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 */
335 l3proto
= __nf_ct_l3proto_find(ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.src
.l3num
);
336 if (l3proto
&& l3proto
->destroy
)
337 l3proto
->destroy(ct
);
339 l4proto
= __nf_ct_l4proto_find(ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.src
.l3num
, ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.dst
.protonum
);
340 if (l4proto
&& l4proto
->destroy
)
341 l4proto
->destroy(ct
);
343 if (nf_conntrack_destroyed
)
344 nf_conntrack_destroyed(ct
);
346 write_lock_bh(&nf_conntrack_lock
);
347 /* Expectations will have been removed in clean_from_lists,
348 * except TFTP can create an expectation on the first packet,
349 * before connection is in the list, so we need to clean here,
351 nf_ct_remove_expectations(ct
);
353 /* We overload first tuple to link into unconfirmed list. */
354 if (!nf_ct_is_confirmed(ct
)) {
355 BUG_ON(list_empty(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
));
356 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
359 NF_CT_STAT_INC(delete);
360 write_unlock_bh(&nf_conntrack_lock
);
363 nf_ct_put(ct
->master
);
365 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct
);
366 nf_conntrack_free(ct
);
369 static void death_by_timeout(unsigned long ul_conntrack
)
371 struct nf_conn
*ct
= (void *)ul_conntrack
;
373 write_lock_bh(&nf_conntrack_lock
);
374 /* Inside lock so preempt is disabled on module removal path.
375 * Otherwise we can get spurious warnings. */
376 NF_CT_STAT_INC(delete_list
);
377 clean_from_lists(ct
);
378 write_unlock_bh(&nf_conntrack_lock
);
382 struct nf_conntrack_tuple_hash
*
383 __nf_conntrack_find(const struct nf_conntrack_tuple
*tuple
,
384 const struct nf_conn
*ignored_conntrack
)
386 struct nf_conntrack_tuple_hash
*h
;
387 unsigned int hash
= hash_conntrack(tuple
);
389 list_for_each_entry(h
, &nf_conntrack_hash
[hash
], list
) {
390 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
391 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
392 NF_CT_STAT_INC(found
);
395 NF_CT_STAT_INC(searched
);
400 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
402 /* Find a connection corresponding to a tuple. */
403 struct nf_conntrack_tuple_hash
*
404 nf_conntrack_find_get(const struct nf_conntrack_tuple
*tuple
,
405 const struct nf_conn
*ignored_conntrack
)
407 struct nf_conntrack_tuple_hash
*h
;
409 read_lock_bh(&nf_conntrack_lock
);
410 h
= __nf_conntrack_find(tuple
, ignored_conntrack
);
412 atomic_inc(&nf_ct_tuplehash_to_ctrack(h
)->ct_general
.use
);
413 read_unlock_bh(&nf_conntrack_lock
);
417 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
419 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
421 unsigned int repl_hash
)
423 ct
->id
= ++nf_conntrack_next_id
;
424 list_add(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
,
425 &nf_conntrack_hash
[hash
]);
426 list_add(&ct
->tuplehash
[IP_CT_DIR_REPLY
].list
,
427 &nf_conntrack_hash
[repl_hash
]);
430 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
432 unsigned int hash
, repl_hash
;
434 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
435 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
437 write_lock_bh(&nf_conntrack_lock
);
438 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
439 write_unlock_bh(&nf_conntrack_lock
);
441 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
443 /* Confirm a connection given skb; places it in hash table */
445 __nf_conntrack_confirm(struct sk_buff
**pskb
)
447 unsigned int hash
, repl_hash
;
448 struct nf_conntrack_tuple_hash
*h
;
450 struct nf_conn_help
*help
;
451 enum ip_conntrack_info ctinfo
;
453 ct
= nf_ct_get(*pskb
, &ctinfo
);
455 /* ipt_REJECT uses nf_conntrack_attach to attach related
456 ICMP/TCP RST packets in other direction. Actual packet
457 which created connection will be IP_CT_NEW or for an
458 expected connection, IP_CT_RELATED. */
