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_conntrack_ecache.h>
43 #include <net/netfilter/nf_nat.h>
44 #include <net/netfilter/nf_nat_core.h>
46 #define NF_CONNTRACK_VERSION "0.5.0"
48 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
49 enum nf_nat_manip_type manip,
50 struct nlattr *attr) __read_mostly;
51 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
53 DEFINE_SPINLOCK(nf_conntrack_lock);
54 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
56 unsigned int nf_conntrack_htable_size __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
59 unsigned int nf_conntrack_max __read_mostly;
60 EXPORT_SYMBOL_GPL(nf_conntrack_max);
62 struct nf_conn nf_conntrack_untracked __read_mostly;
63 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
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_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
169 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
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 net *net = nf_ct_net(ct);
180 struct nf_conntrack_l4proto *l4proto;
182 pr_debug("destroy_conntrack(%p)\n", ct);
183 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
184 NF_CT_ASSERT(!timer_pending(&ct->timeout));
186 if (!test_bit(IPS_DYING_BIT, &ct->status))
187 nf_conntrack_event(IPCT_DESTROY, ct);
188 set_bit(IPS_DYING_BIT, &ct->status);
190 /* To make sure we don't get any weird locking issues here:
191 * destroy_conntrack() MUST NOT be called with a write lock
192 * to nf_conntrack_lock!!! -HW */
194 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
195 if (l4proto && l4proto->destroy)
196 l4proto->destroy(ct);
200 spin_lock_bh(&nf_conntrack_lock);
201 /* Expectations will have been removed in clean_from_lists,
202 * except TFTP can create an expectation on the first packet,
203 * before connection is in the list, so we need to clean here,
205 nf_ct_remove_expectations(ct);
207 /* We overload first tuple to link into unconfirmed list. */
208 if (!nf_ct_is_confirmed(ct)) {
209 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
210 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
213 NF_CT_STAT_INC(net, delete);
214 spin_unlock_bh(&nf_conntrack_lock);
217 nf_ct_put(ct->master);
219 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
220 nf_conntrack_free(ct);
223 static void death_by_timeout(unsigned long ul_conntrack)
225 struct nf_conn *ct = (void *)ul_conntrack;
226 struct net *net = nf_ct_net(ct);
227 struct nf_conn_help *help = nfct_help(ct);
228 struct nf_conntrack_helper *helper;
232 helper = rcu_dereference(help->helper);
233 if (helper && helper->destroy)
238 spin_lock_bh(&nf_conntrack_lock);
239 /* Inside lock so preempt is disabled on module removal path.
240 * Otherwise we can get spurious warnings. */
241 NF_CT_STAT_INC(net, delete_list);
242 clean_from_lists(ct);
243 spin_unlock_bh(&nf_conntrack_lock);
249 * - Caller must take a reference on returned object
250 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
252 * - Caller must lock nf_conntrack_lock before calling this function
254 struct nf_conntrack_tuple_hash *
255 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
257 struct nf_conntrack_tuple_hash *h;
258 struct hlist_nulls_node *n;
259 unsigned int hash = hash_conntrack(tuple);
261 /* Disable BHs the entire time since we normally need to disable them
262 * at least once for the stats anyway.
266 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
267 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
268 NF_CT_STAT_INC(net, found);
272 NF_CT_STAT_INC(net, searched);
275 * if the nulls value we got at the end of this lookup is
276 * not the expected one, we must restart lookup.
277 * We probably met an item that was moved to another chain.
