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>
40 #define NF_CONNTRACK_VERSION "0.5.0"
45 #define DEBUGP(format, args...)
48 DEFINE_RWLOCK(nf_conntrack_lock
);
49 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
51 /* nf_conntrack_standalone needs this */
52 atomic_t nf_conntrack_count
= ATOMIC_INIT(0);
53 EXPORT_SYMBOL_GPL(nf_conntrack_count
);
55 void (*nf_conntrack_destroyed
)(struct nf_conn
*conntrack
);
56 EXPORT_SYMBOL_GPL(nf_conntrack_destroyed
);
58 unsigned int nf_conntrack_htable_size __read_mostly
;
59 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
61 int nf_conntrack_max __read_mostly
;
62 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
64 struct list_head
*nf_conntrack_hash __read_mostly
;
65 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
67 struct nf_conn nf_conntrack_untracked __read_mostly
;
68 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
70 unsigned int nf_ct_log_invalid __read_mostly
;
71 LIST_HEAD(unconfirmed
);
72 static int nf_conntrack_vmalloc __read_mostly
;
74 static unsigned int nf_conntrack_next_id
;
76 DEFINE_PER_CPU(struct ip_conntrack_stat
, nf_conntrack_stat
);
77 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat
);
80 * This scheme offers various size of "struct nf_conn" dependent on
81 * features(helper, nat, ...)
84 #define NF_CT_FEATURES_NAMELEN 256
86 /* name of slab cache. printed in /proc/slabinfo */
89 /* size of slab cache */
92 /* slab cache pointer */
93 struct kmem_cache
*cachep
;
95 /* allocated slab cache + modules which uses this slab cache */
98 } nf_ct_cache
[NF_CT_F_NUM
];
100 /* protect members of nf_ct_cache except of "use" */
101 DEFINE_RWLOCK(nf_ct_cache_lock
);
103 /* This avoids calling kmem_cache_create() with same name simultaneously */
104 static DEFINE_MUTEX(nf_ct_cache_mutex
);
106 static int nf_conntrack_hash_rnd_initted
;
107 static unsigned int nf_conntrack_hash_rnd
;
109 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
110 unsigned int size
, unsigned int rnd
)
114 a
= jhash2(tuple
->src
.u3
.all
, ARRAY_SIZE(tuple
->src
.u3
.all
),
115 (tuple
->src
.l3num
<< 16) | tuple
->dst
.protonum
);
116 b
= jhash2(tuple
->dst
.u3
.all
, ARRAY_SIZE(tuple
->dst
.u3
.all
),
117 (tuple
->src
.u
.all
<< 16) | tuple
->dst
.u
.all
);
119 return jhash_2words(a
, b
, rnd
) % size
;
122 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
124 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
125 nf_conntrack_hash_rnd
);
128 int nf_conntrack_register_cache(u_int32_t features
, const char *name
,
133 struct kmem_cache
*cachep
;
135 DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n",
136 features
, name
, size
);
138 if (features
< NF_CT_F_BASIC
|| features
>= NF_CT_F_NUM
) {
139 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n",
144 mutex_lock(&nf_ct_cache_mutex
);
146 write_lock_bh(&nf_ct_cache_lock
);
147 /* e.g: multiple helpers are loaded */
148 if (nf_ct_cache
[features
].use
> 0) {
149 DEBUGP("nf_conntrack_register_cache: already resisterd.\n");
150 if ((!strncmp(nf_ct_cache
[features
].name
, name
,
151 NF_CT_FEATURES_NAMELEN
))
152 && nf_ct_cache
[features
].size
== size
) {
153 DEBUGP("nf_conntrack_register_cache: reusing.\n");
154 nf_ct_cache
[features
].use
++;
159 write_unlock_bh(&nf_ct_cache_lock
);
160 mutex_unlock(&nf_ct_cache_mutex
);
163 write_unlock_bh(&nf_ct_cache_lock
);
166 * The memory space for name of slab cache must be alive until
167 * cache is destroyed.
