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.
13 * 23 Apr 2001: Harald Welte <laforge@gnumonks.org>
14 * - new API and handling of conntrack/nat helpers
15 * - now capable of multiple expectations for one master
16 * 16 Jul 2002: Harald Welte <laforge@gnumonks.org>
17 * - add usage/reference counts to ip_conntrack_expect
18 * - export ip_conntrack[_expect]_{find_get,put} functions
19 * 16 Dec 2003: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
20 * - generalize L3 protocol denendent part.
21 * 23 Mar 2004: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
22 * - add support various size of conntrack structures.
23 * 26 Jan 2006: Harald Welte <laforge@netfilter.org>
24 * - restructure nf_conn (introduce nf_conn_help)
25 * - redesign 'features' how they were originally intended
26 * 26 Feb 2006: Pablo Neira Ayuso <pablo@eurodev.net>
27 * - add support for L3 protocol module load on demand.
29 * Derived from net/ipv4/netfilter/ip_conntrack_core.c
32 #include <linux/types.h>
33 #include <linux/netfilter.h>
34 #include <linux/module.h>
35 #include <linux/skbuff.h>
36 #include <linux/proc_fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/stddef.h>
39 #include <linux/slab.h>
40 #include <linux/random.h>
41 #include <linux/jhash.h>
42 #include <linux/err.h>
43 #include <linux/percpu.h>
44 #include <linux/moduleparam.h>
45 #include <linux/notifier.h>
46 #include <linux/kernel.h>
47 #include <linux/netdevice.h>
48 #include <linux/socket.h>
51 #include <net/netfilter/nf_conntrack.h>
52 #include <net/netfilter/nf_conntrack_l3proto.h>
53 #include <net/netfilter/nf_conntrack_l4proto.h>
54 #include <net/netfilter/nf_conntrack_expect.h>
55 #include <net/netfilter/nf_conntrack_helper.h>
56 #include <net/netfilter/nf_conntrack_core.h>
58 #define NF_CONNTRACK_VERSION "0.5.0"
63 #define DEBUGP(format, args...)
66 DEFINE_RWLOCK(nf_conntrack_lock);
67 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
69 /* nf_conntrack_standalone needs this */
70 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
71 EXPORT_SYMBOL_GPL(nf_conntrack_count);
73 void (*nf_conntrack_destroyed)(struct nf_conn *conntrack);
74 EXPORT_SYMBOL_GPL(nf_conntrack_destroyed);
76 unsigned int nf_conntrack_htable_size __read_mostly;
77 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
79 int nf_conntrack_max __read_mostly;
80 EXPORT_SYMBOL_GPL(nf_conntrack_max);
82 struct list_head *nf_conntrack_hash __read_mostly;
83 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
85 struct nf_conn nf_conntrack_untracked __read_mostly;
86 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
88 unsigned int nf_ct_log_invalid __read_mostly;
89 LIST_HEAD(unconfirmed);
90 static int nf_conntrack_vmalloc __read_mostly;
92 static unsigned int nf_conntrack_next_id;
94 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
95 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
98 * This scheme offers various size of "struct nf_conn" dependent on
99 * features(helper, nat, ...)
102 #define NF_CT_FEATURES_NAMELEN 256
104 /* name of slab cache. printed in /proc/slabinfo */
107 /* size of slab cache */
110 /* slab cache pointer */
111 struct kmem_cache *cachep;
113 /* allocated slab cache + modules which uses this slab cache */
116 } nf_ct_cache[NF_CT_F_NUM];
118 /* protect members of nf_ct_cache except of "use" */
119 DEFINE_RWLOCK(nf_ct_cache_lock);
121 /* This avoids calling kmem_cache_create() with same name simultaneously */
122 static DEFINE_MUTEX(nf_ct_cache_mutex);
124 static int nf_conntrack_hash_rnd_initted;
125 static unsigned int nf_conntrack_hash_rnd;
127 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
128 unsigned int size, unsigned int rnd)
132 a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
133 (tuple->src.l3num << 16) | tuple->dst.protonum);
134 b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
135 (tuple->src.u.all << 16) | tuple->dst.u.all);
137 return jhash_2words(a, b, rnd) % size;
140 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
142 return __hash_conntrack(tuple, nf_conntrack_htable_size,
143 nf_conntrack_hash_rnd);
146 int nf_conntrack_register_cache(u_int32_t features, const char *name,
151 struct kmem_cache *cachep;
153 DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n",
154 features, name, size);
156 if (features < NF_CT_F_BASIC || features >= NF_CT_F_NUM) {
157 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n",
162 mutex_lock(&nf_ct_cache_mutex);
164 write_lock_bh(&nf_ct_cache_lock);
165 /* e.g: multiple helpers are loaded */
166 if (nf_ct_cache[features].use > 0) {
167 DEBUGP("nf_conntrack_register_cache: already resisterd.\n");
168 if ((!strncmp(nf_ct_cache[features].name, name,
169 NF_CT_FEATURES_NAMELEN))
170 && nf_ct_cache[features].size == size) {
171 DEBUGP("nf_conntrack_register_cache: reusing.\n");
172 nf_ct_cache[features].use++;
177 write_unlock_bh(&nf_ct_cache_lock);
178 mutex_unlock(&nf_ct_cache_mutex);
181 write_unlock_bh(&nf_ct_cache_lock);
184 * The memory space for name of slab cache must be alive until
185 * cache is destroyed.
