1 /* Connection state tracking for netfilter. This is separated from,
2 but required by, the NAT layer; it can also be used by an iptables
5 /* (C) 1999-2001 Paul `Rusty' Russell
6 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
7 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/proc_fs.h>
19 #include <linux/vmalloc.h>
20 #include <linux/stddef.h>
21 #include <linux/slab.h>
22 #include <linux/random.h>
23 #include <linux/jhash.h>
24 #include <linux/err.h>
25 #include <linux/percpu.h>
26 #include <linux/moduleparam.h>
27 #include <linux/notifier.h>
28 #include <linux/kernel.h>
29 #include <linux/netdevice.h>
30 #include <linux/socket.h>
33 #include <net/netfilter/nf_conntrack.h>
34 #include <net/netfilter/nf_conntrack_l3proto.h>
35 #include <net/netfilter/nf_conntrack_l4proto.h>
36 #include <net/netfilter/nf_conntrack_expect.h>
37 #include <net/netfilter/nf_conntrack_helper.h>
38 #include <net/netfilter/nf_conntrack_core.h>
39 #include <net/netfilter/nf_conntrack_extend.h>
41 #define NF_CONNTRACK_VERSION "0.5.0"
43 DEFINE_SPINLOCK(nf_conntrack_lock);
44 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
46 /* nf_conntrack_standalone needs this */
47 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
48 EXPORT_SYMBOL_GPL(nf_conntrack_count);
50 unsigned int nf_conntrack_htable_size __read_mostly;
51 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
53 int nf_conntrack_max __read_mostly;
54 EXPORT_SYMBOL_GPL(nf_conntrack_max);
56 struct hlist_head *nf_conntrack_hash __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
59 struct nf_conn nf_conntrack_untracked __read_mostly;
60 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
62 unsigned int nf_ct_log_invalid __read_mostly;
63 HLIST_HEAD(unconfirmed);
64 static int nf_conntrack_vmalloc __read_mostly;
65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
67 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
68 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
70 static int nf_conntrack_hash_rnd_initted;
71 static unsigned int nf_conntrack_hash_rnd;
73 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
74 unsigned int size, unsigned int rnd)
79 /* The direction must be ignored, so we hash everything up to the
80 * destination ports (which is a multiple of 4) and treat the last
81 * three bytes manually.
83 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
84 h = jhash2((u32 *)tuple, n,
85 rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
86 tuple->dst.protonum));
88 return ((u64)h * size) >> 32;
91 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
93 return __hash_conntrack(tuple, nf_conntrack_htable_size,
94 nf_conntrack_hash_rnd);
98 nf_ct_get_tuple(const struct sk_buff *skb,
100 unsigned int dataoff,
103 struct nf_conntrack_tuple *tuple,
104 const struct nf_conntrack_l3proto *l3proto,
105 const struct nf_conntrack_l4proto *l4proto)
107 NF_CT_TUPLE_U_BLANK(tuple);
109 tuple->src.l3num = l3num;
110 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
113 tuple->dst.protonum = protonum;
114 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
116 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
118 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
120 int nf_ct_get_tuplepr(const struct sk_buff *skb,
123 struct nf_conntrack_tuple *tuple)
125 struct nf_conntrack_l3proto *l3proto;
126 struct nf_conntrack_l4proto *l4proto;
127 unsigned int protoff;
133 l3proto = __nf_ct_l3proto_find(l3num);
134 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
135 if (ret != NF_ACCEPT) {
140 l4proto = __nf_ct_l4proto_find(l3num, protonum);
142 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
148 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
151 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
152 const struct nf_conntrack_tuple *orig,
153 const struct nf_conntrack_l3proto *l3proto,
154 const struct nf_conntrack_l4proto *l4proto)
156 NF_CT_TUPLE_U_BLANK(inverse);
158 inverse->src.l3num = orig->src.l3num;
159 if (l3proto->invert_tuple(inverse, orig) == 0)
162 inverse->dst.dir = !orig->dst.dir;
164 inverse->dst.protonum = orig->dst.protonum;
165 return l4proto->invert_tuple(inverse, orig);
167 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
170 clean_from_lists(struct nf_conn *ct)
172 pr_debug("clean_from_lists(%p)\n", ct);
173 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