169 cache_name
= kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN
, GFP_ATOMIC
);
170 if (cache_name
== NULL
) {
171 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n");
176 if (strlcpy(cache_name
, name
, NF_CT_FEATURES_NAMELEN
)
177 >= NF_CT_FEATURES_NAMELEN
) {
178 printk("nf_conntrack_register_cache: name too long\n");
183 cachep
= kmem_cache_create(cache_name
, size
, 0, 0,
186 printk("nf_conntrack_register_cache: Can't create slab cache "
187 "for the features = 0x%x\n", features
);
192 write_lock_bh(&nf_ct_cache_lock
);
193 nf_ct_cache
[features
].use
= 1;
194 nf_ct_cache
[features
].size
= size
;
195 nf_ct_cache
[features
].cachep
= cachep
;
196 nf_ct_cache
[features
].name
= cache_name
;
197 write_unlock_bh(&nf_ct_cache_lock
);
204 mutex_unlock(&nf_ct_cache_mutex
);
207 EXPORT_SYMBOL_GPL(nf_conntrack_register_cache
);
209 /* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */
210 void nf_conntrack_unregister_cache(u_int32_t features
)
212 struct kmem_cache
*cachep
;
216 * This assures that kmem_cache_create() isn't called before destroying
219 DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features
);
220 mutex_lock(&nf_ct_cache_mutex
);
222 write_lock_bh(&nf_ct_cache_lock
);
223 if (--nf_ct_cache
[features
].use
> 0) {
224 write_unlock_bh(&nf_ct_cache_lock
);
225 mutex_unlock(&nf_ct_cache_mutex
);
228 cachep
= nf_ct_cache
[features
].cachep
;
229 name
= nf_ct_cache
[features
].name
;
230 nf_ct_cache
[features
].cachep
= NULL
;
231 nf_ct_cache
[features
].name
= NULL
;
232 nf_ct_cache
[features
].size
= 0;
233 write_unlock_bh(&nf_ct_cache_lock
);
237 kmem_cache_destroy(cachep
);
240 mutex_unlock(&nf_ct_cache_mutex
);
242 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_cache
);
245 nf_ct_get_tuple(const struct sk_buff
*skb
,
247 unsigned int dataoff
,
250 struct nf_conntrack_tuple
*tuple
,
251 const struct nf_conntrack_l3proto
*l3proto
,
252 const struct nf_conntrack_l4proto
*l4proto
)
254 NF_CT_TUPLE_U_BLANK(tuple
);
256 tuple
->src
.l3num
= l3num
;
257 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
260 tuple
->dst
.protonum
= protonum
;
261 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
263 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
265 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
268 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
269 const struct nf_conntrack_tuple
*orig
,
270 const struct nf_conntrack_l3proto
*l3proto
,
271 const struct nf_conntrack_l4proto
*l4proto
)
273 NF_CT_TUPLE_U_BLANK(inverse
);
275 inverse
->src
.l3num
= orig
->src
.l3num
;
276 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
279 inverse
->dst
.dir
= !orig
->dst
.dir
;
281 inverse
->dst
.protonum
= orig
->dst
.protonum
;
282 return l4proto
->invert_tuple(inverse
, orig
);
284 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
287 clean_from_lists(struct nf_conn
*ct
)
289 DEBUGP("clean_from_lists(%p)\n", ct
);
290 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
291 list_del(&ct
->tuplehash
[IP_CT_DIR_REPLY
].list
);
293 /* Destroy all pending expectations */
294 nf_ct_remove_expectations(ct
);
298 destroy_conntrack(struct nf_conntrack
*nfct
)
300 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
301 struct nf_conn_help
*help
= nfct_help(ct
);
302 struct nf_conntrack_l3proto
*l3proto
;
303 struct nf_conntrack_l4proto
*l4proto
;
304 typeof(nf_conntrack_destroyed
) destroyed
;
306 DEBUGP("destroy_conntrack(%p)\n", ct
);
307 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
308 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
310 nf_conntrack_event(IPCT_DESTROY
, ct
);
311 set_bit(IPS_DYING_BIT
, &ct
->status
);
313 if (help
&& help
->helper
&& help
->helper
->destroy
)
314 help
->helper
->destroy(ct
);
316 /* To make sure we don't get any weird locking issues here:
317 * destroy_conntrack() MUST NOT be called with a write lock
318 * to nf_conntrack_lock!!! -HW */
320 l3proto
= __nf_ct_l3proto_find(ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.src
.l3num
);
321 if (l3proto
&& l3proto
->destroy
)
322 l3proto
->destroy(ct
);
324 l4proto
= __nf_ct_l4proto_find(ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.src
.l3num
,
325 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.dst
.protonum
);
326 if (l4proto
&& l4proto
->destroy
)
327 l4proto
->destroy(ct
);
329 destroyed
= rcu_dereference(nf_conntrack_destroyed
);
335 write_lock_bh(&nf_conntrack_lock
);
336 /* Expectations will have been removed in clean_from_lists,
337 * except TFTP can create an expectation on the first packet,
338 * before connection is in the list, so we need to clean here,
340 nf_ct_remove_expectations(ct
);
342 /* We overload first tuple to link into unconfirmed list. */
343 if (!nf_ct_is_confirmed(ct
)) {
344 BUG_ON(list_empty(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
));
345 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
348 NF_CT_STAT_INC(delete);
349 write_unlock_bh(&nf_conntrack_lock
);
352 nf_ct_put(ct
->master
);
354 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct
);
355 nf_conntrack_free(ct
);
358 static void death_by_timeout(unsigned long ul_conntrack
)
360 struct nf_conn
*ct
= (void *)ul_conntrack
;
362 write_lock_bh(&nf_conntrack_lock
);
363 /* Inside lock so preempt is disabled on module removal path.
