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);
402 nf_conntrack_event_cache(master_ct(ct) ?
403 IPCT_RELATED : IPCT_NEW, ct);
407 NF_CT_STAT_INC(net, insert_failed);
408 spin_unlock_bh(&nf_conntrack_lock);
411 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
413 /* Returns true if a connection correspondings to the tuple (required
416 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
417 const struct nf_conn *ignored_conntrack)
419 struct net *net = nf_ct_net(ignored_conntrack);
420 struct nf_conntrack_tuple_hash *h;
421 struct hlist_nulls_node *n;
422 unsigned int hash = hash_conntrack(tuple);
424 /* Disable BHs the entire time since we need to disable them at
425 * least once for the stats anyway.
428 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
429 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
430 nf_ct_tuple_equal(tuple, &h->tuple)) {
431 NF_CT_STAT_INC(net, found);
432 rcu_read_unlock_bh();
435 NF_CT_STAT_INC(net, searched);
437 rcu_read_unlock_bh();
441 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
443 #define NF_CT_EVICTION_RANGE 8
445 /* There's a small race here where we may free a just-assured
446 connection. Too bad: we're in trouble anyway. */
447 static noinline int early_drop(struct net *net, unsigned int hash)
449 /* Use oldest entry, which is roughly LRU */
450 struct nf_conntrack_tuple_hash *h;
451 struct nf_conn *ct = NULL, *tmp;
452 struct hlist_nulls_node *n;
453 unsigned int i, cnt = 0;
457 for (i = 0; i < nf_conntrack_htable_size; i++) {
458 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
460 tmp = nf_ct_tuplehash_to_ctrack(h);
461 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
466 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
468 if (ct || cnt >= NF_CT_EVICTION_RANGE)
470 hash = (hash + 1) % nf_conntrack_htable_size;
477 if (del_timer(&ct->timeout)) {
478 death_by_timeout((unsigned long)ct);
480 NF_CT_STAT_INC_ATOMIC(net, early_drop);
486 struct nf_conn *nf_conntrack_alloc(struct net *net,
487 const struct nf_conntrack_tuple *orig,
488 const struct nf_conntrack_tuple *repl,
493 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
494 get_random_bytes(&nf_conntrack_hash_rnd,
495 sizeof(nf_conntrack_hash_rnd));
496 nf_conntrack_hash_rnd_initted = 1;
499 /* We don't want any race condition at early drop stage */
500 atomic_inc(&net->ct.count);
502 if (nf_conntrack_max &&
503 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
504 unsigned int hash = hash_conntrack(orig);
505 if (!early_drop(net, hash)) {
506 atomic_dec(&net->ct.count);
509 "nf_conntrack: table full, dropping"
511 return ERR_PTR(-ENOMEM);
515 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
517 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
518 atomic_dec(&net->ct.count);
519 return ERR_PTR(-ENOMEM);
522 atomic_set(&ct->ct_general.use, 1);
523 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
524 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
525 /* Don't set timer yet: wait for confirmation */
526 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
533 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
535 void nf_conntrack_free(struct nf_conn *ct)
537 struct net *net = nf_ct_net(ct);
539 nf_ct_ext_destroy(ct);
540 atomic_dec(&net->ct.count);
542 kmem_cache_free(nf_conntrack_cachep, ct);
544 EXPORT_SYMBOL_GPL(nf_conntrack_free);
546 /* Allocate a new conntrack: we return -ENOMEM if classification
547 failed due to stress. Otherwise it really is unclassifiable. */
548 static struct nf_conntrack_tuple_hash *
549 init_conntrack(struct net *net,
550 const struct nf_conntrack_tuple *tuple,
551 struct nf_conntrack_l3proto *l3proto,
552 struct nf_conntrack_l4proto *l4proto,