459 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
462 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
463 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
465 /* We're not in hash table, and we refuse to set up related
466 connections for unconfirmed conns. But packet copies and
467 REJECT will give spurious warnings here. */
468 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
470 /* No external references means noone else could have
472 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
473 DEBUGP("Confirming conntrack %p\n", ct
);
475 write_lock_bh(&nf_conntrack_lock
);
477 /* See if there's one in the list already, including reverse:
478 NAT could have grabbed it without realizing, since we're
479 not in the hash. If there is, we lost race. */
480 list_for_each_entry(h
, &nf_conntrack_hash
[hash
], list
)
481 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
484 list_for_each_entry(h
, &nf_conntrack_hash
[repl_hash
], list
)
485 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
489 /* Remove from unconfirmed list */
490 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
492 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
493 /* Timer relative to confirmation time, not original
494 setting time, otherwise we'd get timer wrap in
495 weird delay cases. */
496 ct
->timeout
.expires
+= jiffies
;
497 add_timer(&ct
->timeout
);
498 atomic_inc(&ct
->ct_general
.use
);
499 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
500 NF_CT_STAT_INC(insert
);
501 write_unlock_bh(&nf_conntrack_lock
);
502 help
= nfct_help(ct
);
503 if (help
&& help
->helper
)
504 nf_conntrack_event_cache(IPCT_HELPER
, *pskb
);
505 #ifdef CONFIG_NF_NAT_NEEDED
506 if (test_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
) ||
507 test_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
))
508 nf_conntrack_event_cache(IPCT_NATINFO
, *pskb
);
510 nf_conntrack_event_cache(master_ct(ct
) ?
511 IPCT_RELATED
: IPCT_NEW
, *pskb
);
515 NF_CT_STAT_INC(insert_failed
);
516 write_unlock_bh(&nf_conntrack_lock
);
519 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
521 /* Returns true if a connection correspondings to the tuple (required
524 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
525 const struct nf_conn
*ignored_conntrack
)
527 struct nf_conntrack_tuple_hash
*h
;
529 read_lock_bh(&nf_conntrack_lock
);
530 h
= __nf_conntrack_find(tuple
, ignored_conntrack
);
531 read_unlock_bh(&nf_conntrack_lock
);
535 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
537 /* There's a small race here where we may free a just-assured
538 connection. Too bad: we're in trouble anyway. */
539 static int early_drop(struct list_head
*chain
)
541 /* Traverse backwards: gives us oldest, which is roughly LRU */
542 struct nf_conntrack_tuple_hash
*h
;
543 struct nf_conn
*ct
= NULL
, *tmp
;
546 read_lock_bh(&nf_conntrack_lock
);
547 list_for_each_entry_reverse(h
, chain
, list
) {
548 tmp
= nf_ct_tuplehash_to_ctrack(h
);
549 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
)) {
551 atomic_inc(&ct
->ct_general
.use
);
555 read_unlock_bh(&nf_conntrack_lock
);
560 if (del_timer(&ct
->timeout
)) {
561 death_by_timeout((unsigned long)ct
);
563 NF_CT_STAT_INC(early_drop
);
569 static struct nf_conn
*
570 __nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
571 const struct nf_conntrack_tuple
*repl
,
572 const struct nf_conntrack_l3proto
*l3proto
,
575 struct nf_conn
*conntrack
= NULL
;
576 struct nf_conntrack_helper
*helper
;
578 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
579 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
580 nf_conntrack_hash_rnd_initted
= 1;
583 /* We don't want any race condition at early drop stage */
584 atomic_inc(&nf_conntrack_count
);
587 && atomic_read(&nf_conntrack_count
) > nf_conntrack_max
) {
588 unsigned int hash
= hash_conntrack(orig
);
589 /* Try dropping from this hash chain. */
590 if (!early_drop(&nf_conntrack_hash