174 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
176 /* Destroy all pending expectations */
177 nf_ct_remove_expectations(ct);
181 destroy_conntrack(struct nf_conntrack *nfct)
183 struct nf_conn *ct = (struct nf_conn *)nfct;
184 struct nf_conntrack_l4proto *l4proto;
186 pr_debug("destroy_conntrack(%p)\n", ct);
187 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
188 NF_CT_ASSERT(!timer_pending(&ct->timeout));
190 nf_conntrack_event(IPCT_DESTROY, ct);
191 set_bit(IPS_DYING_BIT, &ct->status);
193 /* To make sure we don't get any weird locking issues here:
194 * destroy_conntrack() MUST NOT be called with a write lock
195 * to nf_conntrack_lock!!! -HW */
197 l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
198 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
199 if (l4proto && l4proto->destroy)
200 l4proto->destroy(ct);
202 nf_ct_ext_destroy(ct);
206 spin_lock_bh(&nf_conntrack_lock);
207 /* Expectations will have been removed in clean_from_lists,
208 * except TFTP can create an expectation on the first packet,
209 * before connection is in the list, so we need to clean here,
211 nf_ct_remove_expectations(ct);
213 /* We overload first tuple to link into unconfirmed list. */
214 if (!nf_ct_is_confirmed(ct)) {
215 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
216 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
219 NF_CT_STAT_INC(delete);
220 spin_unlock_bh(&nf_conntrack_lock);
223 nf_ct_put(ct->master);
225 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
226 nf_conntrack_free(ct);
229 static void death_by_timeout(unsigned long ul_conntrack)
231 struct nf_conn *ct = (void *)ul_conntrack;
232 struct nf_conn_help *help = nfct_help(ct);
233 struct nf_conntrack_helper *helper;
237 helper = rcu_dereference(help->helper);
238 if (helper && helper->destroy)
243 spin_lock_bh(&nf_conntrack_lock);
244 /* Inside lock so preempt is disabled on module removal path.
245 * Otherwise we can get spurious warnings. */
246 NF_CT_STAT_INC(delete_list);
247 clean_from_lists(ct);
248 spin_unlock_bh(&nf_conntrack_lock);
252 struct nf_conntrack_tuple_hash *
253 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple)
255 struct nf_conntrack_tuple_hash *h;
256 struct hlist_node *n;
257 unsigned int hash = hash_conntrack(tuple);
259 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
260 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
261 NF_CT_STAT_INC(found);
264 NF_CT_STAT_INC(searched);
269 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
271 /* Find a connection corresponding to a tuple. */
272 struct nf_conntrack_tuple_hash *
273 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
275 struct nf_conntrack_tuple_hash *h;
279 h = __nf_conntrack_find(tuple);
281 ct = nf_ct_tuplehash_to_ctrack(h);
282 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
289 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
291 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
293 unsigned int repl_hash)
295 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
296 &nf_conntrack_hash[hash]);
297 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
298 &nf_conntrack_hash[repl_hash]);
301 void nf_conntrack_hash_insert(struct nf_conn *ct)
303 unsigned int hash, repl_hash;
305 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
306 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
308 spin_lock_bh(&nf_conntrack_lock);
309 __nf_conntrack_hash_insert(ct, hash, repl_hash);
310 spin_unlock_bh(&nf_conntrack_lock);
312 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
314 /* Confirm a connection given skb; places it in hash table */
316 __nf_conntrack_confirm(struct sk_buff *skb)
318 unsigned int hash, repl_hash;
319 struct nf_conntrack_tuple_hash *h;
321 struct nf_conn_help *help;
322 struct hlist_node *n;
323 enum ip_conntrack_info ctinfo;
325 ct = nf_ct_get(skb, &ctinfo);
327 /* ipt_REJECT uses nf_conntrack_attach to attach related
328 ICMP/TCP RST packets in other direction. Actual packet
329 which created connection will be IP_CT_NEW or for an
330 expected connection, IP_CT_RELATED. */
331 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
334 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
335 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
337 /* We're not in hash table, and we refuse to set up related
338 connections for unconfirmed conns. But packet copies and