187 cache_name = kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN, GFP_ATOMIC);
188 if (cache_name == NULL) {
189 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n");
194 if (strlcpy(cache_name, name, NF_CT_FEATURES_NAMELEN)
195 >= NF_CT_FEATURES_NAMELEN) {
196 printk("nf_conntrack_register_cache: name too long\n");
201 cachep = kmem_cache_create(cache_name, size, 0, 0,
204 printk("nf_conntrack_register_cache: Can't create slab cache "
205 "for the features = 0x%x\n", features);
210 write_lock_bh(&nf_ct_cache_lock);
211 nf_ct_cache[features].use = 1;
212 nf_ct_cache[features].size = size;
213 nf_ct_cache[features].cachep = cachep;
214 nf_ct_cache[features].name = cache_name;
215 write_unlock_bh(&nf_ct_cache_lock);
222 mutex_unlock(&nf_ct_cache_mutex);
225 EXPORT_SYMBOL_GPL(nf_conntrack_register_cache);
227 /* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */
228 void nf_conntrack_unregister_cache(u_int32_t features)
230 struct kmem_cache *cachep;
234 * This assures that kmem_cache_create() isn't called before destroying
237 DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features);
238 mutex_lock(&nf_ct_cache_mutex);
240 write_lock_bh(&nf_ct_cache_lock);
241 if (--nf_ct_cache[features].use > 0) {
242 write_unlock_bh(&nf_ct_cache_lock);
243 mutex_unlock(&nf_ct_cache_mutex);
246 cachep = nf_ct_cache[features].cachep;
247 name = nf_ct_cache[features].name;
248 nf_ct_cache[features].cachep = NULL;
249 nf_ct_cache[features].name = NULL;
250 nf_ct_cache[features].size = 0;
251 write_unlock_bh(&nf_ct_cache_lock);
255 kmem_cache_destroy(cachep);
258 mutex_unlock(&nf_ct_cache_mutex);
260 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_cache);
263 nf_ct_get_tuple(const struct sk_buff *skb,
265 unsigned int dataoff,
268 struct nf_conntrack_tuple *tuple,
269 const struct nf_conntrack_l3proto *l3proto,
270 const struct nf_conntrack_l4proto *l4proto)
272 NF_CT_TUPLE_U_BLANK(tuple);
274 tuple->src.l3num = l3num;
275 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
278 tuple->dst.protonum = protonum;
279 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
281 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
283 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
286 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
287 const struct nf_conntrack_tuple *orig,
288 const struct nf_conntrack_l3proto *l3proto,
289 const struct nf_conntrack_l4proto *l4proto)
291 NF_CT_TUPLE_U_BLANK(inverse);
293 inverse->src.l3num = orig->src.l3num;
294 if (l3proto->invert_tuple(inverse, orig) == 0)
297 inverse->dst.dir = !orig->dst.dir;
299 inverse->dst.protonum = orig->dst.protonum;
300 return l4proto->invert_tuple(inverse, orig);
302 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
305 clean_from_lists(struct nf_conn *ct)
307 DEBUGP("clean_from_lists(%p)\n", ct);
308 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
309 list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list);
311 /* Destroy all pending expectations */
312 nf_ct_remove_expectations(ct);
316 destroy_conntrack(struct nf_conntrack *nfct)
318 struct nf_conn *ct = (struct nf_conn *)nfct;
319 struct nf_conn_help *help = nfct_help(ct);
320 struct nf_conntrack_l3proto *l3proto;
321 struct nf_conntrack_l4proto *l4proto;
322 typeof(nf_conntrack_destroyed) destroyed;
324 DEBUGP("destroy_conntrack(%p)\n", ct);
325 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
326 NF_CT_ASSERT(!timer_pending(&ct->timeout));
328 nf_conntrack_event(IPCT_DESTROY, ct);
329 set_bit(IPS_DYING_BIT, &ct->status);
331 if (help && help->helper && help->helper->destroy)
332 help->helper->destroy(ct);
334 /* To make sure we don't get any weird locking issues here:
335 * destroy_conntrack() MUST NOT be called with a write lock
336 * to nf_conntrack_lock!!! -HW */
338 l3proto = __nf_ct_l3proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num);
339 if (l3proto && l3proto->destroy)
340 l3proto->destroy(ct);
342 l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
343 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
344 if (l4proto && l4proto->destroy)
345 l4proto->destroy(ct);
347 destroyed = rcu_dereference(nf_conntrack_destroyed);