[hash
])) {
591 atomic_dec(&nf_conntrack_count
);
594 "nf_conntrack: table full, dropping"
596 return ERR_PTR(-ENOMEM
);
600 /* find features needed by this conntrack. */
601 features
|= l3proto
->get_features(orig
);
603 /* FIXME: protect helper list per RCU */
604 read_lock_bh(&nf_conntrack_lock
);
605 helper
= __nf_ct_helper_find(repl
);
606 /* NAT might want to assign a helper later */
607 if (helper
|| features
& NF_CT_F_NAT
)
608 features
|= NF_CT_F_HELP
;
609 read_unlock_bh(&nf_conntrack_lock
);
611 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features
);
613 read_lock_bh(&nf_ct_cache_lock
);
615 if (unlikely(!nf_ct_cache
[features
].use
)) {
616 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n",
621 conntrack
= kmem_cache_alloc(nf_ct_cache
[features
].cachep
, GFP_ATOMIC
);
622 if (conntrack
== NULL
) {
623 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n");
627 memset(conntrack
, 0, nf_ct_cache
[features
].size
);
628 conntrack
->features
= features
;
629 atomic_set(&conntrack
->ct_general
.use
, 1);
630 conntrack
->ct_general
.destroy
= destroy_conntrack
;
631 conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
632 conntrack
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
633 /* Don't set timer yet: wait for confirmation */
634 init_timer(&conntrack
->timeout
);
635 conntrack
->timeout
.data
= (unsigned long)conntrack
;
636 conntrack
->timeout
.function
= death_by_timeout
;
637 read_unlock_bh(&nf_ct_cache_lock
);
641 read_unlock_bh(&nf_ct_cache_lock
);
642 atomic_dec(&nf_conntrack_count
);
646 struct nf_conn
*nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
647 const struct nf_conntrack_tuple
*repl
)
649 struct nf_conntrack_l3proto
*l3proto
;
651 l3proto
= __nf_ct_l3proto_find(orig
->src
.l3num
);
652 return __nf_conntrack_alloc(orig
, repl
, l3proto
, 0);
654 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
656 void nf_conntrack_free(struct nf_conn
*conntrack
)
658 u_int32_t features
= conntrack
->features
;
659 NF_CT_ASSERT(features
>= NF_CT_F_BASIC
&& features
< NF_CT_F_NUM
);
660 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features
,
662 kmem_cache_free(nf_ct_cache
[features
].cachep
, conntrack
);
663 atomic_dec(&nf_conntrack_count
);
665 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
667 /* Allocate a new conntrack: we return -ENOMEM if classification
668 failed due to stress. Otherwise it really is unclassifiable. */
669 static struct nf_conntrack_tuple_hash
*
670 init_conntrack(const struct nf_conntrack_tuple
*tuple
,
671 struct nf_conntrack_l3proto
*l3proto
,
672 struct nf_conntrack_l4proto
*l4proto
,
674 unsigned int dataoff
)
676 struct nf_conn
*conntrack
;
677 struct nf_conntrack_tuple repl_tuple
;
678 struct nf_conntrack_expect
*exp
;
679 u_int32_t features
= 0;
681 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
682 DEBUGP("Can't invert tuple.\n");
686 read_lock_bh(&nf_conntrack_lock
);
687 exp
= __nf_conntrack_expect_find(tuple
);
688 if (exp
&& exp
->helper
)
689 features
= NF_CT_F_HELP
;
690 read_unlock_bh(&nf_conntrack_lock
);
692 conntrack
= __nf_conntrack_alloc(tuple
, &repl_tuple
, l3proto
, features
);
693 if (conntrack
== NULL
|| IS_ERR(conntrack
)) {
694 DEBUGP("Can't allocate conntrack.\n");
695 return (struct nf_conntrack_tuple_hash
*)conntrack
;
698 if (!l4proto
->new(conntrack
, skb
, dataoff
)) {
699 nf_conntrack_free(conntrack
);
700 DEBUGP("init conntrack: can't track with proto module\n");
704 write_lock_bh(&nf_conntrack_lock
);
705 exp
= find_expectation(tuple
);
708 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
710 /* Welcome, Mr. Bond. We've been expecting you... */
711 __set_bit(IPS_EXPECTED_BIT
, &conntrack
->status
);
712 conntrack
->master
= exp
->master
;
714 nfct_help(conntrack
)->helper
= exp
->helper
;
715 #ifdef CONFIG_NF_CONNTRACK_MARK
716 conntrack
->mark
= exp
->master
->mark
;
718 #ifdef CONFIG_NF_CONNTRACK_SECMARK