554 unsigned int dataoff)
557 struct nf_conn_help *help;
558 struct nf_conntrack_tuple repl_tuple;
559 struct nf_conntrack_expect *exp;
561 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
562 pr_debug("Can't invert tuple.\n");
566 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
568 pr_debug("Can't allocate conntrack.\n");
569 return (struct nf_conntrack_tuple_hash *)ct;
572 if (!l4proto->new(ct, skb, dataoff)) {
573 nf_conntrack_free(ct);
574 pr_debug("init conntrack: can't track with proto module\n");
578 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
580 spin_lock_bh(&nf_conntrack_lock);
581 exp = nf_ct_find_expectation(net, tuple);
583 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
585 /* Welcome, Mr. Bond. We've been expecting you... */
586 __set_bit(IPS_EXPECTED_BIT, &ct->status);
587 ct->master = exp->master;
589 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
591 rcu_assign_pointer(help->helper, exp->helper);
594 #ifdef CONFIG_NF_CONNTRACK_MARK
595 ct->mark = exp->master->mark;
597 #ifdef CONFIG_NF_CONNTRACK_SECMARK
598 ct->secmark = exp->master->secmark;
600 nf_conntrack_get(&ct->master->ct_general);
601 NF_CT_STAT_INC(net, expect_new);
603 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
604 NF_CT_STAT_INC(net, new);
607 /* Overload tuple linked list to put us in unconfirmed list. */
608 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
609 &net->ct.unconfirmed);
611 spin_unlock_bh(&nf_conntrack_lock);
615 exp->expectfn(ct, exp);
616 nf_ct_expect_put(exp);
619 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
622 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
623 static inline struct nf_conn *
624 resolve_normal_ct(struct net *net,
626 unsigned int dataoff,
629 struct nf_conntrack_l3proto *l3proto,
630 struct nf_conntrack_l4proto *l4proto,
632 enum ip_conntrack_info *ctinfo)
634 struct nf_conntrack_tuple tuple;
635 struct nf_conntrack_tuple_hash *h;
638 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
639 dataoff, l3num, protonum, &tuple, l3proto,
641 pr_debug("resolve_normal_ct: Can't get tuple\n");
645 /* look for tuple match */
646 h = nf_conntrack_find_get(net, &tuple);
648 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
654 ct = nf_ct_tuplehash_to_ctrack(h);
656 /* It exists; we have (non-exclusive) reference. */
657 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
658 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
659 /* Please set reply bit if this packet OK */
662 /* Once we've had two way comms, always ESTABLISHED. */
663 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
664 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
665 *ctinfo = IP_CT_ESTABLISHED;
666 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
667 pr_debug("nf_conntrack_in: related packet for %p\n",
669 *ctinfo = IP_CT_RELATED;
671 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
676 skb->nfct = &ct->ct_general;
677 skb->nfctinfo = *ctinfo;
682 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
686 enum ip_conntrack_info ctinfo;
687 struct nf_conntrack_l3proto *l3proto;
688 struct nf_conntrack_l4proto *l4proto;
689 unsigned int dataoff;
694 /* Previously seen (loopback or untracked)? Ignore. */
696 NF_CT_STAT_INC_ATOMIC(net, ignore);
700 /* rcu_read_lock()ed by nf_hook_slow */
701 l3proto = __nf_ct_l3proto_find(pf);
702 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
703 &dataoff, &protonum);
705 pr_debug("not prepared to track yet or error occured\n");
706 NF_CT_STAT_INC_ATOMIC(net, error);
707 NF_CT_STAT_INC_ATOMIC(net, invalid);
711 l4proto = __nf_ct_l4proto_find(pf, protonum);
713 /* It may be an special packet, error, unclean...
