);
162 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
165 clean_from_lists(struct nf_conn *ct)
167 pr_debug("clean_from_lists(%p)\n", ct);
168 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
169 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
171 /* Destroy all pending expectations */
172 nf_ct_remove_expectations(ct);
176 destroy_conntrack(struct nf_conntrack *nfct)
178 struct nf_conn *ct = (struct nf_conn *)nfct;
179 struct nf_conntrack_l4proto *l4proto;
181 pr_debug("destroy_conntrack(%p)\n", ct);
182 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
183 NF_CT_ASSERT(!timer_pending(&ct->timeout));
185 nf_conntrack_event(IPCT_DESTROY, ct);
186 set_bit(IPS_DYING_BIT, &ct->status);
188 /* To make sure we don't get any weird locking issues here:
189 * destroy_conntrack() MUST NOT be called with a write lock
190 * to nf_conntrack_lock!!! -HW */
192 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
193 if (l4proto && l4proto->destroy)
194 l4proto->destroy(ct);
198 spin_lock_bh(&nf_conntrack_lock);
199 /* Expectations will have been removed in clean_from_lists,
200 * except TFTP can create an expectation on the first packet,
201 * before connection is in the list, so we need to clean here,
203 nf_ct_remove_expectations(ct);
205 /* We overload first tuple to link into unconfirmed list. */
206 if (!nf_ct_is_confirmed(ct)) {
207 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
208 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
211 NF_CT_STAT_INC(delete);
212 spin_unlock_bh(&nf_conntrack_lock);
215 nf_ct_put(ct->master);
217 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
218 nf_conntrack_free(ct);
221 static void death_by_timeout(unsigned long ul_conntrack)
223 struct nf_conn *ct = (void *)ul_conntrack;
224 struct nf_conn_help *help = nfct_help(ct);
225 struct nf_conntrack_helper *helper;
229 helper = rcu_dereference(help->helper);
230 if (helper && helper->destroy)
235 spin_lock_bh(&nf_conntrack_lock);
236 /* Inside lock so preempt is disabled on module removal path.
237 * Otherwise we can get spurious warnings. */
238 NF_CT_STAT_INC(delete_list);
239 clean_from_lists(ct);
240 spin_unlock_bh(&nf_conntrack_lock);
244 struct nf_conntrack_tuple_hash *
245 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple)
247 struct nf_conntrack_tuple_hash *h;
248 struct hlist_node *n;
249 unsigned int hash = hash_conntrack(tuple);
251 /* Disable BHs the entire time since we normally need to disable them
252 * at least once for the stats anyway.
255 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
256 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
257 NF_CT_STAT_INC(found);
261 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;
277 h = __nf_conntrack_find(tuple);
279 ct = nf_ct_tuplehash_to_ctrack(h);
280 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
287 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
289 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
291 unsigned int repl_hash)
293 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
294 &nf_conntrack_hash[hash]);
295 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
296 &nf_conntrack_hash[repl_hash]);
299 void nf_conntrack_hash_insert(struct nf_conn *ct)
301 unsigned int hash, repl_hash;
303 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
304 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
306 spin_lock_bh(&nf_conntrack_lock);
307 __nf_conntrack_hash_insert(ct, hash, repl_hash);
308 spin_unlock_bh(&nf_conntrack_lock);
310 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
312 /* Confirm a connection given skb; places it in hash table */
314 __nf_conntrack_confirm(struct sk_buff *skb)
316 unsigned int hash, repl_hash;
317 struct nf_conntrack_tuple_hash *h;
319 struct nf_conn_help *help;
320 struct hlist_node *n;
321 enum ip_conntrack_info ctinfo;
323 ct = nf_ct_get(skb, &ctinfo);
325 /* ipt_REJECT uses nf_conntrack_attach to attach related
326 ICMP/TCP RST packets in other direction. Actual packet
327 which created connection will be IP_CT_NEW or for an
328 expected connection, IP_CT_RELATED. */
329 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
332 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
333 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
335 /* We're not in hash table, and we refuse to set up related
336 connections for unconfirmed conns. But packet copies and
337 REJECT will give spurious warnings here. */
338 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
340 /* No external references means noone else could have
342 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
343 pr_debug("Confirming conntrack %p\n", ct);
345 spin_lock_bh(&nf_conntrack_lock);
347 /* See if there's one in the list already, including reverse:
348 NAT could have grabbed it without realizing, since we're
349 not in the hash. If there is, we lost race. */
350 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
351 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
354 hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
355 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
359 /* Remove from unconfirmed list */
360 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
362 __nf_conntrack_hash_insert(ct, hash, repl_hash);
363 /* Timer relative to confirmation time, not original
364 setting time, otherwise we'd get timer wrap in
365 weird delay cases. */
366 ct->timeout.expires += jiffies;
367 add_timer(&ct->timeout);
368 atomic_inc(&ct->ct_general.use);
369 set_bit(IPS_CONFIRMED_BIT, &ct->status);
370 NF_CT_STAT_INC(insert);
371 spin_unlock_bh(&nf_conntrack_lock);
372 help = nfct_help(ct);
373 if (help && help->helper)
374 nf_conntrack_event_cache(IPCT_HELPER, skb);
375 #ifdef CONFIG_NF_NAT_NEEDED
376 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
377 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
378 nf_conntrack_event_cache(IPCT_NATINFO, skb);
380 nf_conntrack_event_cache(master_ct(ct) ?
