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)
131 a = jhash((void *)tuple->src.u3.all, sizeof(tuple->src.u3.all),
132 ((tuple->src.l3num) << 16) | tuple->dst.protonum);
133 b = jhash((void *)tuple->dst.u3.all, sizeof(tuple->dst.u3.all),
134 (tuple->src.u.all << 16) | tuple->dst.u.all);
136 return jhash_2words(a, b, rnd) % size;
139 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
141 return __hash_conntrack(tuple, nf_conntrack_htable_size,
142 nf_conntrack_hash_rnd);
145 int nf_conntrack_register_cache(u_int32_t features, const char *name,
150 struct kmem_cache *cachep;
152 DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n",
153 features, name, size);
155 if (features < NF_CT_F_BASIC || features >= NF_CT_F_NUM) {
156 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n",
161 mutex_lock(&nf_ct_cache_mutex);
163 write_lock_bh(&nf_ct_cache_lock);
164 /* e.g: multiple helpers are loaded */
165 if (nf_ct_cache[features].use > 0) {
166 DEBUGP("nf_conntrack_register_cache: already resisterd.\n");
167 if ((!strncmp(nf_ct_cache[features].name, name,
168 NF_CT_FEATURES_NAMELEN))
169 && nf_ct_cache[features].size == size) {
170 DEBUGP("nf_conntrack_register_cache: reusing.\n");
171 nf_ct_cache[features].use++;
176 write_unlock_bh(&nf_ct_cache_lock);
177 mutex_unlock(&nf_ct_cache_mutex);
180 write_unlock_bh(&nf_ct_cache_lock);
183 * The memory space for name of slab cache must be alive until
184 * cache is destroyed.
186 cache_name = kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN, GFP_ATOMIC);
187 if (cache_name == NULL) {
188 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n");
193 if (strlcpy(cache_name, name, NF_CT_FEATURES_NAMELEN)
194 >= NF_CT_FEATURES_NAMELEN) {
195 printk("nf_conntrack_register_cache: name too long\n");
200 cachep = kmem_cache_create(cache_name, size, 0, 0,
203 printk("nf_conntrack_register_cache: Can't create slab cache "
204 "for the features = 0x%x\n", features);
209 write_lock_bh(&nf_ct_cache_lock);
210 nf_ct_cache[features].use = 1;
211 nf_ct_cache[features].size = size;
212 nf_ct_cache[features].cachep = cachep;
213 nf_ct_cache[features].name = cache_name;
214 write_unlock_bh(&nf_ct_cache_lock);
221 mutex_unlock(&nf_ct_cache_mutex);
224 EXPORT_SYMBOL_GPL(nf_conntrack_register_cache);
226 /* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */
227 void nf_conntrack_unregister_cache(u_int32_t features)
229 struct kmem_cache *cachep;
233 * This assures that kmem_cache_create() isn't called before destroying
236 DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features);
237 mutex_lock(&nf_ct_cache_mutex);
239 write_lock_bh(&nf_ct_cache_lock);
240 if (--nf_ct_cache[features].use > 0) {
241 write_unlock_bh(&nf_ct_cache_lock);
242 mutex_unlock(&nf_ct_cache_mutex);
245 cachep = nf_ct_cache[features].cachep;
246 name = nf_ct_cache[features].name;
247 nf_ct_cache[features].cachep = NULL;
248 nf_ct_cache[features].name = NULL;
249 nf_ct_cache[features].size = 0;
250 write_unlock_bh(&nf_ct_cache_lock);
254 kmem_cache_destroy(cachep);
257 mutex_unlock(&nf_ct_cache_mutex);
259 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_cache);
262 nf_ct_get_tuple(const struct sk_buff *skb,
264 unsigned int dataoff,
267 struct nf_conntrack_tuple *tuple,
