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>
32 #include <linux/rculist_nulls.h>
34 #include <net/netfilter/nf_conntrack.h>
35 #include <net/netfilter/nf_conntrack_l3proto.h>
36 #include <net/netfilter/nf_conntrack_l4proto.h>
37 #include <net/netfilter/nf_conntrack_expect.h>
38 #include <net/netfilter/nf_conntrack_helper.h>
39 #include <net/netfilter/nf_conntrack_core.h>
40 #include <net/netfilter/nf_conntrack_extend.h>
41 #include <net/netfilter/nf_conntrack_acct.h>
42 #include <net/netfilter/nf_nat.h>
43 #include <net/netfilter/nf_nat_core.h>
45 #define NF_CONNTRACK_VERSION "0.5.0"
47 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
48 enum nf_nat_manip_type manip,
49 struct nlattr *attr) __read_mostly;
50 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
52 DEFINE_SPINLOCK(nf_conntrack_lock);
53 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
55 unsigned int nf_conntrack_htable_size __read_mostly;
56 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
58 unsigned int nf_conntrack_max __read_mostly;
59 EXPORT_SYMBOL_GPL(nf_conntrack_max);
61 struct nf_conn nf_conntrack_untracked __read_mostly;
62 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
64 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
66 static int nf_conntrack_hash_rnd_initted;
67 static unsigned int nf_conntrack_hash_rnd;
69 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
70 unsigned int size, unsigned int rnd)
75 /* The direction must be ignored, so we hash everything up to the
76 * destination ports (which is a multiple of 4) and treat the last
77 * three bytes manually.
79 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
80 h = jhash2((u32 *)tuple, n,
81 rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
82 tuple->dst.protonum));
84 return ((u64)h * 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 memset(tuple, 0, sizeof(*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 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
117 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
119 struct nf_conntrack_l3proto *l3proto;
120 struct nf_conntrack_l4proto *l4proto;
121 unsigned int protoff;
127 l3proto = __nf_ct_l3proto_find(l3num);
128 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
129 if (ret != NF_ACCEPT) {
134 l4proto = __nf_ct_l4proto_find(l3num, protonum);
136 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
142 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
145 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
146 const struct nf_conntrack_tuple *orig,
147 const struct nf_conntrack_l3proto *l3proto,
148 const struct nf_conntrack_l4proto *l4proto)
150 memset(inverse, 0, sizeof(*inverse));
152 inverse->src.l3num = orig->src.l3num;
153 if (l3proto->invert_tuple(inverse, orig) == 0)
156 inverse->dst.dir = !orig->dst.dir;
158 inverse->dst.protonum = orig->dst.protonum;
159 return l4proto->invert_tuple(inverse, orig);
161 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
164 clean_from_lists(struct nf_conn *ct)
166 pr_debug("clean_from_lists(%p)\n", ct);
167 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
168 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
170 /* Destroy all pending expectations */
171 nf_ct_remove_expectations(ct);
175 destroy_conntrack(struct nf_conntrack *nfct)
177 struct nf_conn *ct = (struct nf_conn *)nfct;
178 struct net *net = nf_ct_net(ct);
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 if (!test_bit(IPS_DYING_BIT, &ct->status))
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(nf_ct_l3num(ct), nf_ct_protonum(ct));
194 if (l4proto && l4proto->destroy)
195 l4proto->destroy(ct);
199 spin_lock_bh(&nf_conntrack_lock);
200 /* Expectations will have been removed in clean_from_lists,
201 * except TFTP can create an expectation on the first packet,
202 * before connection is in the list, so we need to clean here,
204 nf_ct_remove_expectations(ct);
206 /* We overload first tuple to link into unconfirmed list. */
207 if (!nf_ct_is_confirmed(ct)) {
208 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
209 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
212 NF_CT_STAT_INC(net, delete);
213 spin_unlock_bh(&nf_conntrack_lock);
216 nf_ct_put(ct->master);
218 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
219 nf_conntrack_free(ct);
222 static void death_by_timeout(unsigned long ul_conntrack)
224 struct nf_conn *ct = (void *)ul_conntrack;
225 struct net *net = nf_ct_net(ct);
226 struct nf_conn_help *help = nfct_help(ct);
227 struct nf_conntrack_helper *helper;
231 helper = rcu_dereference(help->helper);
232 if (helper && helper->destroy)
237 spin_lock_bh(&nf_conntrack_lock);
238 /* Inside lock so preempt is disabled on module removal path.