353 write_lock_bh(&nf_conntrack_lock);
354 /* Expectations will have been removed in clean_from_lists,
355 * except TFTP can create an expectation on the first packet,
356 * before connection is in the list, so we need to clean here,
358 nf_ct_remove_expectations(ct);
360 /* We overload first tuple to link into unconfirmed list. */
361 if (!nf_ct_is_confirmed(ct)) {
362 BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list));
363 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
366 NF_CT_STAT_INC(delete);
367 write_unlock_bh(&nf_conntrack_lock);
370 nf_ct_put(ct->master);
372 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct);
373 nf_conntrack_free(ct);
376 static void death_by_timeout(unsigned long ul_conntrack)
378 struct nf_conn *ct = (void *)ul_conntrack;
380 write_lock_bh(&nf_conntrack_lock);
381 /* Inside lock so preempt is disabled on module removal path.
382 * Otherwise we can get spurious warnings. */
383 NF_CT_STAT_INC(delete_list);
384 clean_from_lists(ct);
385 write_unlock_bh(&nf_conntrack_lock);
389 struct nf_conntrack_tuple_hash *
390 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
391 const struct nf_conn *ignored_conntrack)
393 struct nf_conntrack_tuple_hash *h;
394 unsigned int hash = hash_conntrack(tuple);
396 list_for_each_entry(h, &nf_conntrack_hash[hash], list) {
397 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
398 nf_ct_tuple_equal(tuple, &h->tuple)) {
399 NF_CT_STAT_INC(found);
402 NF_CT_STAT_INC(searched);
407 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
409 /* Find a connection corresponding to a tuple. */
410 struct nf_conntrack_tuple_hash *
411 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple,
412 const struct nf_conn *ignored_conntrack)
414 struct nf_conntrack_tuple_hash *h;
416 read_lock_bh(&nf_conntrack_lock);
417 h = __nf_conntrack_find(tuple, ignored_conntrack);
419 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
420 read_unlock_bh(&nf_conntrack_lock);
424 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
426 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
428 unsigned int repl_hash)
430 ct->id = ++nf_conntrack_next_id;
431 list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list,
432 &nf_conntrack_hash[hash]);
433 list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list,
434 &nf_conntrack_hash[repl_hash]);
437 void nf_conntrack_hash_insert(struct nf_conn *ct)
439 unsigned int hash, repl_hash;
441 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
442 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
444 write_lock_bh(&nf_conntrack_lock);
445 __nf_conntrack_hash_insert(ct, hash, repl_hash);
446 write_unlock_bh(&nf_conntrack_lock);
448 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
450 /* Confirm a connection given skb; places it in hash table */
452 __nf_conntrack_confirm(struct sk_buff **pskb)
454 unsigned int hash, repl_hash;
455 struct nf_conntrack_tuple_hash *h;
457 struct nf_conn_help *help;
458 enum ip_conntrack_info ctinfo;
460 ct = nf_ct_get(*pskb, &ctinfo);
462 /* ipt_REJECT uses nf_conntrack_attach to attach related
463 ICMP/TCP RST packets in other direction. Actual packet
464 which created connection will be IP_CT_NEW or for an
465 expected connection, IP_CT_RELATED. */
466 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
469 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
470 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
472 /* We're not in hash table, and we refuse to set up related
473 connections for unconfirmed conns. But packet copies and
474 REJECT will give spurious warnings here. */
475 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
477 /* No external references means noone else could have
479 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
480 DEBUGP("Confirming conntrack %p\n", ct);
482 write_lock_bh(&nf_conntrack_lock);
484 /* See if there's one in the list already, including reverse:
485 NAT could have grabbed it without realizing, since we're
486 not in the hash. If there is, we lost race. */
487 list_for_each_entry(h, &nf_conntrack_hash[hash], list)
488 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