381 IPCT_RELATED : IPCT_NEW, skb);
385 NF_CT_STAT_INC(insert_failed);
386 spin_unlock_bh(&nf_conntrack_lock);
389 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
391 /* Returns true if a connection correspondings to the tuple (required
394 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
395 const struct nf_conn *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, &nf_conntrack_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(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, &nf_conntrack_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(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(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, &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 sk_buff *skb,
614 unsigned int dataoff,
617 struct nf_conntrack_l3proto *l3proto,
618 struct nf_conntrack_l4proto *l4proto,
620 enum ip_conntrack_info *ctinfo)
622 struct nf_conntrack_tuple tuple;
623 struct nf_conntrack_tuple_hash *h;
626 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
627 dataoff, l3num, protonum, &tuple, l3proto,
629 pr_debug("resolve_normal_ct: Can't get tuple\n");
633 /* look for tuple match */
634 h = nf_conntrack_find_get(&tuple);
636 h = init_conntrack(&init_net, &tuple, l3proto, l4proto, skb,
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(u_int8_t pf, unsigned int hooknum, struct sk_buff *skb)
674 enum ip_conntrack_info ctinfo;
675 struct nf_conntrack_l3proto *l3proto;
676 struct nf_conntrack_l4proto *l4proto;
677 unsigned int dataoff;
682 /* Previously seen (loopback or untracked)? Ignore. */
684 NF_CT_STAT_INC_ATOMIC(ignore);
688 /* rcu_read_lock()ed by nf_hook_slow */
689 l3proto = __nf_ct_l3proto_find(pf);
690 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
691 &dataoff, &protonum);
693 pr_debug("not prepared to track yet or error occured\n");
694 NF_CT_STAT_INC_ATOMIC(error);
695 NF_CT_STAT_INC_ATOMIC(invalid);
699 l4proto = __nf_ct_l4proto_find(pf, protonum);
701 /* It may be an special packet, error, unclean...
702 * inverse of the return code tells to the netfilter
703 * core what to do with the packet. */
704 if (l4proto->error != NULL &&
705 (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
706 NF_CT_STAT_INC_ATOMIC(error);
707 NF_CT_STAT_INC_ATOMIC(invalid);
711 ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
712 &set_reply, &ctinfo);
714 /* Not valid part of a connection */
715 NF_CT_STAT_INC_ATOMIC(invalid);
720 /* Too stressed to deal. */
721 NF_CT_STAT_INC_ATOMIC(drop);
725 NF_CT_ASSERT(skb->nfct);
727 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
729 /* Invalid: inverse of the return code tells
730 * the netfilter core what to do */
731 pr_debug("nf_conntrack_in: Can't track with proto module\n");
732 nf_conntrack_put(skb->nfct);
734 NF_CT_STAT_INC_ATOMIC(invalid);
738 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
739 nf_conntrack_event_cache(IPCT_STATUS, skb);
743 EXPORT_SYMBOL_GPL(nf_conntrack_in);
745 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
746 const struct nf_conntrack_tuple *orig)
751 ret = nf_ct_invert_tuple(inverse, orig,
752 __nf_ct_l3proto_find(orig->src.l3num),
753 __nf_ct_l4proto_find(orig->src.l3num,
754 orig->dst.protonum));
758 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
760 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
761 implicitly racy: see __nf_conntrack_confirm */
762 void nf_conntrack_alter_reply(struct nf_conn *ct,
763 const struct nf_conntrack_tuple *newreply)
765 struct nf_conn_help *help = nfct_help(ct);
766 struct nf_conntrack_helper *helper;
768 /* Should be unconfirmed, so not in hash table yet */
769 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
771 pr_debug("Altering reply tuple of %p to ", ct);
772 nf_ct_dump_tuple(newreply);
774 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
775 if (ct->master || (help && !hlist_empty(&help->expectations)))
779 helper = __nf_ct_helper_find(newreply);
780 if (helper == NULL) {
782 rcu_assign_pointer(help->helper, NULL);
787 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
791 memset(&help->help, 0, sizeof(help->help));
794 rcu_assign_pointer(help->helper, helper);
798 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
800 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
801 void __nf_ct_refresh_acct(struct nf_conn *ct,
802 enum ip_conntrack_info ctinfo,
803 const struct sk_buff *skb,
804 unsigned long extra_jiffies,
809 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
812 spin_lock_bh(&nf_conntrack_lock);
814 /* Only update if this is not a fixed timeout */
815 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
818 /* If not in hash table, timer will not be active yet */
819 if (!nf_ct_is_confirmed(ct)) {
820 ct->timeout.expires = extra_jiffies;
821 event = IPCT_REFRESH;
823 unsigned long newtime = jiffies + extra_jiffies;