279 if (get_nulls_value(n) != hash)
285 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
287 /* Find a connection corresponding to a tuple. */
288 struct nf_conntrack_tuple_hash *
289 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
291 struct nf_conntrack_tuple_hash *h;
296 h = __nf_conntrack_find(net, tuple);
298 ct = nf_ct_tuplehash_to_ctrack(h);
299 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
302 if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
312 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
314 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
316 unsigned int repl_hash)
318 struct net *net = nf_ct_net(ct);
320 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
321 &net->ct.hash[hash]);
322 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
323 &net->ct.hash[repl_hash]);
326 void nf_conntrack_hash_insert(struct nf_conn *ct)
328 unsigned int hash, repl_hash;
330 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
331 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
333 __nf_conntrack_hash_insert(ct, hash, repl_hash);
335 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
337 /* Confirm a connection given skb; places it in hash table */
339 __nf_conntrack_confirm(struct sk_buff *skb)
341 unsigned int hash, repl_hash;
342 struct nf_conntrack_tuple_hash *h;
344 struct nf_conn_help *help;
345 struct hlist_nulls_node *n;
346 enum ip_conntrack_info ctinfo;
349 ct = nf_ct_get(skb, &ctinfo);
352 /* ipt_REJECT uses nf_conntrack_attach to attach related
353 ICMP/TCP RST packets in other direction. Actual packet
354 which created connection will be IP_CT_NEW or for an
355 expected connection, IP_CT_RELATED. */
356 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
359 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
360 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
362 /* We're not in hash table, and we refuse to set up related
363 connections for unconfirmed conns. But packet copies and
364 REJECT will give spurious warnings here. */
365 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
367 /* No external references means noone else could have
369 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
370 pr_debug("Confirming conntrack %p\n", ct);
372 spin_lock_bh(&nf_conntrack_lock);
374 /* See if there's one in the list already, including reverse:
375 NAT could have grabbed it without realizing, since we're
376 not in the hash. If there is, we lost race. */
377 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
378 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
381 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
382 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
386 /* Remove from unconfirmed list */
387 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
389 __nf_conntrack_hash_insert(ct, hash, repl_hash);
390 /* Timer relative to confirmation time, not original
391 setting time, otherwise we'd get timer wrap in
392 weird delay cases. */
393 ct->timeout.expires += jiffies;
394 add_timer(&ct->timeout);
395 atomic_inc(&ct->ct_general.use);
396 set_bit(IPS_CONFIRMED_BIT, &ct->status);
397 NF_CT_STAT_INC(net, insert);
398 spin_unlock_bh(&nf_conntrack_lock);
399 help = nfct_help(ct);
400 if (help && help->helper)
401 nf_conntrack_event_cache(IPCT_HELPER, ct);
403 nf_conntrack_event_cache(master_ct(ct) ?
404 IPCT_RELATED : IPCT_NEW, ct);
408 NF_CT_STAT_INC(net, insert_failed);
409 spin_unlock_bh(&nf_conntrack_lock);
412 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
414 /* Returns true if a connection correspondings to the tuple (required
417 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
418 const struct nf_conn *ignored_conntrack)
420 struct net *net = nf_ct_net(ignored_conntrack);
421 struct nf_conntrack_tuple_hash *h;
422 struct hlist_nulls_node *n;
423 unsigned int hash = hash_conntrack(tuple);
425 /* Disable BHs the entire time since we need to disable them at
426 * least once for the stats anyway.
429 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
430 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
431 nf_ct_tuple_equal(tuple, &h->tuple)) {
432 NF_CT_STAT_INC(net, found);
433 rcu_read_unlock_bh();
436 NF_CT_STAT_INC(net, searched);
438 rcu_read_unlock_bh();
442 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
444 #define NF_CT_EVICTION_RANGE 8
446 /* There's a small race here where we may free a just-assured
447 connection. Too bad: we're in trouble anyway. */
448 static noinline int early_drop(struct net *net, unsigned int hash)
450 /* Use oldest entry, which is roughly LRU */
451 struct nf_conntrack_tuple_hash *h;
452 struct nf_conn *ct = NULL, *tmp;
453 struct hlist_nulls_node *n;
454 unsigned int i, cnt = 0;
458 for (i = 0; i < nf_conntrack_htable_size; i++) {
459 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
461 tmp = nf_ct_tuplehash_to_ctrack(h);
462 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
467 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
469 if (ct || cnt >= NF_CT_EVICTION_RANGE)
471 hash = (hash + 1) % nf_conntrack_htable_size;
478 if (del_timer(&ct->timeout)) {
479 death_by_timeout((unsigned long)ct);
481 NF_CT_STAT_INC_ATOMIC(net, early_drop);
487 struct nf_conn *nf_conntrack_alloc(struct net *net,
488 const struct nf_conntrack_tuple *orig,
489 const struct nf_conntrack_tuple *repl,