239 * Otherwise we can get spurious warnings. */
240 NF_CT_STAT_INC(net, delete_list);
241 clean_from_lists(ct);
242 spin_unlock_bh(&nf_conntrack_lock);
248 * - Caller must take a reference on returned object
249 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
251 * - Caller must lock nf_conntrack_lock before calling this function
253 struct nf_conntrack_tuple_hash *
254 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
256 struct nf_conntrack_tuple_hash *h;
257 struct hlist_nulls_node *n;
258 unsigned int hash = hash_conntrack(tuple);
260 /* Disable BHs the entire time since we normally need to disable them
261 * at least once for the stats anyway.
265 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
266 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
267 NF_CT_STAT_INC(net, found);
271 NF_CT_STAT_INC(net, searched);
274 * if the nulls value we got at the end of this lookup is
275 * not the expected one, we must restart lookup.
276 * We probably met an item that was moved to another chain.
278 if (get_nulls_value(n) != hash)
284 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
286 /* Find a connection corresponding to a tuple. */
287 struct nf_conntrack_tuple_hash *
288 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
290 struct nf_conntrack_tuple_hash *h;
295 h = __nf_conntrack_find(net, tuple);
297 ct = nf_ct_tuplehash_to_ctrack(h);
298 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
301 if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
311 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
313 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
315 unsigned int repl_hash)
317 struct net *net = nf_ct_net(ct);
319 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
320 &net->ct.hash[hash]);
321 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
322 &net->ct.hash[repl_hash]);
325 void nf_conntrack_hash_insert(struct nf_conn *ct)
327 unsigned int hash, repl_hash;
329 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
330 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
332 __nf_conntrack_hash_insert(ct, hash, repl_hash);
334 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
336 /* Confirm a connection given skb; places it in hash table */
338 __nf_conntrack_confirm(struct sk_buff *skb)
340 unsigned int hash, repl_hash;
341 struct nf_conntrack_tuple_hash *h;
343 struct nf_conn_help *help;
344 struct hlist_nulls_node *n;
345 enum ip_conntrack_info ctinfo;
348 ct = nf_ct_get(skb, &ctinfo);
351 /* ipt_REJECT uses nf_conntrack_attach to attach related
352 ICMP/TCP RST packets in other direction. Actual packet
353 which created connection will be IP_CT_NEW or for an
354 expected connection, IP_CT_RELATED. */
355 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
358 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
359 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
361 /* We're not in hash table, and we refuse to set up related
362 connections for unconfirmed conns. But packet copies and
363 REJECT will give spurious warnings here. */
364 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
366 /* No external references means noone else could have
368 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
369 pr_debug("Confirming conntrack %p\n", ct);
371 spin_lock_bh(&nf_conntrack_lock);
373 /* See if there's one in the list already, including reverse:
374 NAT could have grabbed it without realizing, since we're
375 not in the hash. If there is, we lost race. */
376 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
377 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
380 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
381 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
385 /* Remove from unconfirmed list */
386 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
388 __nf_conntrack_hash_insert(ct, hash, repl_hash);
389 /* Timer relative to confirmation time, not original
390 setting time, otherwise we'd get timer wrap in
391 weird delay cases. */
392 ct->timeout.expires += jiffies;
393 add_timer(&ct->timeout);
394 atomic_inc(&ct->ct_general.use);
395 set_bit(IPS_CONFIRMED_BIT, &ct->status);
396 NF_CT_STAT_INC(net, insert);
397 spin_unlock_bh(&nf_conntrack_lock);
398 help = nfct_help(ct);
399 if (help && help->helper)
400 nf_conntrack_event_cache(IPCT_HELPER, ct);
401 #ifdef CONFIG_NF_NAT_NEEDED
402 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
403 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
404 nf_conntrack_event_cache(IPCT_NATINFO, ct);
406 nf_conntrack_event_cache(master_ct(ct) ?