825 /* Only update the timeout if the new timeout is at least
826 HZ jiffies from the old timeout. Need del_timer for race
827 avoidance (may already be dying). */
828 if (newtime - ct->timeout.expires >= HZ
829 && del_timer(&ct->timeout)) {
830 ct->timeout.expires = newtime;
831 add_timer(&ct->timeout);
832 event = IPCT_REFRESH;
838 struct nf_conn_counter *acct;
840 acct = nf_conn_acct_find(ct);
842 acct[CTINFO2DIR(ctinfo)].packets++;
843 acct[CTINFO2DIR(ctinfo)].bytes +=
844 skb->len - skb_network_offset(skb);
848 spin_unlock_bh(&nf_conntrack_lock);
850 /* must be unlocked when calling event cache */
852 nf_conntrack_event_cache(event, skb);
854 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
856 bool __nf_ct_kill_acct(struct nf_conn *ct,
857 enum ip_conntrack_info ctinfo,
858 const struct sk_buff *skb,
862 struct nf_conn_counter *acct;
864 spin_lock_bh(&nf_conntrack_lock);
865 acct = nf_conn_acct_find(ct);
867 acct[CTINFO2DIR(ctinfo)].packets++;
868 acct[CTINFO2DIR(ctinfo)].bytes +=
869 skb->len - skb_network_offset(skb);
871 spin_unlock_bh(&nf_conntrack_lock);
874 if (del_timer(&ct->timeout)) {
875 ct->timeout.function((unsigned long)ct);
880 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
882 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
884 #include <linux/netfilter/nfnetlink.h>
885 #include <linux/netfilter/nfnetlink_conntrack.h>
886 #include <linux/mutex.h>
888 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
889 * in ip_conntrack_core, since we don't want the protocols to autoload
890 * or depend on ctnetlink */
891 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
892 const struct nf_conntrack_tuple *tuple)
894 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
895 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
901 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
903 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
904 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
905 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
907 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
909 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
910 struct nf_conntrack_tuple *t)
912 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
915 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
916 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
920 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
923 /* Used by ipt_REJECT and ip6t_REJECT. */
924 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
927 enum ip_conntrack_info ctinfo;
929 /* This ICMP is in reverse direction to the packet which caused it */
930 ct = nf_ct_get(skb, &ctinfo);
931 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
932 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
934 ctinfo = IP_CT_RELATED;
936 /* Attach to new skbuff, and increment count */
937 nskb->nfct = &ct->ct_general;
938 nskb->nfctinfo = ctinfo;
939 nf_conntrack_get(nskb->nfct);
942 /* Bring out ya dead! */
943 static struct nf_conn *
944 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
945 void *data, unsigned int *bucket)
947 struct nf_conntrack_tuple_hash *h;
949 struct hlist_node *n;
951 spin_lock_bh(&nf_conntrack_lock);
952 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
953 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
954 ct = nf_ct_tuplehash_to_ctrack(h);
959 hlist_for_each_entry(h, n, &unconfirmed, hnode) {
960 ct = nf_ct_tuplehash_to_ctrack(h);
962 set_bit(IPS_DYING_BIT, &ct->status);
964 spin_unlock_bh(&nf_conntrack_lock);
967 atomic_inc(&ct->ct_general.use);
968 spin_unlock_bh(&nf_conntrack_lock);
973 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
976 unsigned int bucket = 0;
978 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
979 /* Time to push up daises... */
980 if (del_timer(&ct->timeout))
981 death_by_timeout((unsigned long)ct);
982 /* ... else the timer will get him soon. */
987 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
989 static int kill_all(struct nf_conn *i, void *data)
994 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
999 free_pages((unsigned long)hash,
1000 get_order(sizeof(struct hlist_head) * size));
1002 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1004 void nf_conntrack_flush(void)
1006 nf_ct_iterate_cleanup(kill_all, NULL);
1008 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1010 /* Mishearing the voices in his head, our hero wonders how he's
1011 supposed to kill the mall. */
1012 void nf_conntrack_cleanup(struct net *net)
1014 rcu_assign_pointer(ip_ct_attach, NULL);
1016 /* This makes sure all current packets have passed through
1017 netfilter framework. Roll on, two-stage module
1021 nf_ct_event_cache_flush();
1023 nf_conntrack_flush();
1024 if (atomic_read(&net->ct.count) != 0) {
1026 goto i_see_dead_people;
1028 /* wait until all references to nf_conntrack_untracked are dropped */