491 list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list)
492 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
496 /* Remove from unconfirmed list */
497 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
499 __nf_conntrack_hash_insert(ct, hash, repl_hash);
500 /* Timer relative to confirmation time, not original
501 setting time, otherwise we'd get timer wrap in
502 weird delay cases. */
503 ct->timeout.expires += jiffies;
504 add_timer(&ct->timeout);
505 atomic_inc(&ct->ct_general.use);
506 set_bit(IPS_CONFIRMED_BIT, &ct->status);
507 NF_CT_STAT_INC(insert);
508 write_unlock_bh(&nf_conntrack_lock);
509 help = nfct_help(ct);
510 if (help && help->helper)
511 nf_conntrack_event_cache(IPCT_HELPER, *pskb);
512 #ifdef CONFIG_NF_NAT_NEEDED
513 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
514 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
515 nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
517 nf_conntrack_event_cache(master_ct(ct) ?
518 IPCT_RELATED : IPCT_NEW, *pskb);
522 NF_CT_STAT_INC(insert_failed);
523 write_unlock_bh(&nf_conntrack_lock);
526 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
528 /* Returns true if a connection correspondings to the tuple (required
531 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
532 const struct nf_conn *ignored_conntrack)
534 struct nf_conntrack_tuple_hash *h;
536 read_lock_bh(&nf_conntrack_lock);
537 h = __nf_conntrack_find(tuple, ignored_conntrack);
538 read_unlock_bh(&nf_conntrack_lock);
542 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
544 /* There's a small race here where we may free a just-assured
545 connection. Too bad: we're in trouble anyway. */
546 static int early_drop(struct list_head *chain)
548 /* Traverse backwards: gives us oldest, which is roughly LRU */
549 struct nf_conntrack_tuple_hash *h;
550 struct nf_conn *ct = NULL, *tmp;
553 read_lock_bh(&nf_conntrack_lock);
554 list_for_each_entry_reverse(h, chain, list) {
555 tmp = nf_ct_tuplehash_to_ctrack(h);
556 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) {
558 atomic_inc(&ct->ct_general.use);
562 read_unlock_bh(&nf_conntrack_lock);
567 if (del_timer(&ct->timeout)) {
568 death_by_timeout((unsigned long)ct);
570 NF_CT_STAT_INC_ATOMIC(early_drop);
576 static struct nf_conn *
577 __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
578 const struct nf_conntrack_tuple *repl,
579 const struct nf_conntrack_l3proto *l3proto,
582 struct nf_conn *conntrack = NULL;
583 struct nf_conntrack_helper *helper;
585 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
586 get_random_bytes(&nf_conntrack_hash_rnd, 4);
587 nf_conntrack_hash_rnd_initted = 1;
590 /* We don't want any race condition at early drop stage */
591 atomic_inc(&nf_conntrack_count);
594 && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
595 unsigned int hash = hash_conntrack(orig);
596 /* Try dropping from this hash chain. */
597 if (!early_drop(&nf_conntrack_hash[hash])) {
598 atomic_dec(&nf_conntrack_count);
601 "nf_conntrack: table full, dropping"
603 return ERR_PTR(-ENOMEM);
607 /* find features needed by this conntrack. */
608 features |= l3proto->get_features(orig);
610 /* FIXME: protect helper list per RCU */
611 read_lock_bh(&nf_conntrack_lock);
612 helper = __nf_ct_helper_find(repl);
613 /* NAT might want to assign a helper later */
614 if (helper || features & NF_CT_F_NAT)
615 features |= NF_CT_F_HELP;
616 read_unlock_bh(&nf_conntrack_lock);
618 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features);
620 read_lock_bh(&nf_ct_cache_lock);
622 if (unlikely(!nf_ct_cache[features].use)) {
623 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n",
628 conntrack = kmem_cache_alloc(nf_ct_cache[features].cachep, GFP_ATOMIC);
629 if (conntrack == NULL) {
630 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n");
634 memset(conntrack, 0, nf_ct_cache[features].size);
635 conntrack->features = features;
636 atomic_set(&conntrack->ct_general.use, 1);
637 conntrack->ct_general.destroy = destroy_conntrack;
638 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
639 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
640 /* Don't set timer yet: wait for confirmation */