407 IPCT_RELATED : IPCT_NEW, ct);
411 NF_CT_STAT_INC(net, insert_failed);
412 spin_unlock_bh(&nf_conntrack_lock);
415 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
417 /* Returns true if a connection correspondings to the tuple (required
420 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
421 const struct nf_conn *ignored_conntrack)
423 struct net *net = nf_ct_net(ignored_conntrack);
424 struct nf_conntrack_tuple_hash *h;
425 struct hlist_nulls_node *n;
426 unsigned int hash = hash_conntrack(tuple);
428 /* Disable BHs the entire time since we need to disable them at
429 * least once for the stats anyway.
432 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
433 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
434 nf_ct_tuple_equal(tuple, &h->tuple)) {
435 NF_CT_STAT_INC(net, found);
436 rcu_read_unlock_bh();
439 NF_CT_STAT_INC(net, searched);
441 rcu_read_unlock_bh();
445 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
447 #define NF_CT_EVICTION_RANGE 8
449 /* There's a small race here where we may free a just-assured
450 connection. Too bad: we're in trouble anyway. */
451 static noinline int early_drop(struct net *net, unsigned int hash)
453 /* Use oldest entry, which is roughly LRU */
454 struct nf_conntrack_tuple_hash *h;
455 struct nf_conn *ct = NULL, *tmp;
456 struct hlist_nulls_node *n;
457 unsigned int i, cnt = 0;
461 for (i = 0; i < nf_conntrack_htable_size; i++) {
462 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
464 tmp = nf_ct_tuplehash_to_ctrack(h);
465 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
470 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
472 if (ct || cnt >= NF_CT_EVICTION_RANGE)
474 hash = (hash + 1) % nf_conntrack_htable_size;
481 if (del_timer(&ct->timeout)) {
482 death_by_timeout((unsigned long)ct);
484 NF_CT_STAT_INC_ATOMIC(net, early_drop);
490 struct nf_conn *nf_conntrack_alloc(struct net *net,
491 const struct nf_conntrack_tuple *orig,
492 const struct nf_conntrack_tuple *repl,
497 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
498 get_random_bytes(&nf_conntrack_hash_rnd,
499 sizeof(nf_conntrack_hash_rnd));
500 nf_conntrack_hash_rnd_initted = 1;
503 /* We don't want any race condition at early drop stage */
504 atomic_inc(&net->ct.count);
506 if (nf_conntrack_max &&
507 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
508 unsigned int hash = hash_conntrack(orig);
509 if (!early_drop(net, hash)) {
510 atomic_dec(&net->ct.count);
513 "nf_conntrack: table full, dropping"
515 return ERR_PTR(-ENOMEM);
519 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
521 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
522 atomic_dec(&net->ct.count);
523 return ERR_PTR(-ENOMEM);
526 atomic_set(&ct->ct_general.use, 1);
527 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
528 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
529 /* Don't set timer yet: wait for confirmation */
530 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
537 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
539 void nf_conntrack_free(struct nf_conn *ct)
541 struct net *net = nf_ct_net(ct);
543 nf_ct_ext_destroy(ct);
544 atomic_dec(&net->ct.count);
546 kmem_cache_free(nf_conntrack_cachep, ct);
548 EXPORT_SYMBOL_GPL(nf_conntrack_free);
550 /* Allocate a new conntrack: we return -ENOMEM if classification
551 failed due to stress. Otherwise it really is unclassifiable. */
552 static struct nf_conntrack_tuple_hash *
553 init_conntrack(struct net *net,
554 const struct nf_conntrack_tuple *tuple,
555 struct nf_conntrack_l3proto *l3proto,
556 struct nf_conntrack_l4proto *l4proto,
558 unsigned int dataoff)
561 struct nf_conn_help *help;
562 struct nf_conntrack_tuple repl_tuple;
563 struct nf_conntrack_expect *exp;
565 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
566 pr_debug("Can't invert tuple.\n");
570 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
572 pr_debug("Can't allocate conntrack.\n");
573 return (struct nf_conntrack_tuple_hash *)ct;
576 if (!l4proto->new(ct, skb, dataoff)) {
577 nf_conntrack_free(ct);
578 pr_debug("init conntrack: can't track with proto module\n");
582 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
584 spin_lock_bh(&nf_conntrack_lock);
585 exp = nf_ct_find_expectation(net, tuple);
587 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
589 /* Welcome, Mr. Bond. We've been expecting you... */
590 __set_bit(IPS_EXPECTED_BIT, &ct->status);
591 ct->master = exp->master;
593 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
595 rcu_assign_pointer(help->helper, exp->helper);
598 #ifdef CONFIG_NF_CONNTRACK_MARK
599 ct->mark = exp->master->mark;
601 #ifdef CONFIG_NF_CONNTRACK_SECMARK
602 ct->secmark = exp->master->secmark;
604 nf_conntrack_get(&ct->master->ct_general);
605 NF_CT_STAT_INC(net, expect_new);