494 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
495 get_random_bytes(&nf_conntrack_hash_rnd,
496 sizeof(nf_conntrack_hash_rnd));
497 nf_conntrack_hash_rnd_initted = 1;
500 /* We don't want any race condition at early drop stage */
501 atomic_inc(&net->ct.count);
503 if (nf_conntrack_max &&
504 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
505 unsigned int hash = hash_conntrack(orig);
506 if (!early_drop(net, hash)) {
507 atomic_dec(&net->ct.count);
510 "nf_conntrack: table full, dropping"
512 return ERR_PTR(-ENOMEM);
516 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
518 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
519 atomic_dec(&net->ct.count);
520 return ERR_PTR(-ENOMEM);
523 spin_lock_init(&ct->lock);
524 atomic_set(&ct->ct_general.use, 1);
525 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
526 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
527 /* Don't set timer yet: wait for confirmation */
528 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
535 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
537 void nf_conntrack_free(struct nf_conn *ct)
539 struct net *net = nf_ct_net(ct);
541 nf_ct_ext_destroy(ct);
542 atomic_dec(&net->ct.count);
544 kmem_cache_free(nf_conntrack_cachep, ct);
546 EXPORT_SYMBOL_GPL(nf_conntrack_free);
548 /* Allocate a new conntrack: we return -ENOMEM if classification
549 failed due to stress. Otherwise it really is unclassifiable. */
550 static struct nf_conntrack_tuple_hash *
551 init_conntrack(struct net *net,
552 const struct nf_conntrack_tuple *tuple,
553 struct nf_conntrack_l3proto *l3proto,
554 struct nf_conntrack_l4proto *l4proto,
556 unsigned int dataoff)
559 struct nf_conn_help *help;
560 struct nf_conntrack_tuple repl_tuple;
561 struct nf_conntrack_expect *exp;
563 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
564 pr_debug("Can't invert tuple.\n");
568 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
570 pr_debug("Can't allocate conntrack.\n");
571 return (struct nf_conntrack_tuple_hash *)ct;
574 if (!l4proto->new(ct, skb, dataoff)) {
575 nf_conntrack_free(ct);
576 pr_debug("init conntrack: can't track with proto module\n");
580 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
581 nf_ct_ecache_ext_add(ct, GFP_ATOMIC);
583 spin_lock_bh(&nf_conntrack_lock);
584 exp = nf_ct_find_expectation(net, tuple);
586 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
588 /* Welcome, Mr. Bond. We've been expecting you... */
589 __set_bit(IPS_EXPECTED_BIT, &ct->status);
590 ct->master = exp->master;
592 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
594 rcu_assign_pointer(help->helper, exp->helper);
597 #ifdef CONFIG_NF_CONNTRACK_MARK
598 ct->mark = exp->master->mark;
600 #ifdef CONFIG_NF_CONNTRACK_SECMARK
601 ct->secmark = exp->master->secmark;
603 nf_conntrack_get(&ct->master->ct_general);
604 NF_CT_STAT_INC(net, expect_new);
606 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
607 NF_CT_STAT_INC(net, new);
610 /* Overload tuple linked list to put us in unconfirmed list. */
611 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
612 &net->ct.unconfirmed);
614 spin_unlock_bh(&nf_conntrack_lock);
618 exp->expectfn(ct, exp);
619 nf_ct_expect_put(exp);
622 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
625 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
626 static inline struct nf_conn *
627 resolve_normal_ct(struct net *net,
629 unsigned int dataoff,
632 struct nf_conntrack_l3proto *l3proto,
633 struct nf_conntrack_l4proto *l4proto,
635 enum ip_conntrack_info *ctinfo)
637 struct nf_conntrack_tuple tuple;
638 struct nf_conntrack_tuple_hash *h;
641 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
642 dataoff, l3num, protonum, &tuple, l3proto,
644 pr_debug("resolve_normal_ct: Can't get tuple\n");
648 /* look for tuple match */
649 h = nf_conntrack_find_get(net, &tuple);
651 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
657 ct = nf_ct_tuplehash_to_ctrack(h);
659 /* It exists; we have (non-exclusive) reference. */
660 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
661 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
662 /* Please set reply bit if this packet OK */
665 /* Once we've had two way comms, always ESTABLISHED. */
666 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
667 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
668 *ctinfo = IP_CT_ESTABLISHED;
669 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
670 pr_debug("nf_conntrack_in: related packet for %p\n",
672 *ctinfo = IP_CT_RELATED;
674 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
679 skb->nfct = &ct->ct_general;
680 skb->nfctinfo = *ctinfo;
685 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
689 enum ip_conntrack_info ctinfo;
690 struct nf_conntrack_l3proto *l3proto;
691 struct nf_conntrack_l4proto *l4proto;
692 unsigned int dataoff;
697 /* Previously seen (loopback or untracked)? Ignore. */
699 NF_CT_STAT_INC_ATOMIC(net, ignore);
703 /* rcu_read_lock()ed by nf_hook_slow */
704 l3proto = __nf_ct_l3proto_find(pf);
705 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
706 &dataoff, &protonum);
708 pr_debug("not prepared to track yet or error occured\n");
709 NF_CT_STAT_INC_ATOMIC(net, error);
710 NF_CT_STAT_INC_ATOMIC(net, invalid);
714 l4proto = __nf_ct_l4proto_find(pf, protonum);
716 /* It may be an special packet, error, unclean...