641 init_timer(&conntrack->timeout);
642 conntrack->timeout.data = (unsigned long)conntrack;
643 conntrack->timeout.function = death_by_timeout;
644 read_unlock_bh(&nf_ct_cache_lock);
648 read_unlock_bh(&nf_ct_cache_lock);
649 atomic_dec(&nf_conntrack_count);
653 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
654 const struct nf_conntrack_tuple *repl)
656 struct nf_conntrack_l3proto *l3proto;
660 l3proto = __nf_ct_l3proto_find(orig->src.l3num);
661 ct = __nf_conntrack_alloc(orig, repl, l3proto, 0);
666 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
668 void nf_conntrack_free(struct nf_conn *conntrack)
670 u_int32_t features = conntrack->features;
671 NF_CT_ASSERT(features >= NF_CT_F_BASIC && features < NF_CT_F_NUM);
672 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features,
674 kmem_cache_free(nf_ct_cache[features].cachep, conntrack);
675 atomic_dec(&nf_conntrack_count);
677 EXPORT_SYMBOL_GPL(nf_conntrack_free);
679 /* Allocate a new conntrack: we return -ENOMEM if classification
680 failed due to stress. Otherwise it really is unclassifiable. */
681 static struct nf_conntrack_tuple_hash *
682 init_conntrack(const struct nf_conntrack_tuple *tuple,
683 struct nf_conntrack_l3proto *l3proto,
684 struct nf_conntrack_l4proto *l4proto,
686 unsigned int dataoff)
688 struct nf_conn *conntrack;
689 struct nf_conntrack_tuple repl_tuple;
690 struct nf_conntrack_expect *exp;
691 u_int32_t features = 0;
693 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
694 DEBUGP("Can't invert tuple.\n");
698 read_lock_bh(&nf_conntrack_lock);
699 exp = __nf_conntrack_expect_find(tuple);
700 if (exp && exp->helper)
701 features = NF_CT_F_HELP;
702 read_unlock_bh(&nf_conntrack_lock);
704 conntrack = __nf_conntrack_alloc(tuple, &repl_tuple, l3proto, features);
705 if (conntrack == NULL || IS_ERR(conntrack)) {
706 DEBUGP("Can't allocate conntrack.\n");
707 return (struct nf_conntrack_tuple_hash *)conntrack;
710 if (!l4proto->new(conntrack, skb, dataoff)) {
711 nf_conntrack_free(conntrack);
712 DEBUGP("init conntrack: can't track with proto module\n");
716 write_lock_bh(&nf_conntrack_lock);
717 exp = find_expectation(tuple);
720 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
722 /* Welcome, Mr. Bond. We've been expecting you... */
723 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
724 conntrack->master = exp->master;
726 nfct_help(conntrack)->helper = exp->helper;
727 #ifdef CONFIG_NF_CONNTRACK_MARK
728 conntrack->mark = exp->master->mark;
730 #ifdef CONFIG_NF_CONNTRACK_SECMARK
731 conntrack->secmark = exp->master->secmark;
733 nf_conntrack_get(&conntrack->master->ct_general);
734 NF_CT_STAT_INC(expect_new);
736 struct nf_conn_help *help = nfct_help(conntrack);
739 help->helper = __nf_ct_helper_find(&repl_tuple);
743 /* Overload tuple linked list to put us in unconfirmed list. */
744 list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed);
746 write_unlock_bh(&nf_conntrack_lock);
750 exp->expectfn(conntrack, exp);
751 nf_conntrack_expect_put(exp);
754 return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
757 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
758 static inline struct nf_conn *
759 resolve_normal_ct(struct sk_buff *skb,
760 unsigned int dataoff,
763 struct nf_conntrack_l3proto *l3proto,
764 struct nf_conntrack_l4proto *l4proto,
766 enum ip_conntrack_info *ctinfo)
768 struct nf_conntrack_tuple tuple;
769 struct nf_conntrack_tuple_hash *h;
772 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
773 dataoff, l3num, protonum, &tuple, l3proto,
775 DEBUGP("resolve_normal_ct: Can't get tuple\n");
779 /* look for tuple match */
780 h = nf_conntrack_find_get(&tuple, NULL);
782 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
788 ct = nf_ct_tuplehash_to_ctrack(h);
790 /* It exists; we have (non-exclusive) reference. */
791 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
792 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
793 /* Please set reply bit if this packet OK */
796 /* Once we've had two way comms, always ESTABLISHED. */
797 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
798 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct);