607 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
608 NF_CT_STAT_INC(net, new);
611 /* Overload tuple linked list to put us in unconfirmed list. */
612 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
613 &net->ct.unconfirmed);
615 spin_unlock_bh(&nf_conntrack_lock);
619 exp->expectfn(ct, exp);
620 nf_ct_expect_put(exp);
623 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
626 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
627 static inline struct nf_conn *
628 resolve_normal_ct(struct net *net,
630 unsigned int dataoff,
633 struct nf_conntrack_l3proto *l3proto,
634 struct nf_conntrack_l4proto *l4proto,
636 enum ip_conntrack_info *ctinfo)
638 struct nf_conntrack_tuple tuple;
639 struct nf_conntrack_tuple_hash *h;
642 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
643 dataoff, l3num, protonum, &tuple, l3proto,
645 pr_debug("resolve_normal_ct: Can't get tuple\n");
649 /* look for tuple match */
650 h = nf_conntrack_find_get(net, &tuple);
652 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
658 ct = nf_ct_tuplehash_to_ctrack(h);
660 /* It exists; we have (non-exclusive) reference. */
661 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
662 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
663 /* Please set reply bit if this packet OK */
666 /* Once we've had two way comms, always ESTABLISHED. */
667 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
668 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
669 *ctinfo = IP_CT_ESTABLISHED;
670 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
671 pr_debug("nf_conntrack_in: related packet for %p\n",
673 *ctinfo = IP_CT_RELATED;
675 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
680 skb->nfct = &ct->ct_general;
681 skb->nfctinfo = *ctinfo;
686 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
690 enum ip_conntrack_info ctinfo;
691 struct nf_conntrack_l3proto *l3proto;
692 struct nf_conntrack_l4proto *l4proto;
693 unsigned int dataoff;
698 /* Previously seen (loopback or untracked)? Ignore. */
700 NF_CT_STAT_INC_ATOMIC(net, ignore);
704 /* rcu_read_lock()ed by nf_hook_slow */
705 l3proto = __nf_ct_l3proto_find(pf);
706 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
707 &dataoff, &protonum);
709 pr_debug("not prepared to track yet or error occured\n");
710 NF_CT_STAT_INC_ATOMIC(net, error);
711 NF_CT_STAT_INC_ATOMIC(net, invalid);
715 l4proto = __nf_ct_l4proto_find(pf, protonum);
717 /* It may be an special packet, error, unclean...
718 * inverse of the return code tells to the netfilter
719 * core what to do with the packet. */
720 if (l4proto->error != NULL) {
721 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
723 NF_CT_STAT_INC_ATOMIC(net, error);
724 NF_CT_STAT_INC_ATOMIC(net, invalid);
729 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
730 l3proto, l4proto, &set_reply, &ctinfo);
732 /* Not valid part of a connection */
733 NF_CT_STAT_INC_ATOMIC(net, invalid);
738 /* Too stressed to deal. */
739 NF_CT_STAT_INC_ATOMIC(net, drop);
743 NF_CT_ASSERT(skb->nfct);
745 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
747 /* Invalid: inverse of the return code tells
748 * the netfilter core what to do */
749 pr_debug("nf_conntrack_in: Can't track with proto module\n");
750 nf_conntrack_put(skb->nfct);
752 NF_CT_STAT_INC_ATOMIC(net, invalid);
754 NF_CT_STAT_INC_ATOMIC(net, drop);
758 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
759 nf_conntrack_event_cache(IPCT_STATUS, ct);
763 EXPORT_SYMBOL_GPL(nf_conntrack_in);
765 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
766 const struct nf_conntrack_tuple *orig)
771 ret = nf_ct_invert_tuple(inverse, orig,
772 __nf_ct_l3proto_find(orig->src.l3num),
773 __nf_ct_l4proto_find(orig->src.l3num,
774 orig->dst.protonum));
778 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
780 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
781 implicitly racy: see __nf_conntrack_confirm */
782 void nf_conntrack_alter_reply(struct nf_conn *ct,
783 const struct nf_conntrack_tuple *newreply)
785 struct nf_conn_help *help = nfct_help(ct);
787 /* Should be unconfirmed, so not in hash table yet */
788 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
790 pr_debug("Altering reply tuple of %p to ", ct);
791 nf_ct_dump_tuple(newreply);
793 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
794 if (ct->master || (help && !hlist_empty(&help->expectations)))
798 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
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);
925 int nf_ct_port_nlattr_tuple_size(void)
927 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
929 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
932 /* Used by ipt_REJECT and ip6t_REJECT. */
933 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
936 enum ip_conntrack_info ctinfo;