339 REJECT will give spurious warnings here. */
340 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
342 /* No external references means noone else could have
344 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
345 pr_debug("Confirming conntrack %p\n", ct);
347 spin_lock_bh(&nf_conntrack_lock);
349 /* See if there's one in the list already, including reverse:
350 NAT could have grabbed it without realizing, since we're
351 not in the hash. If there is, we lost race. */
352 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
353 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
356 hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
357 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
361 /* Remove from unconfirmed list */
362 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
364 __nf_conntrack_hash_insert(ct, hash, repl_hash);
365 /* Timer relative to confirmation time, not original
366 setting time, otherwise we'd get timer wrap in
367 weird delay cases. */
368 ct->timeout.expires += jiffies;
369 add_timer(&ct->timeout);
370 atomic_inc(&ct->ct_general.use);
371 set_bit(IPS_CONFIRMED_BIT, &ct->status);
372 NF_CT_STAT_INC(insert);
373 spin_unlock_bh(&nf_conntrack_lock);
374 help = nfct_help(ct);
375 if (help && help->helper)
376 nf_conntrack_event_cache(IPCT_HELPER, skb);
377 #ifdef CONFIG_NF_NAT_NEEDED
378 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
379 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
380 nf_conntrack_event_cache(IPCT_NATINFO, skb);
382 nf_conntrack_event_cache(master_ct(ct) ?
383 IPCT_RELATED : IPCT_NEW, skb);
387 NF_CT_STAT_INC(insert_failed);
388 spin_unlock_bh(&nf_conntrack_lock);
391 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
393 /* Returns true if a connection correspondings to the tuple (required
396 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
397 const struct nf_conn *ignored_conntrack)
399 struct nf_conntrack_tuple_hash *h;
400 struct hlist_node *n;
401 unsigned int hash = hash_conntrack(tuple);
404 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
405 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
406 nf_ct_tuple_equal(tuple, &h->tuple)) {
407 NF_CT_STAT_INC(found);
411 NF_CT_STAT_INC(searched);
417 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
419 #define NF_CT_EVICTION_RANGE 8
421 /* There's a small race here where we may free a just-assured
422 connection. Too bad: we're in trouble anyway. */
423 static noinline int early_drop(unsigned int hash)
425 /* Use oldest entry, which is roughly LRU */
426 struct nf_conntrack_tuple_hash *h;
427 struct nf_conn *ct = NULL, *tmp;
428 struct hlist_node *n;
429 unsigned int i, cnt = 0;
433 for (i = 0; i < nf_conntrack_htable_size; i++) {
434 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash],
436 tmp = nf_ct_tuplehash_to_ctrack(h);
437 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
442 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
444 if (ct || cnt >= NF_CT_EVICTION_RANGE)
446 hash = (hash + 1) % nf_conntrack_htable_size;
453 if (del_timer(&ct->timeout)) {
454 death_by_timeout((unsigned long)ct);
456 NF_CT_STAT_INC_ATOMIC(early_drop);
462 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
463 const struct nf_conntrack_tuple *repl)
465 struct nf_conn *ct = NULL;
467 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
468 get_random_bytes(&nf_conntrack_hash_rnd, 4);
469 nf_conntrack_hash_rnd_initted = 1;
472 /* We don't want any race condition at early drop stage */
473 atomic_inc(&nf_conntrack_count);
475 if (nf_conntrack_max &&
476 unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) {
477 unsigned int hash = hash_conntrack(orig);
478 if (!early_drop(hash)) {
479 atomic_dec(&nf_conntrack_count);
482 "nf_conntrack: table full, dropping"
484 return ERR_PTR(-ENOMEM);
488 ct = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
490 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
491 atomic_dec(&nf_conntrack_count);
492 return ERR_PTR(-ENOMEM);
495 atomic_set(&ct->ct_general.use, 1);
496 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
497 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
498 /* Don't set timer yet: wait for confirmation */
499 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
500 INIT_RCU_HEAD(&ct->rcu);
504 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