714 * inverse of the return code tells to the netfilter
715 * core what to do with the packet. */
716 if (l4proto->error != NULL) {
717 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
719 NF_CT_STAT_INC_ATOMIC(net, error);
720 NF_CT_STAT_INC_ATOMIC(net, invalid);
725 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
726 l3proto, l4proto, &set_reply, &ctinfo);
728 /* Not valid part of a connection */
729 NF_CT_STAT_INC_ATOMIC(net, invalid);
734 /* Too stressed to deal. */
735 NF_CT_STAT_INC_ATOMIC(net, drop);
739 NF_CT_ASSERT(skb->nfct);
741 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
743 /* Invalid: inverse of the return code tells
744 * the netfilter core what to do */
745 pr_debug("nf_conntrack_in: Can't track with proto module\n");
746 nf_conntrack_put(skb->nfct);
748 NF_CT_STAT_INC_ATOMIC(net, invalid);
750 NF_CT_STAT_INC_ATOMIC(net, drop);
754 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
755 nf_conntrack_event_cache(IPCT_STATUS, ct);
759 EXPORT_SYMBOL_GPL(nf_conntrack_in);
761 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
762 const struct nf_conntrack_tuple *orig)
767 ret = nf_ct_invert_tuple(inverse, orig,
768 __nf_ct_l3proto_find(orig->src.l3num),
769 __nf_ct_l4proto_find(orig->src.l3num,
770 orig->dst.protonum));
774 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
776 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
777 implicitly racy: see __nf_conntrack_confirm */
778 void nf_conntrack_alter_reply(struct nf_conn *ct,
779 const struct nf_conntrack_tuple *newreply)
781 struct nf_conn_help *help = nfct_help(ct);
783 /* Should be unconfirmed, so not in hash table yet */
784 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
786 pr_debug("Altering reply tuple of %p to ", ct);
787 nf_ct_dump_tuple(newreply);
789 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
790 if (ct->master || (help && !hlist_empty(&help->expectations)))
794 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
797 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
799 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
800 void __nf_ct_refresh_acct(struct nf_conn *ct,
801 enum ip_conntrack_info ctinfo,
802 const struct sk_buff *skb,
803 unsigned long extra_jiffies,
806 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
809 spin_lock_bh(&nf_conntrack_lock);
811 /* Only update if this is not a fixed timeout */
812 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
815 /* If not in hash table, timer will not be active yet */
816 if (!nf_ct_is_confirmed(ct)) {
817 ct->timeout.expires = extra_jiffies;
819 unsigned long newtime = jiffies + extra_jiffies;
821 /* Only update the timeout if the new timeout is at least
822 HZ jiffies from the old timeout. Need del_timer for race
823 avoidance (may already be dying). */
824 if (newtime - ct->timeout.expires >= HZ
825 && del_timer(&ct->timeout)) {
826 ct->timeout.expires = newtime;
827 add_timer(&ct->timeout);
833 struct nf_conn_counter *acct;
835 acct = nf_conn_acct_find(ct);
837 acct[CTINFO2DIR(ctinfo)].packets++;
838 acct[CTINFO2DIR(ctinfo)].bytes +=
839 skb->len - skb_network_offset(skb);
843 spin_unlock_bh(&nf_conntrack_lock);
845 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
847 bool __nf_ct_kill_acct(struct nf_conn *ct,
848 enum ip_conntrack_info ctinfo,
849 const struct sk_buff *skb,
853 struct nf_conn_counter *acct;
855 spin_lock_bh(&nf_conntrack_lock);
856 acct = nf_conn_acct_find(ct);
858 acct[CTINFO2DIR(ctinfo)].packets++;
859 acct[CTINFO2DIR(ctinfo)].bytes +=
860 skb->len - skb_network_offset(skb);
862 spin_unlock_bh(&nf_conntrack_lock);
865 if (del_timer(&ct->timeout)) {
866 ct->timeout.function((unsigned long)ct);
871 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
873 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
875 #include <linux/netfilter/nfnetlink.h>
876 #include <linux/netfilter/nfnetlink_conntrack.h>
877 #include <linux/mutex.h>
879 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
880 * in ip_conntrack_core, since we don't want the protocols to autoload
881 * or depend on ctnetlink */
882 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
883 const struct nf_conntrack_tuple *tuple)
885 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
886 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
892 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
894 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
895 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
896 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
898 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
900 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
901 struct nf_conntrack_tuple *t)
903 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
906 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
907 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
911 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
913 int nf_ct_port_nlattr_tuple_size(void)
915 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
917 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