717 * inverse of the return code tells to the netfilter
718 * core what to do with the packet. */
719 if (l4proto->error != NULL) {
720 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
722 NF_CT_STAT_INC_ATOMIC(net, error);
723 NF_CT_STAT_INC_ATOMIC(net, invalid);
728 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
729 l3proto, l4proto, &set_reply, &ctinfo);
731 /* Not valid part of a connection */
732 NF_CT_STAT_INC_ATOMIC(net, invalid);
737 /* Too stressed to deal. */
738 NF_CT_STAT_INC_ATOMIC(net, drop);
742 NF_CT_ASSERT(skb->nfct);
744 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
746 /* Invalid: inverse of the return code tells
747 * the netfilter core what to do */
748 pr_debug("nf_conntrack_in: Can't track with proto module\n");
749 nf_conntrack_put(skb->nfct);
751 NF_CT_STAT_INC_ATOMIC(net, invalid);
753 NF_CT_STAT_INC_ATOMIC(net, drop);
757 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
758 nf_conntrack_event_cache(IPCT_STATUS, ct);
762 EXPORT_SYMBOL_GPL(nf_conntrack_in);
764 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
765 const struct nf_conntrack_tuple *orig)
770 ret = nf_ct_invert_tuple(inverse, orig,
771 __nf_ct_l3proto_find(orig->src.l3num),
772 __nf_ct_l4proto_find(orig->src.l3num,
773 orig->dst.protonum));
777 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
779 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
780 implicitly racy: see __nf_conntrack_confirm */
781 void nf_conntrack_alter_reply(struct nf_conn *ct,
782 const struct nf_conntrack_tuple *newreply)
784 struct nf_conn_help *help = nfct_help(ct);
786 /* Should be unconfirmed, so not in hash table yet */
787 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
789 pr_debug("Altering reply tuple of %p to ", ct);
790 nf_ct_dump_tuple(newreply);
792 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
793 if (ct->master || (help && !hlist_empty(&help->expectations)))
797 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
800 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
802 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
803 void __nf_ct_refresh_acct(struct nf_conn *ct,
804 enum ip_conntrack_info ctinfo,
805 const struct sk_buff *skb,
806 unsigned long extra_jiffies,
809 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
812 /* Only update if this is not a fixed timeout */
813 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
816 /* If not in hash table, timer will not be active yet */
817 if (!nf_ct_is_confirmed(ct)) {
818 ct->timeout.expires = extra_jiffies;
820 unsigned long newtime = jiffies + extra_jiffies;
822 /* Only update the timeout if the new timeout is at least
823 HZ jiffies from the old timeout. Need del_timer for race
824 avoidance (may already be dying). */
825 if (newtime - ct->timeout.expires >= HZ)
826 mod_timer_pending(&ct->timeout, newtime);
831 struct nf_conn_counter *acct;
833 acct = nf_conn_acct_find(ct);
835 spin_lock_bh(&ct->lock);
836 acct[CTINFO2DIR(ctinfo)].packets++;
837 acct[CTINFO2DIR(ctinfo)].bytes +=
838 skb->len - skb_network_offset(skb);
839 spin_unlock_bh(&ct->lock);
843 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
845 bool __nf_ct_kill_acct(struct nf_conn *ct,
846 enum ip_conntrack_info ctinfo,
847 const struct sk_buff *skb,
851 struct nf_conn_counter *acct;
853 acct = nf_conn_acct_find(ct);
855 spin_lock_bh(&ct->lock);
856 acct[CTINFO2DIR(ctinfo)].packets++;
857 acct[CTINFO2DIR(ctinfo)].bytes +=
858 skb->len - skb_network_offset(skb);
859 spin_unlock_bh(&ct->lock);
863 if (del_timer(&ct->timeout)) {
864 ct->timeout.function((unsigned long)ct);
869 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
871 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
873 #include <linux/netfilter/nfnetlink.h>
874 #include <linux/netfilter/nfnetlink_conntrack.h>
875 #include <linux/mutex.h>
877 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
878 * in ip_conntrack_core, since we don't want the protocols to autoload
879 * or depend on ctnetlink */
880 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
881 const struct nf_conntrack_tuple *tuple)
883 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
884 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
890 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
892 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
893 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
894 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
896 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
898 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
899 struct nf_conntrack_tuple *t)
901 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
904 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
905 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
909 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
911 int nf_ct_port_nlattr_tuple_size(void)
913 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
915 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