364 * Otherwise we can get spurious warnings. */
365 NF_CT_STAT_INC(delete_list
);
366 clean_from_lists(ct
);
367 write_unlock_bh(&nf_conntrack_lock
);
371 struct nf_conntrack_tuple_hash
*
372 __nf_conntrack_find(const struct nf_conntrack_tuple
*tuple
,
373 const struct nf_conn
*ignored_conntrack
)
375 struct nf_conntrack_tuple_hash
*h
;
376 unsigned int hash
= hash_conntrack(tuple
);
378 list_for_each_entry(h
, &nf_conntrack_hash
[hash
], list
) {
379 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
380 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
381 NF_CT_STAT_INC(found
);
384 NF_CT_STAT_INC(searched
);
389 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
391 /* Find a connection corresponding to a tuple. */
392 struct nf_conntrack_tuple_hash
*
393 nf_conntrack_find_get(const struct nf_conntrack_tuple
*tuple
,
394 const struct nf_conn
*ignored_conntrack
)
396 struct nf_conntrack_tuple_hash
*h
;
398 read_lock_bh(&nf_conntrack_lock
);
399 h
= __nf_conntrack_find(tuple
, ignored_conntrack
);
401 atomic_inc(&nf_ct_tuplehash_to_ctrack(h
)->ct_general
.use
);
402 read_unlock_bh(&nf_conntrack_lock
);
406 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
408 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
410 unsigned int repl_hash
)
412 ct
->id
= ++nf_conntrack_next_id
;
413 list_add(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
,
414 &nf_conntrack_hash
[hash
]);
415 list_add(&ct
->tuplehash
[IP_CT_DIR_REPLY
].list
,
416 &nf_conntrack_hash
[repl_hash
]);
419 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
421 unsigned int hash
, repl_hash
;
423 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
424 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
426 write_lock_bh(&nf_conntrack_lock
);
427 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
428 write_unlock_bh(&nf_conntrack_lock
);
430 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
432 /* Confirm a connection given skb; places it in hash table */
434 __nf_conntrack_confirm(struct sk_buff
**pskb
)
436 unsigned int hash
, repl_hash
;
437 struct nf_conntrack_tuple_hash
*h
;
439 struct nf_conn_help
*help
;
440 enum ip_conntrack_info ctinfo
;
442 ct
= nf_ct_get(*pskb
, &ctinfo
);
444 /* ipt_REJECT uses nf_conntrack_attach to attach related
445 ICMP/TCP RST packets in other direction. Actual packet
446 which created connection will be IP_CT_NEW or for an
447 expected connection, IP_CT_RELATED. */
448 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
451 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
452 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
454 /* We're not in hash table, and we refuse to set up related
455 connections for unconfirmed conns. But packet copies and
456 REJECT will give spurious warnings here. */
457 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
459 /* No external references means noone else could have
461 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
462 DEBUGP("Confirming conntrack %p\n", ct
);
464 write_lock_bh(&nf_conntrack_lock
);
466 /* See if there's one in the list already, including reverse:
467 NAT could have grabbed it without realizing, since we're
468 not in the hash. If there is, we lost race. */
469 list_for_each_entry(h
, &nf_conntrack_hash
[hash
], list
)
470 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
473 list_for_each_entry(h
, &nf_conntrack_hash
[repl_hash
], list
)
474 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
478 /* Remove from unconfirmed list */
479 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
481 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
482 /* Timer relative to confirmation time, not original
483 setting time, otherwise we'd get timer wrap in
484 weird delay cases. */
485 ct
->timeout
.expires
+= jiffies
;
486 add_timer(&ct
->timeout
);
487 atomic_inc(&ct
->ct_general
.use
);
488 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
489 NF_CT_STAT_INC(insert
);
490 write_unlock_bh(&nf_conntrack_lock
);
491 help
= nfct_help(ct
);
492 if (help
&& help
->helper
)
493 nf_conntrack_event_cache(IPCT_HELPER
, *pskb
);
494 #ifdef CONFIG_NF_NAT_NEEDED
495 if (test_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
) ||
496 test_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
))
497 nf_conntrack_event_cache(IPCT_NATINFO
, *pskb
);
499 nf_conntrack_event_cache(master_ct(ct
) ?