719 conntrack
->secmark
= exp
->master
->secmark
;
721 nf_conntrack_get(&conntrack
->master
->ct_general
);
722 NF_CT_STAT_INC(expect_new
);
724 struct nf_conn_help
*help
= nfct_help(conntrack
);
727 help
->helper
= __nf_ct_helper_find(&repl_tuple
);
731 /* Overload tuple linked list to put us in unconfirmed list. */
732 list_add(&conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
, &unconfirmed
);
734 write_unlock_bh(&nf_conntrack_lock
);
738 exp
->expectfn(conntrack
, exp
);
739 nf_conntrack_expect_put(exp
);
742 return &conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
];
745 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
746 static inline struct nf_conn
*
747 resolve_normal_ct(struct sk_buff
*skb
,
748 unsigned int dataoff
,
751 struct nf_conntrack_l3proto
*l3proto
,
752 struct nf_conntrack_l4proto
*l4proto
,
754 enum ip_conntrack_info
*ctinfo
)
756 struct nf_conntrack_tuple tuple
;
757 struct nf_conntrack_tuple_hash
*h
;
760 if (!nf_ct_get_tuple(skb
, (unsigned int)(skb
->nh
.raw
- skb
->data
),
761 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
763 DEBUGP("resolve_normal_ct: Can't get tuple\n");
767 /* look for tuple match */
768 h
= nf_conntrack_find_get(&tuple
, NULL
);
770 h
= init_conntrack(&tuple
, l3proto
, l4proto
, skb
, dataoff
);
776 ct
= nf_ct_tuplehash_to_ctrack(h
);
778 /* It exists; we have (non-exclusive) reference. */
779 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
780 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
781 /* Please set reply bit if this packet OK */
784 /* Once we've had two way comms, always ESTABLISHED. */
785 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
786 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct
);
787 *ctinfo
= IP_CT_ESTABLISHED
;
788 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
789 DEBUGP("nf_conntrack_in: related packet for %p\n", ct
);
790 *ctinfo
= IP_CT_RELATED
;
792 DEBUGP("nf_conntrack_in: new packet for %p\n", ct
);
797 skb
->nfct
= &ct
->ct_general
;
798 skb
->nfctinfo
= *ctinfo
;
803 nf_conntrack_in(int pf
, unsigned int hooknum
, struct sk_buff
**pskb
)
806 enum ip_conntrack_info ctinfo
;
807 struct nf_conntrack_l3proto
*l3proto
;
808 struct nf_conntrack_l4proto
*l4proto
;
809 unsigned int dataoff
;
814 /* Previously seen (loopback or untracked)? Ignore. */
816 NF_CT_STAT_INC(ignore
);
820 l3proto
= __nf_ct_l3proto_find((u_int16_t
)pf
);
821 if ((ret
= l3proto
->prepare(pskb
, hooknum
, &dataoff
, &protonum
)) <= 0) {
822 DEBUGP("not prepared to track yet or error occured\n");
826 l4proto
= __nf_ct_l4proto_find((u_int16_t
)pf
, protonum
);
828 /* It may be an special packet, error, unclean...
829 * inverse of the return code tells to the netfilter
830 * core what to do with the packet. */
831 if (l4proto
->error
!= NULL
&&
832 (ret
= l4proto
->error(*pskb
, dataoff
, &ctinfo
, pf
, hooknum
)) <= 0) {
833 NF_CT_STAT_INC(error
);
834 NF_CT_STAT_INC(invalid
);
838 ct
= resolve_normal_ct(*pskb
, dataoff
, pf
, protonum
, l3proto
, l4proto
,
839 &set_reply
, &ctinfo
);
841 /* Not valid part of a connection */
842 NF_CT_STAT_INC(invalid
);
847 /* Too stressed to deal. */
848 NF_CT_STAT_INC(drop
);
852 NF_CT_ASSERT((*pskb
)->nfct
);
854 ret
= l4proto
->packet(ct
, *pskb
, dataoff
, ctinfo
, pf
, hooknum
);
856 /* Invalid: inverse of the return code tells
857 * the netfilter core what to do */
858 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
859 nf_conntrack_put((*pskb
)->nfct
);
860 (*pskb
)->nfct
= NULL
;
861 NF_CT_STAT_INC(invalid
);
865 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
866 nf_conntrack_event_cache(IPCT_STATUS
, *pskb
);
870 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
872 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
873 const struct nf_conntrack_tuple
*orig
)
875 return nf_ct_invert_tuple(inverse
, orig
,
876 __nf_ct_l3proto_find(orig
->src
.l3num
),