268 const struct nf_conntrack_l3proto *l3proto,
269 const struct nf_conntrack_l4proto *l4proto)
271 NF_CT_TUPLE_U_BLANK(tuple);
273 tuple->src.l3num = l3num;
274 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
277 tuple->dst.protonum = protonum;
278 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
280 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
282 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
285 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
286 const struct nf_conntrack_tuple *orig,
287 const struct nf_conntrack_l3proto *l3proto,
288 const struct nf_conntrack_l4proto *l4proto)
290 NF_CT_TUPLE_U_BLANK(inverse);
292 inverse->src.l3num = orig->src.l3num;
293 if (l3proto->invert_tuple(inverse, orig) == 0)
296 inverse->dst.dir = !orig->dst.dir;
298 inverse->dst.protonum = orig->dst.protonum;
299 return l4proto->invert_tuple(inverse, orig);
301 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
304 clean_from_lists(struct nf_conn *ct)
306 DEBUGP("clean_from_lists(%p)\n", ct);
307 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
308 list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list);
310 /* Destroy all pending expectations */
311 nf_ct_remove_expectations(ct);
315 destroy_conntrack(struct nf_conntrack *nfct)
317 struct nf_conn *ct = (struct nf_conn *)nfct;
318 struct nf_conn_help *help = nfct_help(ct);
319 struct nf_conntrack_l3proto *l3proto;
320 struct nf_conntrack_l4proto *l4proto;
322 DEBUGP("destroy_conntrack(%p)\n", ct);
323 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
324 NF_CT_ASSERT(!timer_pending(&ct->timeout));
326 nf_conntrack_event(IPCT_DESTROY, ct);
327 set_bit(IPS_DYING_BIT, &ct->status);
329 if (help && help->helper && help->helper->destroy)
330 help->helper->destroy(ct);
332 /* To make sure we don't get any weird locking issues here:
333 * destroy_conntrack() MUST NOT be called with a write lock
334 * to nf_conntrack_lock!!! -HW */
336 l3proto = __nf_ct_l3proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num);
337 if (l3proto && l3proto->destroy)
338 l3proto->destroy(ct);
340 l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
341 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
342 if (l4proto && l4proto->destroy)
343 l4proto->destroy(ct);
346 if (nf_conntrack_destroyed)
347 nf_conntrack_destroyed(ct);
349 write_lock_bh(&nf_conntrack_lock);
350 /* Expectations will have been removed in clean_from_lists,
351 * except TFTP can create an expectation on the first packet,
352 * before connection is in the list, so we need to clean here,
354 nf_ct_remove_expectations(ct);
356 /* We overload first tuple to link into unconfirmed list. */
357 if (!nf_ct_is_confirmed(ct)) {
358 BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list));
359 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
362 NF_CT_STAT_INC(delete);
363 write_unlock_bh(&nf_conntrack_lock);
366 nf_ct_put(ct->master);
368 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct);
369 nf_conntrack_free(ct);
372 static void death_by_timeout(unsigned long ul_conntrack)
374 struct nf_conn *ct = (void *)ul_conntrack;
376 write_lock_bh(&nf_conntrack_lock);
377 /* Inside lock so preempt is disabled on module removal path.