500 IPCT_RELATED
: IPCT_NEW
, *pskb
);
504 NF_CT_STAT_INC(insert_failed
);
505 write_unlock_bh(&nf_conntrack_lock
);
508 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
510 /* Returns true if a connection correspondings to the tuple (required
513 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
514 const struct nf_conn
*ignored_conntrack
)
516 struct nf_conntrack_tuple_hash
*h
;
518 read_lock_bh(&nf_conntrack_lock
);
519 h
= __nf_conntrack_find(tuple
, ignored_conntrack
);
520 read_unlock_bh(&nf_conntrack_lock
);
524 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
526 /* There's a small race here where we may free a just-assured
527 connection. Too bad: we're in trouble anyway. */
528 static int early_drop(struct list_head
*chain
)
530 /* Traverse backwards: gives us oldest, which is roughly LRU */
531 struct nf_conntrack_tuple_hash
*h
;
532 struct nf_conn
*ct
= NULL
, *tmp
;
535 read_lock_bh(&nf_conntrack_lock
);
536 list_for_each_entry_reverse(h
, chain
, list
) {
537 tmp
= nf_ct_tuplehash_to_ctrack(h
);
538 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
)) {
540 atomic_inc(&ct
->ct_general
.use
);
544 read_unlock_bh(&nf_conntrack_lock
);
549 if (del_timer(&ct
->timeout
)) {
550 death_by_timeout((unsigned long)ct
);
552 NF_CT_STAT_INC_ATOMIC(early_drop
);
558 static struct nf_conn
*
559 __nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
560 const struct nf_conntrack_tuple
*repl
,
561 const struct nf_conntrack_l3proto
*l3proto
,
564 struct nf_conn
*conntrack
= NULL
;
565 struct nf_conntrack_helper
*helper
;
567 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
568 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
569 nf_conntrack_hash_rnd_initted
= 1;
572 /* We don't want any race condition at early drop stage */
573 atomic_inc(&nf_conntrack_count
);
576 && atomic_read(&nf_conntrack_count
) > nf_conntrack_max
) {
577 unsigned int hash
= hash_conntrack(orig
);
578 /* Try dropping from this hash chain. */
579 if (!early_drop(&nf_conntrack_hash
[hash
])) {
580 atomic_dec(&nf_conntrack_count
);
583 "nf_conntrack: table full, dropping"
585 return ERR_PTR(-ENOMEM
);
589 /* find features needed by this conntrack. */
590 features
|= l3proto
->get_features(orig
);
592 /* FIXME: protect helper list per RCU */
593 read_lock_bh(&nf_conntrack_lock
);
594 helper
= __nf_ct_helper_find(repl
);
595 /* NAT might want to assign a helper later */
596 if (helper
|| features
& NF_CT_F_NAT
)
597 features
|= NF_CT_F_HELP
;
598 read_unlock_bh(&nf_conntrack_lock
);
600 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features
);
602 read_lock_bh(&nf_ct_cache_lock
);
604 if (unlikely(!nf_ct_cache
[features
].use
)) {
605 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n",
610 conntrack
= kmem_cache_alloc(nf_ct_cache
[features
].cachep
, GFP_ATOMIC
);
611 if (conntrack
== NULL
) {
612 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n");
616 memset(conntrack
, 0, nf_ct_cache
[features
].size
);
617 conntrack
->features
= features
;
618 atomic_set(&conntrack
->ct_general
.use
, 1);
619 conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
620 conntrack
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
621 /* Don't set timer yet: wait for confirmation */
622 setup_timer(&conntrack
->timeout
, death_by_timeout
,
623 (unsigned long)conntrack
);
624 read_unlock_bh(&nf_ct_cache_lock
);
628 read_unlock_bh(&nf_ct_cache_lock
);
629 atomic_dec(&nf_conntrack_count
);
633 struct nf_conn
*nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
634 const struct nf_conntrack_tuple
*repl
)
636 struct nf_conntrack_l3proto
*l3proto
;
640 l3proto
= __nf_ct_l3proto_find(orig
->src
.l3num
);
641 ct
= __nf_conntrack_alloc(orig
, repl
, l3proto
, 0);
646 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
648 void nf_conntrack_free(struct nf_conn
*conntrack
)
650 u_int32_t features
= conntrack
->features
;
651 NF_CT_ASSERT(features
>= NF_CT_F_BASIC
&& features
< NF_CT_F_NUM
);
652 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features
,
654 kmem_cache_free(nf_ct_cache
[features
].cachep
, conntrack
);
655 atomic_dec(&nf_conntrack_count
);
657 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
659 /* Allocate a new conntrack: we return -ENOMEM if classification