920 /* Used by ipt_REJECT and ip6t_REJECT. */
921 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
924 enum ip_conntrack_info ctinfo;
926 /* This ICMP is in reverse direction to the packet which caused it */
927 ct = nf_ct_get(skb, &ctinfo);
928 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
929 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
931 ctinfo = IP_CT_RELATED;
933 /* Attach to new skbuff, and increment count */
934 nskb->nfct = &ct->ct_general;
935 nskb->nfctinfo = ctinfo;
936 nf_conntrack_get(nskb->nfct);
939 /* Bring out ya dead! */
940 static struct nf_conn *
941 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
942 void *data, unsigned int *bucket)
944 struct nf_conntrack_tuple_hash *h;
946 struct hlist_nulls_node *n;
948 spin_lock_bh(&nf_conntrack_lock);
949 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
950 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
951 ct = nf_ct_tuplehash_to_ctrack(h);
956 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
957 ct = nf_ct_tuplehash_to_ctrack(h);
959 set_bit(IPS_DYING_BIT, &ct->status);
961 spin_unlock_bh(&nf_conntrack_lock);
964 atomic_inc(&ct->ct_general.use);
965 spin_unlock_bh(&nf_conntrack_lock);
969 void nf_ct_iterate_cleanup(struct net *net,
970 int (*iter)(struct nf_conn *i, void *data),
974 unsigned int bucket = 0;
976 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
977 /* Time to push up daises... */
978 if (del_timer(&ct->timeout))
979 death_by_timeout((unsigned long)ct);
980 /* ... else the timer will get him soon. */
985 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
987 struct __nf_ct_flush_report {
992 static int kill_report(struct nf_conn *i, void *data)
994 struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
996 /* get_next_corpse sets the dying bit for us */
997 nf_conntrack_event_report(IPCT_DESTROY,
1004 static int kill_all(struct nf_conn *i, void *data)
1009 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1014 free_pages((unsigned long)hash,
1015 get_order(sizeof(struct hlist_head) * size));
1017 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1019 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1021 struct __nf_ct_flush_report fr = {
1025 nf_ct_iterate_cleanup(net, kill_report, &fr);
1027 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1029 static void nf_conntrack_cleanup_init_net(void)
1031 nf_conntrack_helper_fini();
1032 nf_conntrack_proto_fini();
1033 kmem_cache_destroy(nf_conntrack_cachep);
1036 static void nf_conntrack_cleanup_net(struct net *net)
1038 nf_ct_event_cache_flush(net);
1039 nf_conntrack_ecache_fini(net);
1041 nf_ct_iterate_cleanup(net, kill_all, NULL);
1042 if (atomic_read(&net->ct.count) != 0) {
1044 goto i_see_dead_people;
1046 /* wait until all references to nf_conntrack_untracked are dropped */
1047 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1050 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1051 nf_conntrack_htable_size);
1052 nf_conntrack_acct_fini(net);
1053 nf_conntrack_expect_fini(net);
1054 free_percpu(net->ct.stat);
1057 /* Mishearing the voices in his head, our hero wonders how he's
1058 supposed to kill the mall. */
1059 void nf_conntrack_cleanup(struct net *net)
1061 if (net_eq(net, &init_net))
1062 rcu_assign_pointer(ip_ct_attach, NULL);
1064 /* This makes sure all current packets have passed through
1065 netfilter framework. Roll on, two-stage module
1069 nf_conntrack_cleanup_net(net);
1071 if (net_eq(net, &init_net)) {
1072 rcu_assign_pointer(nf_ct_destroy, NULL);
1073 nf_conntrack_cleanup_init_net();
1077 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1079 struct hlist_nulls_head *hash;
1080 unsigned int nr_slots, i;
1085 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1086 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1087 sz = nr_slots * sizeof(struct hlist_nulls_head);
1088 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1092 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1093 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1097 for (i = 0; i < nr_slots; i++)
1098 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1102 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1104 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1106 int i, bucket, vmalloced, old_vmalloced;
1107 unsigned int hashsize, old_size;
1109 struct hlist_nulls_head *hash, *old_hash;
1110 struct nf_conntrack_tuple_hash *h;
1112 /* On boot, we can set this without any fancy locking. */
1113 if (!nf_conntrack_htable_size)
1114 return param_set_uint(val, kp);
1116 hashsize = simple_strtoul(val, NULL, 0);
1120 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1124 /* We have to rehahs for the new table anyway, so we also can
1125 * use a newrandom seed */
1126 get_random_bytes(&rnd, sizeof(rnd));
1128 /* Lookups in the old hash might happen in parallel, which means we
1129 * might get false negatives during connection lookup. New connections
1130 * created because of a false negative won't make it into the hash
1131 * though since that required taking the lock.