506 static void nf_conntrack_free_rcu(struct rcu_head *head)
508 struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
511 kmem_cache_free(nf_conntrack_cachep, ct);
512 atomic_dec(&nf_conntrack_count);
515 void nf_conntrack_free(struct nf_conn *ct)
517 call_rcu(&ct->rcu, nf_conntrack_free_rcu);
519 EXPORT_SYMBOL_GPL(nf_conntrack_free);
521 /* Allocate a new conntrack: we return -ENOMEM if classification
522 failed due to stress. Otherwise it really is unclassifiable. */
523 static struct nf_conntrack_tuple_hash *
524 init_conntrack(const struct nf_conntrack_tuple *tuple,
525 struct nf_conntrack_l3proto *l3proto,
526 struct nf_conntrack_l4proto *l4proto,
528 unsigned int dataoff)
531 struct nf_conn_help *help;
532 struct nf_conntrack_tuple repl_tuple;
533 struct nf_conntrack_expect *exp;
535 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
536 pr_debug("Can't invert tuple.\n");
540 ct = nf_conntrack_alloc(tuple, &repl_tuple);
541 if (ct == NULL || IS_ERR(ct)) {
542 pr_debug("Can't allocate conntrack.\n");
543 return (struct nf_conntrack_tuple_hash *)ct;
546 if (!l4proto->new(ct, skb, dataoff)) {
547 nf_conntrack_free(ct);
548 pr_debug("init conntrack: can't track with proto module\n");
552 spin_lock_bh(&nf_conntrack_lock);
553 exp = nf_ct_find_expectation(tuple);
555 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
557 /* Welcome, Mr. Bond. We've been expecting you... */
558 __set_bit(IPS_EXPECTED_BIT, &ct->status);
559 ct->master = exp->master;
561 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
563 rcu_assign_pointer(help->helper, exp->helper);
566 #ifdef CONFIG_NF_CONNTRACK_MARK
567 ct->mark = exp->master->mark;
569 #ifdef CONFIG_NF_CONNTRACK_SECMARK
570 ct->secmark = exp->master->secmark;
572 nf_conntrack_get(&ct->master->ct_general);
573 NF_CT_STAT_INC(expect_new);
575 struct nf_conntrack_helper *helper;
577 helper = __nf_ct_helper_find(&repl_tuple);
579 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
581 rcu_assign_pointer(help->helper, helper);
586 /* Overload tuple linked list to put us in unconfirmed list. */
587 hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed);
589 spin_unlock_bh(&nf_conntrack_lock);
593 exp->expectfn(ct, exp);
594 nf_ct_expect_put(exp);
597 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
600 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
601 static inline struct nf_conn *
602 resolve_normal_ct(struct sk_buff *skb,
603 unsigned int dataoff,
606 struct nf_conntrack_l3proto *l3proto,
607 struct nf_conntrack_l4proto *l4proto,
609 enum ip_conntrack_info *ctinfo)
611 struct nf_conntrack_tuple tuple;
612 struct nf_conntrack_tuple_hash *h;
615 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
616 dataoff, l3num, protonum, &tuple, l3proto,
618 pr_debug("resolve_normal_ct: Can't get tuple\n");
622 /* look for tuple match */
623 h = nf_conntrack_find_get(&tuple);
625 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
631 ct = nf_ct_tuplehash_to_ctrack(h);
633 /* It exists; we have (non-exclusive) reference. */
634 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
635 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
636 /* Please set reply bit if this packet OK */
639 /* Once we've had two way comms, always ESTABLISHED. */
640 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
641 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
642 *ctinfo = IP_CT_ESTABLISHED;
643 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
644 pr_debug("nf_conntrack_in: related packet for %p\n",
646 *ctinfo = IP_CT_RELATED;
648 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
653 skb->nfct = &ct->ct_general;
654 skb->nfctinfo = *ctinfo;
659 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb)
662 enum ip_conntrack_info ctinfo;
663 struct nf_conntrack_l3proto *l3proto;
664 struct nf_conntrack_l4proto *l4proto;
665 unsigned int dataoff;
670 /* Previously seen (loopback or untracked)? Ignore. */
672 NF_CT_STAT_INC_ATOMIC(ignore);
676 /* rcu_read_lock()ed by nf_hook_slow */
677 l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
678 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
679 &dataoff, &protonum);
681 pr_debug("not prepared to track yet or error occured\n");
682 NF_CT_STAT_INC_ATOMIC(error);
683 NF_CT_STAT_INC_ATOMIC(invalid);
687 l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
689 /* It may be an special packet, error, unclean...