660 failed due to stress. Otherwise it really is unclassifiable. */
661 static struct nf_conntrack_tuple_hash
*
662 init_conntrack(const struct nf_conntrack_tuple
*tuple
,
663 struct nf_conntrack_l3proto
*l3proto
,
664 struct nf_conntrack_l4proto
*l4proto
,
666 unsigned int dataoff
)
668 struct nf_conn
*conntrack
;
669 struct nf_conntrack_tuple repl_tuple
;
670 struct nf_conntrack_expect
*exp
;
671 u_int32_t features
= 0;
673 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
674 DEBUGP("Can't invert tuple.\n");
678 read_lock_bh(&nf_conntrack_lock
);
679 exp
= __nf_conntrack_expect_find(tuple
);
680 if (exp
&& exp
->helper
)
681 features
= NF_CT_F_HELP
;
682 read_unlock_bh(&nf_conntrack_lock
);
684 conntrack
= __nf_conntrack_alloc(tuple
, &repl_tuple
, l3proto
, features
);
685 if (conntrack
== NULL
|| IS_ERR(conntrack
)) {
686 DEBUGP("Can't allocate conntrack.\n");
687 return (struct nf_conntrack_tuple_hash
*)conntrack
;
690 if (!l4proto
->new(conntrack
, skb
, dataoff
)) {
691 nf_conntrack_free(conntrack
);
692 DEBUGP("init conntrack: can't track with proto module\n");
696 write_lock_bh(&nf_conntrack_lock
);
697 exp
= find_expectation(tuple
);
700 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
702 /* Welcome, Mr. Bond. We've been expecting you... */
703 __set_bit(IPS_EXPECTED_BIT
, &conntrack
->status
);
704 conntrack
->master
= exp
->master
;
706 nfct_help(conntrack
)->helper
= exp
->helper
;
707 #ifdef CONFIG_NF_CONNTRACK_MARK
708 conntrack
->mark
= exp
->master
->mark
;
710 #ifdef CONFIG_NF_CONNTRACK_SECMARK
711 conntrack
->secmark
= exp
->master
->secmark
;
713 nf_conntrack_get(&conntrack
->master
->ct_general
);
714 NF_CT_STAT_INC(expect_new
);
716 struct nf_conn_help
*help
= nfct_help(conntrack
);
719 help
->helper
= __nf_ct_helper_find(&repl_tuple
);
723 /* Overload tuple linked list to put us in unconfirmed list. */
724 list_add(&conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
, &unconfirmed
);
726 write_unlock_bh(&nf_conntrack_lock
);
730 exp
->expectfn(conntrack
, exp
);
731 nf_conntrack_expect_put(exp
);
734 return &conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
];
737 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
738 static inline struct nf_conn
*
739 resolve_normal_ct(struct sk_buff
*skb
,
740 unsigned int dataoff
,
743 struct nf_conntrack_l3proto
*l3proto
,
744 struct nf_conntrack_l4proto
*l4proto
,
746 enum ip_conntrack_info
*ctinfo
)
748 struct nf_conntrack_tuple tuple
;
749 struct nf_conntrack_tuple_hash
*h
;
752 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
753 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
755 DEBUGP("resolve_normal_ct: Can't get tuple\n");
759 /* look for tuple match */
760 h
= nf_conntrack_find_get(&tuple
, NULL
);
762 h
= init_conntrack(&tuple
, l3proto
, l4proto
, skb
, dataoff
);
768 ct
= nf_ct_tuplehash_to_ctrack(h
);
770 /* It exists; we have (non-exclusive) reference. */
771 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
772 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
773 /* Please set reply bit if this packet OK */
776 /* Once we've had two way comms, always ESTABLISHED. */
777 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
778 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct
);
779 *ctinfo
= IP_CT_ESTABLISHED
;
780 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
781 DEBUGP("nf_conntrack_in: related packet for %p\n", ct
);
782 *ctinfo
= IP_CT_RELATED
;
784 DEBUGP("nf_conntrack_in: new packet for %p\n", ct
);
789 skb
->nfct
= &ct
->ct_general
;
790 skb
->nfctinfo
= *ctinfo
;
795 nf_conntrack_in(int pf
, unsigned int hooknum
, struct sk_buff
**pskb
)
798 enum ip_conntrack_info ctinfo
;
799 struct nf_conntrack_l3proto
*l3proto
;
800 struct nf_conntrack_l4proto
*l4proto
;
801 unsigned int dataoff
;
806 /* Previously seen (loopback or untracked)? Ignore. */
808 NF_CT_STAT_INC_ATOMIC(ignore
);
812 /* rcu_read_lock()ed by nf_hook_slow */
813 l3proto
= __nf_ct_l3proto_find((u_int16_t
)pf
);
815 if ((ret
= l3proto
->prepare(pskb
, hooknum
, &dataoff
, &protonum
)) <= 0) {
816 DEBUGP("not prepared to track yet or error occured\n");
820 l4proto
= __nf_ct_l4proto_find((u_int16_t
)pf
, protonum
);
822 /* It may be an special packet, error, unclean...