1133 spin_lock_bh(&nf_conntrack_lock);
1134 for (i = 0; i < nf_conntrack_htable_size; i++) {
1135 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1136 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1137 struct nf_conntrack_tuple_hash, hnnode);
1138 hlist_nulls_del_rcu(&h->hnnode);
1139 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1140 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1143 old_size = nf_conntrack_htable_size;
1144 old_vmalloced = init_net.ct.hash_vmalloc;
1145 old_hash = init_net.ct.hash;
1147 nf_conntrack_htable_size = hashsize;
1148 init_net.ct.hash_vmalloc = vmalloced;
1149 init_net.ct.hash = hash;
1150 nf_conntrack_hash_rnd = rnd;
1151 spin_unlock_bh(&nf_conntrack_lock);
1153 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1156 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1158 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1159 &nf_conntrack_htable_size, 0600);
1161 static int nf_conntrack_init_init_net(void)
1166 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1167 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1168 if (!nf_conntrack_htable_size) {
1169 nf_conntrack_htable_size
1170 = (((num_physpages << PAGE_SHIFT) / 16384)
1171 / sizeof(struct hlist_head));
1172 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1173 nf_conntrack_htable_size = 16384;
1174 if (nf_conntrack_htable_size < 32)
1175 nf_conntrack_htable_size = 32;
1177 /* Use a max. factor of four by default to get the same max as
1178 * with the old struct list_heads. When a table size is given
1179 * we use the old value of 8 to avoid reducing the max.
1183 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1185 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1186 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1189 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1190 sizeof(struct nf_conn),
1191 0, SLAB_DESTROY_BY_RCU, NULL);
1192 if (!nf_conntrack_cachep) {
1193 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1198 ret = nf_conntrack_proto_init();
1202 ret = nf_conntrack_helper_init();
1209 nf_conntrack_proto_fini();
1211 kmem_cache_destroy(nf_conntrack_cachep);
1216 static int nf_conntrack_init_net(struct net *net)
1220 atomic_set(&net->ct.count, 0);
1221 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, 0);
1222 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1223 if (!net->ct.stat) {
1227 ret = nf_conntrack_ecache_init(net);
1230 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1231 &net->ct.hash_vmalloc, 1);
1232 if (!net->ct.hash) {
1234 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1237 ret = nf_conntrack_expect_init(net);
1240 ret = nf_conntrack_acct_init(net);
1244 /* Set up fake conntrack:
1245 - to never be deleted, not in any hashes */
1246 #ifdef CONFIG_NET_NS
1247 nf_conntrack_untracked.ct_net = &init_net;
1249 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1250 /* - and look it like as a confirmed connection */
1251 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1256 nf_conntrack_expect_fini(net);
1258 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1259 nf_conntrack_htable_size);
1261 nf_conntrack_ecache_fini(net);
1263 free_percpu(net->ct.stat);
1268 int nf_conntrack_init(struct net *net)
1272 if (net_eq(net, &init_net)) {
1273 ret = nf_conntrack_init_init_net();
1277 ret = nf_conntrack_init_net(net);
1281 if (net_eq(net, &init_net)) {
1282 /* For use by REJECT target */
1283 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1284 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1289 if (net_eq(net, &init_net))
1290 nf_conntrack_cleanup_init_net();