690 * inverse of the return code tells to the netfilter
691 * core what to do with the packet. */
692 if (l4proto->error != NULL &&
693 (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
694 NF_CT_STAT_INC_ATOMIC(error);
695 NF_CT_STAT_INC_ATOMIC(invalid);
699 ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
700 &set_reply, &ctinfo);
702 /* Not valid part of a connection */
703 NF_CT_STAT_INC_ATOMIC(invalid);
708 /* Too stressed to deal. */
709 NF_CT_STAT_INC_ATOMIC(drop);
713 NF_CT_ASSERT(skb->nfct);
715 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
717 /* Invalid: inverse of the return code tells
718 * the netfilter core what to do */
719 pr_debug("nf_conntrack_in: Can't track with proto module\n");
720 nf_conntrack_put(skb->nfct);
722 NF_CT_STAT_INC_ATOMIC(invalid);
726 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
727 nf_conntrack_event_cache(IPCT_STATUS, skb);
731 EXPORT_SYMBOL_GPL(nf_conntrack_in);
733 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
734 const struct nf_conntrack_tuple *orig)
739 ret = nf_ct_invert_tuple(inverse, orig,
740 __nf_ct_l3proto_find(orig->src.l3num),
741 __nf_ct_l4proto_find(orig->src.l3num,
742 orig->dst.protonum));
746 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
748 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
749 implicitly racy: see __nf_conntrack_confirm */
750 void nf_conntrack_alter_reply(struct nf_conn *ct,
751 const struct nf_conntrack_tuple *newreply)
753 struct nf_conn_help *help = nfct_help(ct);
754 struct nf_conntrack_helper *helper;
756 /* Should be unconfirmed, so not in hash table yet */
757 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
759 pr_debug("Altering reply tuple of %p to ", ct);
760 NF_CT_DUMP_TUPLE(newreply);
762 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
763 if (ct->master || (help && help->expecting != 0))
767 helper = __nf_ct_helper_find(newreply);
768 if (helper == NULL) {
770 rcu_assign_pointer(help->helper, NULL);
775 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
779 memset(&help->help, 0, sizeof(help->help));
782 rcu_assign_pointer(help->helper, helper);
786 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
788 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
789 void __nf_ct_refresh_acct(struct nf_conn *ct,
790 enum ip_conntrack_info ctinfo,
791 const struct sk_buff *skb,
792 unsigned long extra_jiffies,
797 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
800 spin_lock_bh(&nf_conntrack_lock);
802 /* Only update if this is not a fixed timeout */
803 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
806 /* If not in hash table, timer will not be active yet */
807 if (!nf_ct_is_confirmed(ct)) {
808 ct->timeout.expires = extra_jiffies;
809 event = IPCT_REFRESH;
811 unsigned long newtime = jiffies + extra_jiffies;
813 /* Only update the timeout if the new timeout is at least
814 HZ jiffies from the old timeout. Need del_timer for race
815 avoidance (may already be dying). */
816 if (newtime - ct->timeout.expires >= HZ
817 && del_timer(&ct->timeout)) {
818 ct->timeout.expires = newtime;
819 add_timer(&ct->timeout);
820 event = IPCT_REFRESH;
825 #ifdef CONFIG_NF_CT_ACCT
827 ct->counters[CTINFO2DIR(ctinfo)].packets++;
828 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
829 skb->len - skb_network_offset(skb);
831 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
832 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
833 event |= IPCT_COUNTER_FILLING;
837 spin_unlock_bh(&nf_conntrack_lock);
839 /* must be unlocked when calling event cache */
841 nf_conntrack_event_cache(event, skb);
843 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
845 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
847 #include <linux/netfilter/nfnetlink.h>
848 #include <linux/netfilter/nfnetlink_conntrack.h>
849 #include <linux/mutex.h>
851 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
852 * in ip_conntrack_core, since we don't want the protocols to autoload