823 * inverse of the return code tells to the netfilter
824 * core what to do with the packet. */
825 if (l4proto
->error
!= NULL
&&
826 (ret
= l4proto
->error(*pskb
, dataoff
, &ctinfo
, pf
, hooknum
)) <= 0) {
827 NF_CT_STAT_INC_ATOMIC(error
);
828 NF_CT_STAT_INC_ATOMIC(invalid
);
832 ct
= resolve_normal_ct(*pskb
, dataoff
, pf
, protonum
, l3proto
, l4proto
,
833 &set_reply
, &ctinfo
);
835 /* Not valid part of a connection */
836 NF_CT_STAT_INC_ATOMIC(invalid
);
841 /* Too stressed to deal. */
842 NF_CT_STAT_INC_ATOMIC(drop
);
846 NF_CT_ASSERT((*pskb
)->nfct
);
848 ret
= l4proto
->packet(ct
, *pskb
, dataoff
, ctinfo
, pf
, hooknum
);
850 /* Invalid: inverse of the return code tells
851 * the netfilter core what to do */
852 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
853 nf_conntrack_put((*pskb
)->nfct
);
854 (*pskb
)->nfct
= NULL
;
855 NF_CT_STAT_INC_ATOMIC(invalid
);
859 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
860 nf_conntrack_event_cache(IPCT_STATUS
, *pskb
);
864 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
866 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
867 const struct nf_conntrack_tuple
*orig
)
872 ret
= nf_ct_invert_tuple(inverse
, orig
,
873 __nf_ct_l3proto_find(orig
->src
.l3num
),
874 __nf_ct_l4proto_find(orig
->src
.l3num
,
875 orig
->dst
.protonum
));
879 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
881 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
882 implicitly racy: see __nf_conntrack_confirm */
883 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
884 const struct nf_conntrack_tuple
*newreply
)
886 struct nf_conn_help
*help
= nfct_help(ct
);
888 write_lock_bh(&nf_conntrack_lock
);
889 /* Should be unconfirmed, so not in hash table yet */
890 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
892 DEBUGP("Altering reply tuple of %p to ", ct
);
893 NF_CT_DUMP_TUPLE(newreply
);
895 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
896 if (!ct
->master
&& help
&& help
->expecting
== 0)
897 help
->helper
= __nf_ct_helper_find(newreply
);
898 write_unlock_bh(&nf_conntrack_lock
);
900 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
902 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
903 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
904 enum ip_conntrack_info ctinfo
,
905 const struct sk_buff
*skb
,
906 unsigned long extra_jiffies
,
911 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
914 write_lock_bh(&nf_conntrack_lock
);
916 /* Only update if this is not a fixed timeout */
917 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
)) {
918 write_unlock_bh(&nf_conntrack_lock
);
922 /* If not in hash table, timer will not be active yet */
923 if (!nf_ct_is_confirmed(ct
)) {
924 ct
->timeout
.expires
= extra_jiffies
;
925 event
= IPCT_REFRESH
;
927 unsigned long newtime
= jiffies
+ extra_jiffies
;
929 /* Only update the timeout if the new timeout is at least
930 HZ jiffies from the old timeout. Need del_timer for race
931 avoidance (may already be dying). */
932 if (newtime
- ct
->timeout
.expires
>= HZ
933 && del_timer(&ct
->timeout
)) {
934 ct
->timeout
.expires
= newtime
;
935 add_timer(&ct
->timeout
);
936 event
= IPCT_REFRESH
;
940 #ifdef CONFIG_NF_CT_ACCT
942 ct
->counters
[CTINFO2DIR(ctinfo
)].packets
++;
943 ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
+=
944 skb
->len
- skb_network_offset(skb
);
946 if ((ct
->counters
[CTINFO2DIR(ctinfo
)].packets
& 0x80000000)
947 || (ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
& 0x80000000))
948 event
|= IPCT_COUNTER_FILLING
;
952 write_unlock_bh(&nf_conntrack_lock
);
954 /* must be unlocked when calling event cache */
956 nf_conntrack_event_cache(event
, skb
);
958 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
960 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
962 #include <linux/netfilter/nfnetlink.h>
963 #include <linux/netfilter/nfnetlink_conntrack.h>
964 #include <linux/mutex.h>
967 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
968 * in ip_conntrack_core, since we don't want the protocols to autoload