378 * Otherwise we can get spurious warnings. */
379 NF_CT_STAT_INC(delete_list);
380 clean_from_lists(ct);
381 write_unlock_bh(&nf_conntrack_lock);
385 struct nf_conntrack_tuple_hash *
386 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
387 const struct nf_conn *ignored_conntrack)
389 struct nf_conntrack_tuple_hash *h;
390 unsigned int hash = hash_conntrack(tuple);
392 list_for_each_entry(h, &nf_conntrack_hash[hash], list) {
393 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
394 nf_ct_tuple_equal(tuple, &h->tuple)) {
395 NF_CT_STAT_INC(found);
398 NF_CT_STAT_INC(searched);
403 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
405 /* Find a connection corresponding to a tuple. */
406 struct nf_conntrack_tuple_hash *
407 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple,
408 const struct nf_conn *ignored_conntrack)
410 struct nf_conntrack_tuple_hash *h;
412 read_lock_bh(&nf_conntrack_lock);
413 h = __nf_conntrack_find(tuple, ignored_conntrack);
415 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
416 read_unlock_bh(&nf_conntrack_lock);
420 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
422 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
424 unsigned int repl_hash)
426 ct->id = ++nf_conntrack_next_id;
427 list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list,
428 &nf_conntrack_hash[hash]);
429 list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list,
430 &nf_conntrack_hash[repl_hash]);
433 void nf_conntrack_hash_insert(struct nf_conn *ct)
435 unsigned int hash, repl_hash;
437 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
438 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
440 write_lock_bh(&nf_conntrack_lock);
441 __nf_conntrack_hash_insert(ct, hash, repl_hash);
442 write_unlock_bh(&nf_conntrack_lock);
444 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
446 /* Confirm a connection given skb; places it in hash table */
448 __nf_conntrack_confirm(struct sk_buff **pskb)
450 unsigned int hash, repl_hash;
451 struct nf_conntrack_tuple_hash *h;
453 struct nf_conn_help *help;
454 enum ip_conntrack_info ctinfo;
456 ct = nf_ct_get(*pskb, &ctinfo);
458 /* ipt_REJECT uses nf_conntrack_attach to attach related
459 ICMP/TCP RST packets in other direction. Actual packet
460 which created connection will be IP_CT_NEW or for an
461 expected connection, IP_CT_RELATED. */
462 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
465 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
466 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
468 /* We're not in hash table, and we refuse to set up related
469 connections for unconfirmed conns. But packet copies and
470 REJECT will give spurious warnings here. */
471 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
473 /* No external references means noone else could have
475 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
476 DEBUGP("Confirming conntrack %p\n", ct);
478 write_lock_bh(&nf_conntrack_lock);
480 /* See if there's one in the list already, including reverse:
481 NAT could have grabbed it without realizing, since we're
482 not in the hash. If there is, we lost race. */
483 list_for_each_entry(h, &nf_conntrack_hash[hash], list)
484 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
487 list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list)
488 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
492 /* Remove from unconfirmed list */
493 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
495 __nf_conntrack_hash_insert(ct, hash, repl_hash);
496 /* Timer relative to confirmation time, not original
497 setting time, otherwise we'd get timer wrap in
498 weird delay cases. */
499 ct->timeout.expires += jiffies;
500 add_timer(&ct->timeout);
501 atomic_inc(&ct->ct_general.use);
502 set_bit(IPS_CONFIRMED_BIT, &ct->status);
503 NF_CT_STAT_INC(insert);
504 write_unlock_bh(&nf_conntrack_lock);
505 help = nfct_help(ct);
506 if (help && help->helper)
507 nf_conntrack_event_cache(IPCT_HELPER, *pskb);
508 #ifdef CONFIG_NF_NAT_NEEDED
509 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
510 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
511 nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
513 nf_conntrack_event_cache(master_ct(ct) ?