1029 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1032 rcu_assign_pointer(nf_ct_destroy, NULL);
1034 kmem_cache_destroy(nf_conntrack_cachep);
1035 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1036 nf_conntrack_htable_size);
1038 nf_conntrack_acct_fini();
1039 nf_conntrack_expect_fini();
1040 nf_conntrack_helper_fini();
1041 nf_conntrack_proto_fini();
1044 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1046 struct hlist_head *hash;
1047 unsigned int size, i;
1051 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1052 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1053 get_order(sizeof(struct hlist_head)
1057 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1058 hash = vmalloc(sizeof(struct hlist_head) * size);
1062 for (i = 0; i < size; i++)
1063 INIT_HLIST_HEAD(&hash[i]);
1067 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1069 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1071 int i, bucket, vmalloced, old_vmalloced;
1072 unsigned int hashsize, old_size;
1074 struct hlist_head *hash, *old_hash;
1075 struct nf_conntrack_tuple_hash *h;
1077 /* On boot, we can set this without any fancy locking. */
1078 if (!nf_conntrack_htable_size)
1079 return param_set_uint(val, kp);
1081 hashsize = simple_strtoul(val, NULL, 0);
1085 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1089 /* We have to rehahs for the new table anyway, so we also can
1090 * use a newrandom seed */
1091 get_random_bytes(&rnd, 4);
1093 /* Lookups in the old hash might happen in parallel, which means we
1094 * might get false negatives during connection lookup. New connections
1095 * created because of a false negative won't make it into the hash
1096 * though since that required taking the lock.
1098 spin_lock_bh(&nf_conntrack_lock);
1099 for (i = 0; i < nf_conntrack_htable_size; i++) {
1100 while (!hlist_empty(&nf_conntrack_hash[i])) {
1101 h = hlist_entry(nf_conntrack_hash[i].first,
1102 struct nf_conntrack_tuple_hash, hnode);
1103 hlist_del_rcu(&h->hnode);
1104 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1105 hlist_add_head(&h->hnode, &hash[bucket]);
1108 old_size = nf_conntrack_htable_size;
1109 old_vmalloced = nf_conntrack_vmalloc;
1110 old_hash = nf_conntrack_hash;
1112 nf_conntrack_htable_size = hashsize;
1113 nf_conntrack_vmalloc = vmalloced;
1114 nf_conntrack_hash = hash;
1115 nf_conntrack_hash_rnd = rnd;
1116 spin_unlock_bh(&nf_conntrack_lock);
1118 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1121 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1123 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1124 &nf_conntrack_htable_size, 0600);
1126 int nf_conntrack_init(struct net *net)
1131 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1132 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1133 if (!nf_conntrack_htable_size) {
1134 nf_conntrack_htable_size
1135 = (((num_physpages << PAGE_SHIFT) / 16384)
1136 / sizeof(struct hlist_head));
1137 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1138 nf_conntrack_htable_size = 16384;
1139 if (nf_conntrack_htable_size < 32)
1140 nf_conntrack_htable_size = 32;
1142 /* Use a max. factor of four by default to get the same max as
1143 * with the old struct list_heads. When a table size is given
1144 * we use the old value of 8 to avoid reducing the max.
1148 atomic_set(&net->ct.count, 0);
1149 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1150 &nf_conntrack_vmalloc);
1151 if (!nf_conntrack_hash) {
1152 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1156 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1158 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1159 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1162 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1163 sizeof(struct nf_conn),
1165 if (!nf_conntrack_cachep) {
1166 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1170 ret = nf_conntrack_proto_init();
1172 goto err_free_conntrack_slab;
1174 ret = nf_conntrack_expect_init();
1176 goto out_fini_proto;
1178 ret = nf_conntrack_helper_init();
1180 goto out_fini_expect;
1182 ret = nf_conntrack_acct_init();
1184 goto out_fini_helper;
1186 /* For use by REJECT target */
1187 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1188 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1190 /* Set up fake conntrack:
1191 - to never be deleted, not in any hashes */
1192 #ifdef CONFIG_NET_NS
1193 nf_conntrack_untracked.ct_net = &init_net;
1195 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1196 /* - and look it like as a confirmed connection */
1197 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1202 nf_conntrack_helper_fini();
1204 nf_conntrack_expect_fini();
1206 nf_conntrack_proto_fini();
1207 err_free_conntrack_slab:
1208 kmem_cache_destroy(nf_conntrack_cachep);
1210 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1211 nf_conntrack_htable_size);