969 * or depend on ctnetlink */
970 int nf_ct_port_tuple_to_nfattr(struct sk_buff
*skb
,
971 const struct nf_conntrack_tuple
*tuple
)
973 NFA_PUT(skb
, CTA_PROTO_SRC_PORT
, sizeof(u_int16_t
),
974 &tuple
->src
.u
.tcp
.port
);
975 NFA_PUT(skb
, CTA_PROTO_DST_PORT
, sizeof(u_int16_t
),
976 &tuple
->dst
.u
.tcp
.port
);
982 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr
);
984 static const size_t cta_min_proto
[CTA_PROTO_MAX
] = {
985 [CTA_PROTO_SRC_PORT
-1] = sizeof(u_int16_t
),
986 [CTA_PROTO_DST_PORT
-1] = sizeof(u_int16_t
)
989 int nf_ct_port_nfattr_to_tuple(struct nfattr
*tb
[],
990 struct nf_conntrack_tuple
*t
)
992 if (!tb
[CTA_PROTO_SRC_PORT
-1] || !tb
[CTA_PROTO_DST_PORT
-1])
995 if (nfattr_bad_size(tb
, CTA_PROTO_MAX
, cta_min_proto
))
998 t
->src
.u
.tcp
.port
= *(__be16
*)NFA_DATA(tb
[CTA_PROTO_SRC_PORT
-1]);
999 t
->dst
.u
.tcp
.port
= *(__be16
*)NFA_DATA(tb
[CTA_PROTO_DST_PORT
-1]);
1003 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple
);
1006 /* Used by ipt_REJECT and ip6t_REJECT. */
1007 void __nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
1010 enum ip_conntrack_info ctinfo
;
1012 /* This ICMP is in reverse direction to the packet which caused it */
1013 ct
= nf_ct_get(skb
, &ctinfo
);
1014 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1015 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
1017 ctinfo
= IP_CT_RELATED
;
1019 /* Attach to new skbuff, and increment count */
1020 nskb
->nfct
= &ct
->ct_general
;
1021 nskb
->nfctinfo
= ctinfo
;
1022 nf_conntrack_get(nskb
->nfct
);
1024 EXPORT_SYMBOL_GPL(__nf_conntrack_attach
);
1027 do_iter(const struct nf_conntrack_tuple_hash
*i
,
1028 int (*iter
)(struct nf_conn
*i
, void *data
),
1031 return iter(nf_ct_tuplehash_to_ctrack(i
), data
);
1034 /* Bring out ya dead! */
1035 static struct nf_conn
*
1036 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
1037 void *data
, unsigned int *bucket
)
1039 struct nf_conntrack_tuple_hash
*h
;
1042 write_lock_bh(&nf_conntrack_lock
);
1043 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1044 list_for_each_entry(h
, &nf_conntrack_hash
[*bucket
], list
) {
1045 ct
= nf_ct_tuplehash_to_ctrack(h
);
1050 list_for_each_entry(h
, &unconfirmed
, list
) {
1051 ct
= nf_ct_tuplehash_to_ctrack(h
);
1053 set_bit(IPS_DYING_BIT
, &ct
->status
);
1055 write_unlock_bh(&nf_conntrack_lock
);
1058 atomic_inc(&ct
->ct_general
.use
);
1059 write_unlock_bh(&nf_conntrack_lock
);
1064 nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
1067 unsigned int bucket
= 0;
1069 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
1070 /* Time to push up daises... */
1071 if (del_timer(&ct
->timeout
))
1072 death_by_timeout((unsigned long)ct
);
1073 /* ... else the timer will get him soon. */
1078 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1080 static int kill_all(struct nf_conn
*i
, void *data
)
1085 static void free_conntrack_hash(struct list_head
*hash
, int vmalloced
, int size
)
1090 free_pages((unsigned long)hash
,
1091 get_order(sizeof(struct list_head
) * size
));
1094 void nf_conntrack_flush(void)
1096 nf_ct_iterate_cleanup(kill_all
, NULL
);
1098 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
1100 /* Mishearing the voices in his head, our hero wonders how he's
1101 supposed to kill the mall. */
1102 void nf_conntrack_cleanup(void)
1106 rcu_assign_pointer(ip_ct_attach
, NULL
);
1108 /* This makes sure all current packets have passed through
1109 netfilter framework. Roll on, two-stage module
1113 nf_ct_event_cache_flush();
1115 nf_conntrack_flush();
1116 if (atomic_read(&nf_conntrack_count
) != 0) {
1118 goto i_see_dead_people
;
1120 /* wait until all references to nf_conntrack_untracked are dropped */
1121 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1124 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1126 for (i
= 0; i
< NF_CT_F_NUM
; i
++) {
1127 if (nf_ct_cache
[i
].use
== 0)
1130 NF_CT_ASSERT(nf_ct_cache
[i
].use
== 1);
1131 nf_ct_cache
[i
].use
= 1;
1132 nf_conntrack_unregister_cache(i
);