853 * or depend on ctnetlink */
854 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
855 const struct nf_conntrack_tuple *tuple)
857 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
858 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
864 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
866 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
867 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
868 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
870 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
872 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
873 struct nf_conntrack_tuple *t)
875 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
878 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
879 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
883 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
886 /* Used by ipt_REJECT and ip6t_REJECT. */
887 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
890 enum ip_conntrack_info ctinfo;
892 /* This ICMP is in reverse direction to the packet which caused it */
893 ct = nf_ct_get(skb, &ctinfo);
894 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
895 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
897 ctinfo = IP_CT_RELATED;
899 /* Attach to new skbuff, and increment count */
900 nskb->nfct = &ct->ct_general;
901 nskb->nfctinfo = ctinfo;
902 nf_conntrack_get(nskb->nfct);
905 /* Bring out ya dead! */
906 static struct nf_conn *
907 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
908 void *data, unsigned int *bucket)
910 struct nf_conntrack_tuple_hash *h;
912 struct hlist_node *n;
914 spin_lock_bh(&nf_conntrack_lock);
915 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
916 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
917 ct = nf_ct_tuplehash_to_ctrack(h);
922 hlist_for_each_entry(h, n, &unconfirmed, hnode) {
923 ct = nf_ct_tuplehash_to_ctrack(h);
925 set_bit(IPS_DYING_BIT, &ct->status);
927 spin_unlock_bh(&nf_conntrack_lock);
930 atomic_inc(&ct->ct_general.use);
931 spin_unlock_bh(&nf_conntrack_lock);
936 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
939 unsigned int bucket = 0;
941 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
942 /* Time to push up daises... */
943 if (del_timer(&ct->timeout))
944 death_by_timeout((unsigned long)ct);
945 /* ... else the timer will get him soon. */
950 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
952 static int kill_all(struct nf_conn *i, void *data)
957 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
962 free_pages((unsigned long)hash,
963 get_order(sizeof(struct hlist_head) * size));
965 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
967 void nf_conntrack_flush(void)
969 nf_ct_iterate_cleanup(kill_all, NULL);
971 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
973 /* Mishearing the voices in his head, our hero wonders how he's
974 supposed to kill the mall. */
975 void nf_conntrack_cleanup(void)
977 rcu_assign_pointer(ip_ct_attach, NULL);
979 /* This makes sure all current packets have passed through
980 netfilter framework. Roll on, two-stage module
984 nf_ct_event_cache_flush();
986 nf_conntrack_flush();
987 if (atomic_read(&nf_conntrack_count) != 0) {
989 goto i_see_dead_people;
991 /* wait until all references to nf_conntrack_untracked are dropped */
992 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
995 rcu_assign_pointer(nf_ct_destroy, NULL);
997 kmem_cache_destroy(nf_conntrack_cachep);
998 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
999 nf_conntrack_htable_size);
1001 nf_conntrack_proto_fini();
1002 nf_conntrack_helper_fini();
1003 nf_conntrack_expect_fini();
1006 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1008 struct hlist_head *hash;
1009 unsigned int size, i;
1013 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1014 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1015 get_order(sizeof(struct hlist_head)