877 __nf_ct_l4proto_find(orig
->src
.l3num
,
878 orig
->dst
.protonum
));
880 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
882 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
883 implicitly racy: see __nf_conntrack_confirm */
884 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
885 const struct nf_conntrack_tuple
*newreply
)
887 struct nf_conn_help
*help
= nfct_help(ct
);
889 write_lock_bh(&nf_conntrack_lock
);
890 /* Should be unconfirmed, so not in hash table yet */
891 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
893 DEBUGP("Altering reply tuple of %p to ", ct
);
894 NF_CT_DUMP_TUPLE(newreply
);
896 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
897 if (!ct
->master
&& help
&& help
->expecting
== 0)
898 help
->helper
= __nf_ct_helper_find(newreply
);
899 write_unlock_bh(&nf_conntrack_lock
);
901 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
903 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
904 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
905 enum ip_conntrack_info ctinfo
,
906 const struct sk_buff
*skb
,
907 unsigned long extra_jiffies
,
912 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
915 write_lock_bh(&nf_conntrack_lock
);
917 /* Only update if this is not a fixed timeout */
918 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
)) {
919 write_unlock_bh(&nf_conntrack_lock
);
923 /* If not in hash table, timer will not be active yet */
924 if (!nf_ct_is_confirmed(ct
)) {
925 ct
->timeout
.expires
= extra_jiffies
;
926 event
= IPCT_REFRESH
;
928 unsigned long newtime
= jiffies
+ extra_jiffies
;
930 /* Only update the timeout if the new timeout is at least
931 HZ jiffies from the old timeout. Need del_timer for race
932 avoidance (may already be dying). */
933 if (newtime
- ct
->timeout
.expires
>= HZ
934 && del_timer(&ct
->timeout
)) {
935 ct
->timeout
.expires
= newtime
;
936 add_timer(&ct
->timeout
);
937 event
= IPCT_REFRESH
;
941 #ifdef CONFIG_NF_CT_ACCT
943 ct
->counters
[CTINFO2DIR(ctinfo
)].packets
++;
944 ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
+=
945 skb
->len
- (unsigned int)(skb
->nh
.raw
- skb
->data
);
947 if ((ct
->counters
[CTINFO2DIR(ctinfo
)].packets
& 0x80000000)
948 || (ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
& 0x80000000))
949 event
|= IPCT_COUNTER_FILLING
;
953 write_unlock_bh(&nf_conntrack_lock
);
955 /* must be unlocked when calling event cache */
957 nf_conntrack_event_cache(event
, skb
);
959 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
961 #if defined(CONFIG_NF_CT_NETLINK) || \
962 defined(CONFIG_NF_CT_NETLINK_MODULE)
964 #include <linux/netfilter/nfnetlink.h>
965 #include <linux/netfilter/nfnetlink_conntrack.h>
966 #include <linux/mutex.h>
969 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
970 * in ip_conntrack_core, since we don't want the protocols to autoload
971 * or depend on ctnetlink */
972 int nf_ct_port_tuple_to_nfattr(struct sk_buff
*skb
,
973 const struct nf_conntrack_tuple
*tuple
)
975 NFA_PUT(skb
, CTA_PROTO_SRC_PORT
, sizeof(u_int16_t
),
976 &tuple
->src
.u
.tcp
.port
);
977 NFA_PUT(skb
, CTA_PROTO_DST_PORT
, sizeof(u_int16_t
),
978 &tuple
->dst
.u
.tcp
.port
);
984 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr
);
986 static const size_t cta_min_proto
[CTA_PROTO_MAX
] = {
987 [CTA_PROTO_SRC_PORT
-1] = sizeof(u_int16_t
),
988 [CTA_PROTO_DST_PORT
-1] = sizeof(u_int16_t
)
991 int nf_ct_port_nfattr_to_tuple(struct nfattr
*tb
[],
992 struct nf_conntrack_tuple
*t
)
994 if (!tb
[CTA_PROTO_SRC_PORT
-1] || !tb
[CTA_PROTO_DST_PORT
-1])
997 if (nfattr_bad_size(tb
, CTA_PROTO_MAX
, cta_min_proto
))
1000 t
->src
.u
.tcp
.port
= *(__be16
*)NFA_DATA(tb
[CTA_PROTO_SRC_PORT
-1]);
1001 t
->dst
.u
.tcp
.port
= *(__be16
*)NFA_DATA(tb
[CTA_PROTO_DST_PORT
-1]);
1005 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple
);
1008 /* Used by ipt_REJECT and ip6t_REJECT. */