1134 kmem_cache_destroy(nf_conntrack_expect_cachep
);
1135 free_conntrack_hash(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1136 nf_conntrack_htable_size
);
1138 nf_conntrack_proto_fini();
1141 static struct list_head
*alloc_hashtable(int size
, int *vmalloced
)
1143 struct list_head
*hash
;
1147 hash
= (void*)__get_free_pages(GFP_KERNEL
,
1148 get_order(sizeof(struct list_head
)
1152 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1153 hash
= vmalloc(sizeof(struct list_head
) * size
);
1157 for (i
= 0; i
< size
; i
++)
1158 INIT_LIST_HEAD(&hash
[i
]);
1163 int set_hashsize(const char *val
, struct kernel_param
*kp
)
1165 int i
, bucket
, hashsize
, vmalloced
;
1166 int old_vmalloced
, old_size
;
1168 struct list_head
*hash
, *old_hash
;
1169 struct nf_conntrack_tuple_hash
*h
;
1171 /* On boot, we can set this without any fancy locking. */
1172 if (!nf_conntrack_htable_size
)
1173 return param_set_uint(val
, kp
);
1175 hashsize
= simple_strtol(val
, NULL
, 0);
1179 hash
= alloc_hashtable(hashsize
, &vmalloced
);
1183 /* We have to rehahs for the new table anyway, so we also can
1184 * use a newrandom seed */
1185 get_random_bytes(&rnd
, 4);
1187 write_lock_bh(&nf_conntrack_lock
);
1188 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1189 while (!list_empty(&nf_conntrack_hash
[i
])) {
1190 h
= list_entry(nf_conntrack_hash
[i
].next
,
1191 struct nf_conntrack_tuple_hash
, list
);
1193 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1194 list_add_tail(&h
->list
, &hash
[bucket
]);
1197 old_size
= nf_conntrack_htable_size
;
1198 old_vmalloced
= nf_conntrack_vmalloc
;
1199 old_hash
= nf_conntrack_hash
;
1201 nf_conntrack_htable_size
= hashsize
;
1202 nf_conntrack_vmalloc
= vmalloced
;
1203 nf_conntrack_hash
= hash
;
1204 nf_conntrack_hash_rnd
= rnd
;
1205 write_unlock_bh(&nf_conntrack_lock
);
1207 free_conntrack_hash(old_hash
, old_vmalloced
, old_size
);
1211 module_param_call(hashsize
, set_hashsize
, param_get_uint
,
1212 &nf_conntrack_htable_size
, 0600);
1214 int __init
nf_conntrack_init(void)
1218 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1219 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */
1220 if (!nf_conntrack_htable_size
) {
1221 nf_conntrack_htable_size
1222 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1223 / sizeof(struct list_head
));
1224 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1225 nf_conntrack_htable_size
= 8192;
1226 if (nf_conntrack_htable_size
< 16)
1227 nf_conntrack_htable_size
= 16;
1229 nf_conntrack_max
= 8 * nf_conntrack_htable_size
;
1231 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1232 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1235 nf_conntrack_hash
= alloc_hashtable(nf_conntrack_htable_size
,
1236 &nf_conntrack_vmalloc
);
1237 if (!nf_conntrack_hash
) {
1238 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1242 ret
= nf_conntrack_register_cache(NF_CT_F_BASIC
, "nf_conntrack:basic",
1243 sizeof(struct nf_conn
));
1245 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1249 nf_conntrack_expect_cachep
= kmem_cache_create("nf_conntrack_expect",
1250 sizeof(struct nf_conntrack_expect
),
1252 if (!nf_conntrack_expect_cachep
) {
1253 printk(KERN_ERR
"Unable to create nf_expect slab cache\n");
1254 goto err_free_conntrack_slab
;
1257 ret
= nf_conntrack_proto_init();
1259 goto out_free_expect_slab
;
1261 /* For use by REJECT target */
1262 rcu_assign_pointer(ip_ct_attach
, __nf_conntrack_attach
);
1263 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1265 /* Set up fake conntrack:
1266 - to never be deleted, not in any hashes */
1267 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1268 /* - and look it like as a confirmed connection */
1269 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1273 out_free_expect_slab
:
1274 kmem_cache_destroy(nf_conntrack_expect_cachep
);
1275 err_free_conntrack_slab
:
1276 nf_conntrack_unregister_cache(NF_CT_F_BASIC
);
1278 free_conntrack_hash(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1279 nf_conntrack_htable_size
);