938 /* This ICMP is in reverse direction to the packet which caused it */
939 ct = nf_ct_get(skb, &ctinfo);
940 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
941 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
943 ctinfo = IP_CT_RELATED;
945 /* Attach to new skbuff, and increment count */
946 nskb->nfct = &ct->ct_general;
947 nskb->nfctinfo = ctinfo;
948 nf_conntrack_get(nskb->nfct);
951 /* Bring out ya dead! */
952 static struct nf_conn *
953 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
954 void *data, unsigned int *bucket)
956 struct nf_conntrack_tuple_hash *h;
958 struct hlist_nulls_node *n;
960 spin_lock_bh(&nf_conntrack_lock);
961 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
962 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
963 ct = nf_ct_tuplehash_to_ctrack(h);
968 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
969 ct = nf_ct_tuplehash_to_ctrack(h);
971 set_bit(IPS_DYING_BIT, &ct->status);
973 spin_unlock_bh(&nf_conntrack_lock);
976 atomic_inc(&ct->ct_general.use);
977 spin_unlock_bh(&nf_conntrack_lock);
981 void nf_ct_iterate_cleanup(struct net *net,
982 int (*iter)(struct nf_conn *i, void *data),
986 unsigned int bucket = 0;
988 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
989 /* Time to push up daises... */
990 if (del_timer(&ct->timeout))
991 death_by_timeout((unsigned long)ct);
992 /* ... else the timer will get him soon. */
997 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
999 struct __nf_ct_flush_report {
1004 static int kill_all(struct nf_conn *i, void *data)
1006 struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1008 /* get_next_corpse sets the dying bit for us */
1009 nf_conntrack_event_report(IPCT_DESTROY,
1016 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1021 free_pages((unsigned long)hash,
1022 get_order(sizeof(struct hlist_head) * size));
1024 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1026 void nf_conntrack_flush(struct net *net, u32 pid, int report)
1028 struct __nf_ct_flush_report fr = {
1032 nf_ct_iterate_cleanup(net, kill_all, &fr);
1034 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1036 static void nf_conntrack_cleanup_init_net(void)
1038 nf_conntrack_helper_fini();
1039 nf_conntrack_proto_fini();
1040 kmem_cache_destroy(nf_conntrack_cachep);
1043 static void nf_conntrack_cleanup_net(struct net *net)
1045 nf_ct_event_cache_flush(net);
1046 nf_conntrack_ecache_fini(net);
1048 nf_conntrack_flush(net, 0, 0);
1049 if (atomic_read(&net->ct.count) != 0) {
1051 goto i_see_dead_people;
1053 /* wait until all references to nf_conntrack_untracked are dropped */
1054 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1057 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1058 nf_conntrack_htable_size);
1059 nf_conntrack_acct_fini(net);
1060 nf_conntrack_expect_fini(net);
1061 free_percpu(net->ct.stat);
1064 /* Mishearing the voices in his head, our hero wonders how he's
1065 supposed to kill the mall. */
1066 void nf_conntrack_cleanup(struct net *net)
1068 if (net_eq(net, &init_net))
1069 rcu_assign_pointer(ip_ct_attach, NULL);
1071 /* This makes sure all current packets have passed through
1072 netfilter framework. Roll on, two-stage module
1076 nf_conntrack_cleanup_net(net);
1078 if (net_eq(net, &init_net)) {
1079 rcu_assign_pointer(nf_ct_destroy, NULL);
1080 nf_conntrack_cleanup_init_net();
1084 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1086 struct hlist_nulls_head *hash;
1087 unsigned int nr_slots, i;
1092 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1093 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1094 sz = nr_slots * sizeof(struct hlist_nulls_head);
1095 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1099 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1100 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1104 for (i = 0; i < nr_slots; i++)
1105 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1109 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1111 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1113 int i, bucket, vmalloced, old_vmalloced;
1114 unsigned int hashsize, old_size;
1116 struct hlist_nulls_head *hash, *old_hash;
1117 struct nf_conntrack_tuple_hash *h;
1119 /* On boot, we can set this without any fancy locking. */
1120 if (!nf_conntrack_htable_size)
1121 return param_set_uint(val, kp);
1123 hashsize = simple_strtoul(val, NULL, 0);
1127 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1131 /* We have to rehahs for the new table anyway, so we also can
1132 * use a newrandom seed */
1133 get_random_bytes(&rnd, sizeof(rnd));
1135 /* Lookups in the old hash might happen in parallel, which means we
1136 * might get false negatives during connection lookup. New connections
1137 * created because of a false negative won't make it into the hash
1138 * though since that required taking the lock.