514 IPCT_RELATED : IPCT_NEW, *pskb);
518 NF_CT_STAT_INC(insert_failed);
519 write_unlock_bh(&nf_conntrack_lock);
522 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
524 /* Returns true if a connection correspondings to the tuple (required
527 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
528 const struct nf_conn *ignored_conntrack)
530 struct nf_conntrack_tuple_hash *h;
532 read_lock_bh(&nf_conntrack_lock);
533 h = __nf_conntrack_find(tuple, ignored_conntrack);
534 read_unlock_bh(&nf_conntrack_lock);
538 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
540 /* There's a small race here where we may free a just-assured
541 connection. Too bad: we're in trouble anyway. */
542 static int early_drop(struct list_head *chain)
544 /* Traverse backwards: gives us oldest, which is roughly LRU */
545 struct nf_conntrack_tuple_hash *h;
546 struct nf_conn *ct = NULL, *tmp;
549 read_lock_bh(&nf_conntrack_lock);
550 list_for_each_entry_reverse(h, chain, list) {
551 tmp = nf_ct_tuplehash_to_ctrack(h);
552 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) {
554 atomic_inc(&ct->ct_general.use);
558 read_unlock_bh(&nf_conntrack_lock);
563 if (del_timer(&ct->timeout)) {
564 death_by_timeout((unsigned long)ct);
566 NF_CT_STAT_INC(early_drop);
572 static struct nf_conn *
573 __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
574 const struct nf_conntrack_tuple *repl,
575 const struct nf_conntrack_l3proto *l3proto,
578 struct nf_conn *conntrack = NULL;
579 struct nf_conntrack_helper *helper;
581 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
582 get_random_bytes(&nf_conntrack_hash_rnd, 4);
583 nf_conntrack_hash_rnd_initted = 1;
586 /* We don't want any race condition at early drop stage */
587 atomic_inc(&nf_conntrack_count);
590 && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
591 unsigned int hash = hash_conntrack(orig);
592 /* Try dropping from this hash chain. */
593 if (!early_drop(&nf_conntrack_hash[hash])) {
594 atomic_dec(&nf_conntrack_count);
597 "nf_conntrack: table full, dropping"
599 return ERR_PTR(-ENOMEM);
603 /* find features needed by this conntrack. */
604 features |= l3proto->get_features(orig);
606 /* FIXME: protect helper list per RCU */
607 read_lock_bh(&nf_conntrack_lock);
608 helper = __nf_ct_helper_find(repl);
609 /* NAT might want to assign a helper later */
610 if (helper || features & NF_CT_F_NAT)
611 features |= NF_CT_F_HELP;
612 read_unlock_bh(&nf_conntrack_lock);
614 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features);
616 read_lock_bh(&nf_ct_cache_lock);
618 if (unlikely(!nf_ct_cache[features].use)) {
619 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n",
624 conntrack = kmem_cache_alloc(nf_ct_cache[features].cachep, GFP_ATOMIC);
625 if (conntrack == NULL) {
626 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n");
630 memset(conntrack, 0, nf_ct_cache[features].size);
631 conntrack->features = features;
632 atomic_set(&conntrack->ct_general.use, 1);
633 conntrack->ct_general.destroy = destroy_conntrack;
634 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
635 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
636 /* Don't set timer yet: wait for confirmation */
637 init_timer(&conntrack->timeout);
638 conntrack->timeout.data = (unsigned long)conntrack;
639 conntrack->timeout.function = death_by_timeout;
640 read_unlock_bh(&nf_ct_cache_lock);
644 read_unlock_bh(&nf_ct_cache_lock);
645 atomic_dec(&nf_conntrack_count);
649 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
650 const struct nf_conntrack_tuple *repl)
652 struct nf_conntrack_l3proto *l3proto;
656 l3proto = __nf_ct_l3proto_find(orig->src.l3num);
657 ct = __nf_conntrack_alloc(orig, repl, l3proto, 0);
662 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
664 void nf_conntrack_free(struct nf_conn *conntrack)
666 u_int32_t features = conntrack->features;
667 NF_CT_ASSERT(features >= NF_CT_F_BASIC && features < NF_CT_F_NUM);
668 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features,
670 kmem_cache_free(nf_ct_cache[features].cachep, conntrack);
671 atomic_dec(&nf_conntrack_count);
673 EXPORT_SYMBOL_GPL(nf_conntrack_free);