1019 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1020 hash = vmalloc(sizeof(struct hlist_head) * size);
1024 for (i = 0; i < size; i++)
1025 INIT_HLIST_HEAD(&hash[i]);
1029 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1031 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1033 int i, bucket, vmalloced, old_vmalloced;
1034 unsigned int hashsize, old_size;
1036 struct hlist_head *hash, *old_hash;
1037 struct nf_conntrack_tuple_hash *h;
1039 /* On boot, we can set this without any fancy locking. */
1040 if (!nf_conntrack_htable_size)
1041 return param_set_uint(val, kp);
1043 hashsize = simple_strtoul(val, NULL, 0);
1047 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1051 /* We have to rehahs for the new table anyway, so we also can
1052 * use a newrandom seed */
1053 get_random_bytes(&rnd, 4);
1055 /* Lookups in the old hash might happen in parallel, which means we
1056 * might get false negatives during connection lookup. New connections
1057 * created because of a false negative won't make it into the hash
1058 * though since that required taking the lock.
1060 spin_lock_bh(&nf_conntrack_lock);
1061 for (i = 0; i < nf_conntrack_htable_size; i++) {
1062 while (!hlist_empty(&nf_conntrack_hash[i])) {
1063 h = hlist_entry(nf_conntrack_hash[i].first,
1064 struct nf_conntrack_tuple_hash, hnode);
1065 hlist_del_rcu(&h->hnode);
1066 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1067 hlist_add_head(&h->hnode, &hash[bucket]);
1070 old_size = nf_conntrack_htable_size;
1071 old_vmalloced = nf_conntrack_vmalloc;
1072 old_hash = nf_conntrack_hash;
1074 nf_conntrack_htable_size = hashsize;
1075 nf_conntrack_vmalloc = vmalloced;
1076 nf_conntrack_hash = hash;
1077 nf_conntrack_hash_rnd = rnd;
1078 spin_unlock_bh(&nf_conntrack_lock);
1080 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1083 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1085 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1086 &nf_conntrack_htable_size, 0600);
1088 int __init nf_conntrack_init(void)
1093 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1094 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1095 if (!nf_conntrack_htable_size) {
1096 nf_conntrack_htable_size
1097 = (((num_physpages << PAGE_SHIFT) / 16384)
1098 / sizeof(struct hlist_head));
1099 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1100 nf_conntrack_htable_size = 16384;
1101 if (nf_conntrack_htable_size < 32)
1102 nf_conntrack_htable_size = 32;
1104 /* Use a max. factor of four by default to get the same max as
1105 * with the old struct list_heads. When a table size is given
1106 * we use the old value of 8 to avoid reducing the max.
1110 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1111 &nf_conntrack_vmalloc);
1112 if (!nf_conntrack_hash) {
1113 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1117 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1119 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1120 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1123 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1124 sizeof(struct nf_conn),
1126 if (!nf_conntrack_cachep) {
1127 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1131 ret = nf_conntrack_proto_init();
1133 goto err_free_conntrack_slab;
1135 ret = nf_conntrack_expect_init();
1137 goto out_fini_proto;
1139 ret = nf_conntrack_helper_init();
1141 goto out_fini_expect;
1143 /* For use by REJECT target */
1144 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1145 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1147 /* Set up fake conntrack:
1148 - to never be deleted, not in any hashes */
1149 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1150 /* - and look it like as a confirmed connection */
1151 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1156 nf_conntrack_expect_fini();
1158 nf_conntrack_proto_fini();
1159 err_free_conntrack_slab:
1160 kmem_cache_destroy(nf_conntrack_cachep);
1162 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1163 nf_conntrack_htable_size);