1140 spin_lock_bh(&nf_conntrack_lock);
1141 for (i = 0; i < nf_conntrack_htable_size; i++) {
1142 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1143 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1144 struct nf_conntrack_tuple_hash, hnnode);
1145 hlist_nulls_del_rcu(&h->hnnode);
1146 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1147 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1150 old_size = nf_conntrack_htable_size;
1151 old_vmalloced = init_net.ct.hash_vmalloc;
1152 old_hash = init_net.ct.hash;
1154 nf_conntrack_htable_size = hashsize;
1155 init_net.ct.hash_vmalloc = vmalloced;
1156 init_net.ct.hash = hash;
1157 nf_conntrack_hash_rnd = rnd;
1158 spin_unlock_bh(&nf_conntrack_lock);
1160 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1163 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1165 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1166 &nf_conntrack_htable_size, 0600);
1168 static int nf_conntrack_init_init_net(void)
1173 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1174 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1175 if (!nf_conntrack_htable_size) {
1176 nf_conntrack_htable_size
1177 = (((num_physpages << PAGE_SHIFT) / 16384)
1178 / sizeof(struct hlist_head));
1179 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1180 nf_conntrack_htable_size = 16384;
1181 if (nf_conntrack_htable_size < 32)
1182 nf_conntrack_htable_size = 32;
1184 /* Use a max. factor of four by default to get the same max as
1185 * with the old struct list_heads. When a table size is given
1186 * we use the old value of 8 to avoid reducing the max.
1190 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1192 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1193 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1196 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1197 sizeof(struct nf_conn),
1198 0, SLAB_DESTROY_BY_RCU, NULL);
1199 if (!nf_conntrack_cachep) {
1200 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1205 ret = nf_conntrack_proto_init();
1209 ret = nf_conntrack_helper_init();
1216 nf_conntrack_proto_fini();
1218 kmem_cache_destroy(nf_conntrack_cachep);
1223 static int nf_conntrack_init_net(struct net *net)
1227 atomic_set(&net->ct.count, 0);
1228 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, 0);
1229 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1230 if (!net->ct.stat) {
1234 ret = nf_conntrack_ecache_init(net);
1237 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1238 &net->ct.hash_vmalloc, 1);
1239 if (!net->ct.hash) {
1241 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1244 ret = nf_conntrack_expect_init(net);
1247 ret = nf_conntrack_acct_init(net);
1251 /* Set up fake conntrack:
1252 - to never be deleted, not in any hashes */
1253 #ifdef CONFIG_NET_NS
1254 nf_conntrack_untracked.ct_net = &init_net;
1256 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1257 /* - and look it like as a confirmed connection */
1258 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1263 nf_conntrack_expect_fini(net);
1265 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1266 nf_conntrack_htable_size);
1268 nf_conntrack_ecache_fini(net);
1270 free_percpu(net->ct.stat);
1275 int nf_conntrack_init(struct net *net)
1279 if (net_eq(net, &init_net)) {
1280 ret = nf_conntrack_init_init_net();
1284 ret = nf_conntrack_init_net(net);
1288 if (net_eq(net, &init_net)) {
1289 /* For use by REJECT target */
1290 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1291 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1296 if (net_eq(net, &init_net))
1297 nf_conntrack_cleanup_init_net();