675 /* Allocate a new conntrack: we return -ENOMEM if classification
676 failed due to stress. Otherwise it really is unclassifiable. */
677 static struct nf_conntrack_tuple_hash *
678 init_conntrack(const struct nf_conntrack_tuple *tuple,
679 struct nf_conntrack_l3proto *l3proto,
680 struct nf_conntrack_l4proto *l4proto,
682 unsigned int dataoff)
684 struct nf_conn *conntrack;
685 struct nf_conntrack_tuple repl_tuple;
686 struct nf_conntrack_expect *exp;
687 u_int32_t features = 0;
689 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
690 DEBUGP("Can't invert tuple.\n");
694 read_lock_bh(&nf_conntrack_lock);
695 exp = __nf_conntrack_expect_find(tuple);
696 if (exp && exp->helper)
697 features = NF_CT_F_HELP;
698 read_unlock_bh(&nf_conntrack_lock);
700 conntrack = __nf_conntrack_alloc(tuple, &repl_tuple, l3proto, features);
701 if (conntrack == NULL || IS_ERR(conntrack)) {
702 DEBUGP("Can't allocate conntrack.\n");
703 return (struct nf_conntrack_tuple_hash *)conntrack;
706 if (!l4proto->new(conntrack, skb, dataoff)) {
707 nf_conntrack_free(conntrack);
708 DEBUGP("init conntrack: can't track with proto module\n");
712 write_lock_bh(&nf_conntrack_lock);
713 exp = find_expectation(tuple);
716 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
718 /* Welcome, Mr. Bond. We've been expecting you... */
719 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
720 conntrack->master = exp->master;
722 nfct_help(conntrack)->helper = exp->helper;
723 #ifdef CONFIG_NF_CONNTRACK_MARK
724 conntrack->mark = exp->master->mark;
726 #ifdef CONFIG_NF_CONNTRACK_SECMARK
727 conntrack->secmark = exp->master->secmark;
729 nf_conntrack_get(&conntrack->master->ct_general);
730 NF_CT_STAT_INC(expect_new);
732 struct nf_conn_help *help = nfct_help(conntrack);
735 help->helper = __nf_ct_helper_find(&repl_tuple);
739 /* Overload tuple linked list to put us in unconfirmed list. */
740 list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed);
742 write_unlock_bh(&nf_conntrack_lock);
746 exp->expectfn(conntrack, exp);
747 nf_conntrack_expect_put(exp);
750 return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
753 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
754 static inline struct nf_conn *
755 resolve_normal_ct(struct sk_buff *skb,
756 unsigned int dataoff,
759 struct nf_conntrack_l3proto *l3proto,
760 struct nf_conntrack_l4proto *l4proto,
762 enum ip_conntrack_info *ctinfo)
764 struct nf_conntrack_tuple tuple;
765 struct nf_conntrack_tuple_hash *h;
768 if (!nf_ct_get_tuple(skb, (unsigned int)(skb->nh.raw - skb->data),
769 dataoff, l3num, protonum, &tuple, l3proto,
771 DEBUGP("resolve_normal_ct: Can't get tuple\n");
775 /* look for tuple match */
776 h = nf_conntrack_find_get(&tuple, NULL);
778 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
784 ct = nf_ct_tuplehash_to_ctrack(h);
786 /* It exists; we have (non-exclusive) reference. */
787 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
788 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
789 /* Please set reply bit if this packet OK */
792 /* Once we've had two way comms, always ESTABLISHED. */
793 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
794 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct);
795 *ctinfo = IP_CT_ESTABLISHED;
796 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
797 DEBUGP("nf_conntrack_in: related packet for %p\n", ct);
798 *ctinfo = IP_CT_RELATED;
800 DEBUGP("nf_conntrack_in: new packet for %p\n", ct);
805 skb->nfct = &ct->ct_general;
806 skb->nfctinfo = *ctinfo;
811 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
814 enum ip_conntrack_info ctinfo;
815 struct nf_conntrack_l3proto *l3proto;
816 struct nf_conntrack_l4proto *l4proto;
817 unsigned int dataoff;
822 /* Previously seen (loopback or untracked)? Ignore. */
824 NF_CT_STAT_INC(ignore);
828 /* rcu_read_lock()ed by nf_hook_slow */
829 l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
831 if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) {
832 DEBUGP("not prepared to track yet or error occured\n");
836 l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
838 /* It may be an special packet, error, unclean...