918 /* Used by ipt_REJECT and ip6t_REJECT. */
919 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
922 enum ip_conntrack_info ctinfo;
924 /* This ICMP is in reverse direction to the packet which caused it */
925 ct = nf_ct_get(skb, &ctinfo);
926 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
927 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
929 ctinfo = IP_CT_RELATED;
931 /* Attach to new skbuff, and increment count */
932 nskb->nfct = &ct->ct_general;
933 nskb->nfctinfo = ctinfo;
934 nf_conntrack_get(nskb->nfct);
937 /* Bring out ya dead! */
938 static struct nf_conn *
939 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
940 void *data, unsigned int *bucket)
942 struct nf_conntrack_tuple_hash *h;
944 struct hlist_nulls_node *n;
946 spin_lock_bh(&nf_conntrack_lock);
947 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
948 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
949 ct = nf_ct_tuplehash_to_ctrack(h);
954 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
955 ct = nf_ct_tuplehash_to_ctrack(h);
957 set_bit(IPS_DYING_BIT, &ct->status);
959 spin_unlock_bh(&nf_conntrack_lock);
962 atomic_inc(&ct->ct_general.use);
963 spin_unlock_bh(&nf_conntrack_lock);
967 void nf_ct_iterate_cleanup(struct net *net,
968 int (*iter)(struct nf_conn *i, void *data),
972 unsigned int bucket = 0;
974 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
975 /* Time to push up daises... */
976 if (del_timer(&ct->timeout))
977 death_by_timeout((unsigned long)ct);
978 /* ... else the timer will get him soon. */
983 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
985 struct __nf_ct_flush_report {
990 static int kill_report(struct nf_conn *i, void *data)
992 struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
994 /* get_next_corpse sets the dying bit for us */
995 nf_conntrack_event_report(IPCT_DESTROY,
1002 static int kill_all(struct nf_conn *i, void *data)
1007 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1012 free_pages((unsigned long)hash,
1013 get_order(sizeof(struct hlist_head) * size));
1015 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1017 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1019 struct __nf_ct_flush_report fr = {
1023 nf_ct_iterate_cleanup(net, kill_report, &fr);
1025 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1027 static void nf_conntrack_cleanup_init_net(void)
1029 nf_conntrack_helper_fini();
1030 nf_conntrack_proto_fini();
1031 kmem_cache_destroy(nf_conntrack_cachep);
1034 static void nf_conntrack_cleanup_net(struct net *net)
1037 nf_ct_iterate_cleanup(net, kill_all, NULL);
1038 if (atomic_read(&net->ct.count) != 0) {
1040 goto i_see_dead_people;
1042 /* wait until all references to nf_conntrack_untracked are dropped */
1043 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1046 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1047 nf_conntrack_htable_size);
1048 nf_conntrack_ecache_fini(net);
1049 nf_conntrack_acct_fini(net);
1050 nf_conntrack_expect_fini(net);
1051 free_percpu(net->ct.stat);
1054 /* Mishearing the voices in his head, our hero wonders how he's
1055 supposed to kill the mall. */
1056 void nf_conntrack_cleanup(struct net *net)
1058 if (net_eq(net, &init_net))
1059 rcu_assign_pointer(ip_ct_attach, NULL);
1061 /* This makes sure all current packets have passed through
1062 netfilter framework. Roll on, two-stage module
1066 nf_conntrack_cleanup_net(net);
1068 if (net_eq(net, &init_net)) {
1069 rcu_assign_pointer(nf_ct_destroy, NULL);
1070 nf_conntrack_cleanup_init_net();
1074 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1076 struct hlist_nulls_head *hash;
1077 unsigned int nr_slots, i;
1082 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1083 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1084 sz = nr_slots * sizeof(struct hlist_nulls_head);
1085 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1089 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1090 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1094 for (i = 0; i < nr_slots; i++)
1095 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1099 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1101 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1103 int i, bucket, vmalloced, old_vmalloced;
1104 unsigned int hashsize, old_size;
1106 struct hlist_nulls_head *hash, *old_hash;
1107 struct nf_conntrack_tuple_hash *h;
1109 /* On boot, we can set this without any fancy locking. */
1110 if (!nf_conntrack_htable_size)
1111 return param_set_uint(val, kp);
1113 hashsize = simple_strtoul(val, NULL, 0);
1117 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1121 /* We have to rehahs for the new table anyway, so we also can
1122 * use a newrandom seed */
1123 get_random_bytes(&rnd, sizeof(rnd));
1125 /* Lookups in the old hash might happen in parallel, which means we
1126 * might get false negatives during connection lookup. New connections
1127 * created because of a false negative won't make it into the hash
1128 * though since that required taking the lock.