799 *ctinfo = IP_CT_ESTABLISHED;
800 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
801 DEBUGP("nf_conntrack_in: related packet for %p\n", ct);
802 *ctinfo = IP_CT_RELATED;
804 DEBUGP("nf_conntrack_in: new packet for %p\n", ct);
809 skb->nfct = &ct->ct_general;
810 skb->nfctinfo = *ctinfo;
815 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
818 enum ip_conntrack_info ctinfo;
819 struct nf_conntrack_l3proto *l3proto;
820 struct nf_conntrack_l4proto *l4proto;
821 unsigned int dataoff;
826 /* Previously seen (loopback or untracked)? Ignore. */
828 NF_CT_STAT_INC_ATOMIC(ignore);
832 /* rcu_read_lock()ed by nf_hook_slow */
833 l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
835 if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) {
836 DEBUGP("not prepared to track yet or error occured\n");
840 l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
842 /* It may be an special packet, error, unclean...
843 * inverse of the return code tells to the netfilter
844 * core what to do with the packet. */
845 if (l4proto->error != NULL &&
846 (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
847 NF_CT_STAT_INC_ATOMIC(error);
848 NF_CT_STAT_INC_ATOMIC(invalid);
852 ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto,
853 &set_reply, &ctinfo);
855 /* Not valid part of a connection */
856 NF_CT_STAT_INC_ATOMIC(invalid);
861 /* Too stressed to deal. */
862 NF_CT_STAT_INC_ATOMIC(drop);
866 NF_CT_ASSERT((*pskb)->nfct);
868 ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum);
870 /* Invalid: inverse of the return code tells
871 * the netfilter core what to do */
872 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
873 nf_conntrack_put((*pskb)->nfct);
874 (*pskb)->nfct = NULL;
875 NF_CT_STAT_INC_ATOMIC(invalid);
879 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
880 nf_conntrack_event_cache(IPCT_STATUS, *pskb);
884 EXPORT_SYMBOL_GPL(nf_conntrack_in);
886 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
887 const struct nf_conntrack_tuple *orig)
892 ret = nf_ct_invert_tuple(inverse, orig,
893 __nf_ct_l3proto_find(orig->src.l3num),
894 __nf_ct_l4proto_find(orig->src.l3num,
895 orig->dst.protonum));
899 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
901 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
902 implicitly racy: see __nf_conntrack_confirm */
903 void nf_conntrack_alter_reply(struct nf_conn *ct,
904 const struct nf_conntrack_tuple *newreply)
906 struct nf_conn_help *help = nfct_help(ct);
908 write_lock_bh(&nf_conntrack_lock);
909 /* Should be unconfirmed, so not in hash table yet */
910 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
912 DEBUGP("Altering reply tuple of %p to ", ct);
913 NF_CT_DUMP_TUPLE(newreply);
915 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
916 if (!ct->master && help && help->expecting == 0)
917 help->helper = __nf_ct_helper_find(newreply);
918 write_unlock_bh(&nf_conntrack_lock);
920 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
922 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
923 void __nf_ct_refresh_acct(struct nf_conn *ct,
924 enum ip_conntrack_info ctinfo,
925 const struct sk_buff *skb,
926 unsigned long extra_jiffies,
931 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
934 write_lock_bh(&nf_conntrack_lock);
936 /* Only update if this is not a fixed timeout */
937 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
938 write_unlock_bh(&nf_conntrack_lock);
942 /* If not in hash table, timer will not be active yet */
943 if (!nf_ct_is_confirmed(ct)) {
944 ct->timeout.expires = extra_jiffies;
945 event = IPCT_REFRESH;
947 unsigned long newtime = jiffies + extra_jiffies;
949 /* Only update the timeout if the new timeout is at least
950 HZ jiffies from the old timeout. Need del_timer for race
951 avoidance (may already be dying). */
952 if (newtime - ct->timeout.expires >= HZ
953 && del_timer(&ct->timeout)) {
954 ct->timeout.expires = newtime;
955 add_timer(&ct->timeout);
956 event = IPCT_REFRESH;
960 #ifdef CONFIG_NF_CT_ACCT
962 ct->counters[CTINFO2DIR(ctinfo)].packets++;
963 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
964 skb->len - skb_network_offset(skb);