839 * inverse of the return code tells to the netfilter
840 * core what to do with the packet. */
841 if (l4proto->error != NULL &&
842 (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
843 NF_CT_STAT_INC(error);
844 NF_CT_STAT_INC(invalid);
848 ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto,
849 &set_reply, &ctinfo);
851 /* Not valid part of a connection */
852 NF_CT_STAT_INC(invalid);
857 /* Too stressed to deal. */
858 NF_CT_STAT_INC(drop);
862 NF_CT_ASSERT((*pskb)->nfct);
864 ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum);
866 /* Invalid: inverse of the return code tells
867 * the netfilter core what to do */
868 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
869 nf_conntrack_put((*pskb)->nfct);
870 (*pskb)->nfct = NULL;
871 NF_CT_STAT_INC(invalid);
875 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
876 nf_conntrack_event_cache(IPCT_STATUS, *pskb);
880 EXPORT_SYMBOL_GPL(nf_conntrack_in);
882 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
883 const struct nf_conntrack_tuple *orig)
888 ret = nf_ct_invert_tuple(inverse, orig,
889 __nf_ct_l3proto_find(orig->src.l3num),
890 __nf_ct_l4proto_find(orig->src.l3num,
891 orig->dst.protonum));
895 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
897 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
898 implicitly racy: see __nf_conntrack_confirm */
899 void nf_conntrack_alter_reply(struct nf_conn *ct,
900 const struct nf_conntrack_tuple *newreply)
902 struct nf_conn_help *help = nfct_help(ct);
904 write_lock_bh(&nf_conntrack_lock);
905 /* Should be unconfirmed, so not in hash table yet */
906 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
908 DEBUGP("Altering reply tuple of %p to ", ct);
909 NF_CT_DUMP_TUPLE(newreply);
911 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
912 if (!ct->master && help && help->expecting == 0)
913 help->helper = __nf_ct_helper_find(newreply);
914 write_unlock_bh(&nf_conntrack_lock);
916 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
918 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
919 void __nf_ct_refresh_acct(struct nf_conn *ct,
920 enum ip_conntrack_info ctinfo,
921 const struct sk_buff *skb,
922 unsigned long extra_jiffies,
927 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
930 write_lock_bh(&nf_conntrack_lock);
932 /* Only update if this is not a fixed timeout */
933 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
934 write_unlock_bh(&nf_conntrack_lock);
938 /* If not in hash table, timer will not be active yet */
939 if (!nf_ct_is_confirmed(ct)) {
940 ct->timeout.expires = extra_jiffies;
941 event = IPCT_REFRESH;
943 unsigned long newtime = jiffies + extra_jiffies;
945 /* Only update the timeout if the new timeout is at least
946 HZ jiffies from the old timeout. Need del_timer for race
947 avoidance (may already be dying). */
948 if (newtime - ct->timeout.expires >= HZ
949 && del_timer(&ct->timeout)) {
950 ct->timeout.expires = newtime;
951 add_timer(&ct->timeout);
952 event = IPCT_REFRESH;
956 #ifdef CONFIG_NF_CT_ACCT
958 ct->counters[CTINFO2DIR(ctinfo)].packets++;
959 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
960 skb->len - (unsigned int)(skb->nh.raw - skb->data);
962 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
963 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
964 event |= IPCT_COUNTER_FILLING;
968 write_unlock_bh(&nf_conntrack_lock);
970 /* must be unlocked when calling event cache */
972 nf_conntrack_event_cache(event, skb);
974 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
976 #if defined(CONFIG_NF_CT_NETLINK) || \
977 defined(CONFIG_NF_CT_NETLINK_MODULE)
979 #include <linux/netfilter/nfnetlink.h>
980 #include <linux/netfilter/nfnetlink_conntrack.h>
981 #include <linux/mutex.h>
984 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
985 * in ip_conntrack_core, since we don't want the protocols to autoload
986 * or depend on ctnetlink */
987 int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