1130 spin_lock_bh(&nf_conntrack_lock);
1131 for (i = 0; i < nf_conntrack_htable_size; i++) {
1132 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1133 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1134 struct nf_conntrack_tuple_hash, hnnode);
1135 hlist_nulls_del_rcu(&h->hnnode);
1136 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1137 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1140 old_size = nf_conntrack_htable_size;
1141 old_vmalloced = init_net.ct.hash_vmalloc;
1142 old_hash = init_net.ct.hash;
1144 nf_conntrack_htable_size = hashsize;
1145 init_net.ct.hash_vmalloc = vmalloced;
1146 init_net.ct.hash = hash;
1147 nf_conntrack_hash_rnd = rnd;
1148 spin_unlock_bh(&nf_conntrack_lock);
1150 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1153 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1155 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1156 &nf_conntrack_htable_size, 0600);
1158 static int nf_conntrack_init_init_net(void)
1163 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1164 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1165 if (!nf_conntrack_htable_size) {
1166 nf_conntrack_htable_size
1167 = (((num_physpages << PAGE_SHIFT) / 16384)
1168 / sizeof(struct hlist_head));
1169 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1170 nf_conntrack_htable_size = 16384;
1171 if (nf_conntrack_htable_size < 32)
1172 nf_conntrack_htable_size = 32;
1174 /* Use a max. factor of four by default to get the same max as
1175 * with the old struct list_heads. When a table size is given
1176 * we use the old value of 8 to avoid reducing the max.
1180 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1182 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1183 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1186 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1187 sizeof(struct nf_conn),
1188 0, SLAB_DESTROY_BY_RCU, NULL);
1189 if (!nf_conntrack_cachep) {
1190 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1195 ret = nf_conntrack_proto_init();
1199 ret = nf_conntrack_helper_init();
1206 nf_conntrack_proto_fini();
1208 kmem_cache_destroy(nf_conntrack_cachep);
1213 static int nf_conntrack_init_net(struct net *net)
1217 atomic_set(&net->ct.count, 0);
1218 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, 0);
1219 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1220 if (!net->ct.stat) {
1224 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1225 &net->ct.hash_vmalloc, 1);
1226 if (!net->ct.hash) {
1228 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1231 ret = nf_conntrack_expect_init(net);
1234 ret = nf_conntrack_acct_init(net);
1237 ret = nf_conntrack_ecache_init(net);
1241 /* Set up fake conntrack:
1242 - to never be deleted, not in any hashes */
1243 #ifdef CONFIG_NET_NS
1244 nf_conntrack_untracked.ct_net = &init_net;
1246 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1247 /* - and look it like as a confirmed connection */
1248 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1253 nf_conntrack_acct_fini(net);
1255 nf_conntrack_expect_fini(net);
1257 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1258 nf_conntrack_htable_size);
1260 free_percpu(net->ct.stat);
1265 int nf_conntrack_init(struct net *net)
1269 if (net_eq(net, &init_net)) {
1270 ret = nf_conntrack_init_init_net();
1274 ret = nf_conntrack_init_net(net);
1278 if (net_eq(net, &init_net)) {
1279 /* For use by REJECT target */
1280 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1281 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1286 if (net_eq(net, &init_net))
1287 nf_conntrack_cleanup_init_net();