966 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
967 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
968 event |= IPCT_COUNTER_FILLING;
972 write_unlock_bh(&nf_conntrack_lock);
974 /* must be unlocked when calling event cache */
976 nf_conntrack_event_cache(event, skb);
978 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
980 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
982 #include <linux/netfilter/nfnetlink.h>
983 #include <linux/netfilter/nfnetlink_conntrack.h>
984 #include <linux/mutex.h>
987 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
988 * in ip_conntrack_core, since we don't want the protocols to autoload
989 * or depend on ctnetlink */
990 int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
991 const struct nf_conntrack_tuple *tuple)
993 NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
994 &tuple->src.u.tcp.port);
995 NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
996 &tuple->dst.u.tcp.port);
1002 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr);
1004 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
1005 [CTA_PROTO_SRC_PORT-1] = sizeof(u_int16_t),
1006 [CTA_PROTO_DST_PORT-1] = sizeof(u_int16_t)
1009 int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[],
1010 struct nf_conntrack_tuple *t)
1012 if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1])
1015 if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
1018 t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]);
1019 t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]);
1023 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple);
1026 /* Used by ipt_REJECT and ip6t_REJECT. */
1027 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1030 enum ip_conntrack_info ctinfo;
1032 /* This ICMP is in reverse direction to the packet which caused it */
1033 ct = nf_ct_get(skb, &ctinfo);
1034 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1035 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
1037 ctinfo = IP_CT_RELATED;
1039 /* Attach to new skbuff, and increment count */
1040 nskb->nfct = &ct->ct_general;
1041 nskb->nfctinfo = ctinfo;
1042 nf_conntrack_get(nskb->nfct);
1044 EXPORT_SYMBOL_GPL(__nf_conntrack_attach);
1047 do_iter(const struct nf_conntrack_tuple_hash *i,
1048 int (*iter)(struct nf_conn *i, void *data),
1051 return iter(nf_ct_tuplehash_to_ctrack(i), data);
1054 /* Bring out ya dead! */
1055 static struct nf_conn *
1056 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
1057 void *data, unsigned int *bucket)
1059 struct nf_conntrack_tuple_hash *h;
1062 write_lock_bh(&nf_conntrack_lock);
1063 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1064 list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) {
1065 ct = nf_ct_tuplehash_to_ctrack(h);
1070 list_for_each_entry(h, &unconfirmed, list) {
1071 ct = nf_ct_tuplehash_to_ctrack(h);
1073 set_bit(IPS_DYING_BIT, &ct->status);
1075 write_unlock_bh(&nf_conntrack_lock);
1078 atomic_inc(&ct->ct_general.use);
1079 write_unlock_bh(&nf_conntrack_lock);
1084 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
1087 unsigned int bucket = 0;
1089 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
1090 /* Time to push up daises... */
1091 if (del_timer(&ct->timeout))
1092 death_by_timeout((unsigned long)ct);
1093 /* ... else the timer will get him soon. */
1098 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1100 static int kill_all(struct nf_conn *i, void *data)
1105 static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size)
1110 free_pages((unsigned long)hash,
1111 get_order(sizeof(struct list_head) * size));
1114 void nf_conntrack_flush(void)
1116 nf_ct_iterate_cleanup(kill_all, NULL);
1118 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1120 /* Mishearing the voices in his head, our hero wonders how he's
1121 supposed to kill the mall. */
1122 void nf_conntrack_cleanup(void)
1126 rcu_assign_pointer(ip_ct_attach, NULL);
1128 /* This makes sure all current packets have passed through
1129 netfilter framework. Roll on, two-stage module
1133 nf_ct_event_cache_flush();
1135 nf_conntrack_flush();
1136 if (atomic_read(&nf_conntrack_count) != 0) {
1138 goto i_see_dead_people;