988 const struct nf_conntrack_tuple *tuple)
990 NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
991 &tuple->src.u.tcp.port);
992 NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
993 &tuple->dst.u.tcp.port);
999 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr);
1001 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
1002 [CTA_PROTO_SRC_PORT-1] = sizeof(u_int16_t),
1003 [CTA_PROTO_DST_PORT-1] = sizeof(u_int16_t)
1006 int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[],
1007 struct nf_conntrack_tuple *t)
1009 if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1])
1012 if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
1015 t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]);
1016 t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]);
1020 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple);
1023 /* Used by ipt_REJECT and ip6t_REJECT. */
1024 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1027 enum ip_conntrack_info ctinfo;
1029 /* This ICMP is in reverse direction to the packet which caused it */
1030 ct = nf_ct_get(skb, &ctinfo);
1031 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1032 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
1034 ctinfo = IP_CT_RELATED;
1036 /* Attach to new skbuff, and increment count */
1037 nskb->nfct = &ct->ct_general;
1038 nskb->nfctinfo = ctinfo;
1039 nf_conntrack_get(nskb->nfct);
1041 EXPORT_SYMBOL_GPL(__nf_conntrack_attach);
1044 do_iter(const struct nf_conntrack_tuple_hash *i,
1045 int (*iter)(struct nf_conn *i, void *data),
1048 return iter(nf_ct_tuplehash_to_ctrack(i), data);
1051 /* Bring out ya dead! */
1052 static struct nf_conn *
1053 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
1054 void *data, unsigned int *bucket)
1056 struct nf_conntrack_tuple_hash *h;
1059 write_lock_bh(&nf_conntrack_lock);
1060 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1061 list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) {
1062 ct = nf_ct_tuplehash_to_ctrack(h);
1067 list_for_each_entry(h, &unconfirmed, list) {
1068 ct = nf_ct_tuplehash_to_ctrack(h);
1072 write_unlock_bh(&nf_conntrack_lock);
1075 atomic_inc(&ct->ct_general.use);
1076 write_unlock_bh(&nf_conntrack_lock);
1081 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
1084 unsigned int bucket = 0;
1086 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
1087 /* Time to push up daises... */
1088 if (del_timer(&ct->timeout))
1089 death_by_timeout((unsigned long)ct);
1090 /* ... else the timer will get him soon. */
1095 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1097 static int kill_all(struct nf_conn *i, void *data)
1102 static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size)
1107 free_pages((unsigned long)hash,
1108 get_order(sizeof(struct list_head) * size));
1111 void nf_conntrack_flush(void)
1113 nf_ct_iterate_cleanup(kill_all, NULL);
1115 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1117 /* Mishearing the voices in his head, our hero wonders how he's
1118 supposed to kill the mall. */
1119 void nf_conntrack_cleanup(void)
1123 rcu_assign_pointer(ip_ct_attach, NULL);
1125 /* This makes sure all current packets have passed through
1126 netfilter framework. Roll on, two-stage module
1130 nf_ct_event_cache_flush();
1132 nf_conntrack_flush();
1133 if (atomic_read(&nf_conntrack_count) != 0) {
1135 goto i_see_dead_people;
1137 /* wait until all references to nf_conntrack_untracked are dropped */
1138 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1141 for (i = 0; i < NF_CT_F_NUM; i++) {
1142 if (nf_ct_cache[i].use == 0)
1145 NF_CT_ASSERT(nf_ct_cache[i].use == 1);
1146 nf_ct_cache[i].use = 1;
1147 nf_conntrack_unregister_cache(i);
1149 kmem_cache_destroy(nf_conntrack_expect_cachep);
1150 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1151 nf_conntrack_htable_size);
1153 nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_generic);
1155 /* free l3proto protocol tables */
1156 for (i = 0; i < PF_MAX; i++)
1157 if (nf_ct_protos[i]) {
1158 kfree(nf_ct_protos[i]);
1159 nf_ct_protos[i] = NULL;
1163 static struct list_head *alloc_hashtable(int size, int *vmalloced)
1165 struct list_head *hash;
1169 hash = (void*)__get_free_pages(GFP_KERNEL,
1170 get_order(sizeof(struct list_head)
1174 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1175 hash = vmalloc(sizeof(struct list_head) * size);
1179 for (i = 0; i < size; i++)
1180 INIT_LIST_HEAD(&hash[i]);
1185 int set_hashsize(const char *val, struct kernel_param *kp)
1187 int i, bucket, hashsize, vmalloced;
1188 int old_vmalloced, old_size;
1190 struct list_head *hash, *old_hash;
1191 struct nf_conntrack_tuple_hash *h;
1193 /* On boot, we can set this without any fancy locking. */
1194 if (!nf_conntrack_htable_size)
1195 return param_set_uint(val, kp);
1197 hashsize = simple_strtol(val, NULL, 0);
1201 hash = alloc_hashtable(hashsize, &vmalloced);
1205 /* We have to rehahs for the new table anyway, so we also can
1206 * use a newrandom seed */
1207 get_random_bytes(&rnd, 4);
1209 write_lock_bh(&nf_conntrack_lock);
1210 for (i = 0; i < nf_conntrack_htable_size; i++) {
1211 while (!list_empty(&nf_conntrack_hash[i])) {
1212 h = list_entry(nf_conntrack_hash[i].next,
1213 struct nf_conntrack_tuple_hash, list);
1215 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1216 list_add_tail(&h->list, &hash[bucket]);
1219 old_size = nf_conntrack_htable_size;
1220 old_vmalloced = nf_conntrack_vmalloc;
1221 old_hash = nf_conntrack_hash;
1223 nf_conntrack_htable_size = hashsize;
1224 nf_conntrack_vmalloc = vmalloced;
1225 nf_conntrack_hash = hash;
1226 nf_conntrack_hash_rnd = rnd;
1227 write_unlock_bh(&nf_conntrack_lock);
1229 free_conntrack_hash(old_hash, old_vmalloced, old_size);
1233 module_param_call(hashsize, set_hashsize, param_get_uint,
1234 &nf_conntrack_htable_size, 0600);
1236 int __init nf_conntrack_init(void)
1241 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1242 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */
1243 if (!nf_conntrack_htable_size) {
1244 nf_conntrack_htable_size
1245 = (((num_physpages << PAGE_SHIFT) / 16384)
1246 / sizeof(struct list_head));
1247 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1248 nf_conntrack_htable_size = 8192;
1249 if (nf_conntrack_htable_size < 16)
1250 nf_conntrack_htable_size = 16;
1252 nf_conntrack_max = 8 * nf_conntrack_htable_size;
1254 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1255 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1258 nf_conntrack_hash = alloc_hashtable(nf_conntrack_htable_size,
1259 &nf_conntrack_vmalloc);
1260 if (!nf_conntrack_hash) {
1261 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1265 ret = nf_conntrack_register_cache(NF_CT_F_BASIC, "nf_conntrack:basic",
1266 sizeof(struct nf_conn));
1268 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1272 nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect",
1273 sizeof(struct nf_conntrack_expect),
1275 if (!nf_conntrack_expect_cachep) {
1276 printk(KERN_ERR "Unable to create nf_expect slab cache\n");
1277 goto err_free_conntrack_slab;
1280 ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_generic);
1282 goto out_free_expect_slab;
1284 /* Don't NEED lock here, but good form anyway. */
1285 write_lock_bh(&nf_conntrack_lock);
1286 for (i = 0; i < AF_MAX; i++)
1287 nf_ct_l3protos[i] = &nf_conntrack_l3proto_generic;
1288 write_unlock_bh(&nf_conntrack_lock);
1290 /* For use by REJECT target */
1291 rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
1293 /* Set up fake conntrack:
1294 - to never be deleted, not in any hashes */
1295 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1296 /* - and look it like as a confirmed connection */
1297 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1301 out_free_expect_slab:
1302 kmem_cache_destroy(nf_conntrack_expect_cachep);
1303 err_free_conntrack_slab:
1304 nf_conntrack_unregister_cache(NF_CT_F_BASIC);
1306 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1307 nf_conntrack_htable_size);