1140 /* wait until all references to nf_conntrack_untracked are dropped */
1141 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1144 for (i = 0; i < NF_CT_F_NUM; i++) {
1145 if (nf_ct_cache[i].use == 0)
1148 NF_CT_ASSERT(nf_ct_cache[i].use == 1);
1149 nf_ct_cache[i].use = 1;
1150 nf_conntrack_unregister_cache(i);
1152 kmem_cache_destroy(nf_conntrack_expect_cachep);
1153 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1154 nf_conntrack_htable_size);
1156 nf_conntrack_proto_fini();
1159 static struct list_head *alloc_hashtable(int size, int *vmalloced)
1161 struct list_head *hash;
1165 hash = (void*)__get_free_pages(GFP_KERNEL,
1166 get_order(sizeof(struct list_head)
1170 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1171 hash = vmalloc(sizeof(struct list_head) * size);
1175 for (i = 0; i < size; i++)
1176 INIT_LIST_HEAD(&hash[i]);
1181 int set_hashsize(const char *val, struct kernel_param *kp)
1183 int i, bucket, hashsize, vmalloced;
1184 int old_vmalloced, old_size;
1186 struct list_head *hash, *old_hash;
1187 struct nf_conntrack_tuple_hash *h;
1189 /* On boot, we can set this without any fancy locking. */
1190 if (!nf_conntrack_htable_size)
1191 return param_set_uint(val, kp);
1193 hashsize = simple_strtol(val, NULL, 0);
1197 hash = alloc_hashtable(hashsize, &vmalloced);
1201 /* We have to rehahs for the new table anyway, so we also can
1202 * use a newrandom seed */
1203 get_random_bytes(&rnd, 4);
1205 write_lock_bh(&nf_conntrack_lock);
1206 for (i = 0; i < nf_conntrack_htable_size; i++) {
1207 while (!list_empty(&nf_conntrack_hash[i])) {
1208 h = list_entry(nf_conntrack_hash[i].next,
1209 struct nf_conntrack_tuple_hash, list);
1211 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1212 list_add_tail(&h->list, &hash[bucket]);
1215 old_size = nf_conntrack_htable_size;
1216 old_vmalloced = nf_conntrack_vmalloc;
1217 old_hash = nf_conntrack_hash;
1219 nf_conntrack_htable_size = hashsize;
1220 nf_conntrack_vmalloc = vmalloced;
1221 nf_conntrack_hash = hash;
1222 nf_conntrack_hash_rnd = rnd;
1223 write_unlock_bh(&nf_conntrack_lock);
1225 free_conntrack_hash(old_hash, old_vmalloced, old_size);
1229 module_param_call(hashsize, set_hashsize, param_get_uint,
1230 &nf_conntrack_htable_size, 0600);
1232 int __init nf_conntrack_init(void)
1236 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1237 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */
1238 if (!nf_conntrack_htable_size) {
1239 nf_conntrack_htable_size
1240 = (((num_physpages << PAGE_SHIFT) / 16384)
1241 / sizeof(struct list_head));
1242 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1243 nf_conntrack_htable_size = 8192;
1244 if (nf_conntrack_htable_size < 16)
1245 nf_conntrack_htable_size = 16;
1247 nf_conntrack_max = 8 * nf_conntrack_htable_size;
1249 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1250 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1253 nf_conntrack_hash = alloc_hashtable(nf_conntrack_htable_size,
1254 &nf_conntrack_vmalloc);
1255 if (!nf_conntrack_hash) {
1256 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1260 ret = nf_conntrack_register_cache(NF_CT_F_BASIC, "nf_conntrack:basic",
1261 sizeof(struct nf_conn));
1263 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1267 nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect",
1268 sizeof(struct nf_conntrack_expect),
1270 if (!nf_conntrack_expect_cachep) {
1271 printk(KERN_ERR "Unable to create nf_expect slab cache\n");
1272 goto err_free_conntrack_slab;
1275 ret = nf_conntrack_proto_init();
1277 goto out_free_expect_slab;
1279 /* For use by REJECT target */
1280 rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
1282 /* Set up fake conntrack:
1283 - to never be deleted, not in any hashes */
1284 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1285 /* - and look it like as a confirmed connection */
1286 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1290 out_free_expect_slab:
1291 kmem_cache_destroy(nf_conntrack_expect_cachep);
1292 err_free_conntrack_slab:
1293 nf_conntrack_unregister_cache(NF_CT_F_BASIC);
1295 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1296 nf_conntrack_htable_size);