1009 void __nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
1012 enum ip_conntrack_info ctinfo
;
1014 /* This ICMP is in reverse direction to the packet which caused it */
1015 ct
= nf_ct_get(skb
, &ctinfo
);
1016 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1017 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
1019 ctinfo
= IP_CT_RELATED
;
1021 /* Attach to new skbuff, and increment count */
1022 nskb
->nfct
= &ct
->ct_general
;
1023 nskb
->nfctinfo
= ctinfo
;
1024 nf_conntrack_get(nskb
->nfct
);
1026 EXPORT_SYMBOL_GPL(__nf_conntrack_attach
);
1029 do_iter(const struct nf_conntrack_tuple_hash
*i
,
1030 int (*iter
)(struct nf_conn
*i
, void *data
),
1033 return iter(nf_ct_tuplehash_to_ctrack(i
), data
);
1036 /* Bring out ya dead! */
1037 static struct nf_conn
*
1038 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
1039 void *data
, unsigned int *bucket
)
1041 struct nf_conntrack_tuple_hash
*h
;
1044 write_lock_bh(&nf_conntrack_lock
);
1045 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1046 list_for_each_entry(h
, &nf_conntrack_hash
[*bucket
], list
) {
1047 ct
= nf_ct_tuplehash_to_ctrack(h
);
1052 list_for_each_entry(h
, &unconfirmed
, list
) {
1053 ct
= nf_ct_tuplehash_to_ctrack(h
);
1057 write_unlock_bh(&nf_conntrack_lock
);
1060 atomic_inc(&ct
->ct_general
.use
);
1061 write_unlock_bh(&nf_conntrack_lock
);
1066 nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
1069 unsigned int bucket
= 0;
1071 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
1072 /* Time to push up daises... */
1073 if (del_timer(&ct
->timeout
))
1074 death_by_timeout((unsigned long)ct
);
1075 /* ... else the timer will get him soon. */
1080 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1082 static int kill_all(struct nf_conn
*i
, void *data
)
1087 static void free_conntrack_hash(struct list_head
*hash
, int vmalloced
, int size
)
1092 free_pages((unsigned long)hash
,
1093 get_order(sizeof(struct list_head
) * size
));
1096 void nf_conntrack_flush(void)
1098 nf_ct_iterate_cleanup(kill_all
, NULL
);
1100 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
1102 /* Mishearing the voices in his head, our hero wonders how he's
1103 supposed to kill the mall. */
1104 void nf_conntrack_cleanup(void)
1108 rcu_assign_pointer(ip_ct_attach
, NULL
);
1110 /* This makes sure all current packets have passed through
1111 netfilter framework. Roll on, two-stage module
1115 nf_ct_event_cache_flush();
1117 nf_conntrack_flush();
1118 if (atomic_read(&nf_conntrack_count
) != 0) {
1120 goto i_see_dead_people
;
1122 /* wait until all references to nf_conntrack_untracked are dropped */
1123 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
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_l4proto_unregister(&nf_conntrack_l4proto_generic
);
1140 /* free l3proto protocol tables */
1141 for (i
= 0; i
< PF_MAX
; i
++)
1142 if (nf_ct_protos
[i
]) {
1143 kfree(nf_ct_protos
[i
]);
1144 nf_ct_protos
[i
] = NULL
;
1148 static struct list_head
*alloc_hashtable(int size
, int *vmalloced
)
1150 struct list_head
*hash
;
1154 hash
= (void*)__get_free_pages(GFP_KERNEL
,
1155 get_order(sizeof(struct list_head
)
1159 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1160 hash
= vmalloc(sizeof(struct list_head
) * size
);
1164 for (i
= 0; i
< size
; i
++)
1165 INIT_LIST_HEAD(&hash
[i
]);
1170 int set_hashsize(const char *val
, struct kernel_param
*kp
)
1172 int i
, bucket
, hashsize
, vmalloced
;
1173 int old_vmalloced
, old_size
;
1175 struct list_head
*hash
, *old_hash
;
1176 struct nf_conntrack_tuple_hash
*h
;
1178 /* On boot, we can set this without any fancy locking. */
1179 if (!nf_conntrack_htable_size
)
1180 return param_set_uint(val
, kp
);
1182 hashsize
= simple_strtol(val
, NULL
, 0);
1186 hash
= alloc_hashtable(hashsize
, &vmalloced
);
1190 /* We have to rehahs for the new table anyway, so we also can
1191 * use a newrandom seed */
1192 get_random_bytes(&rnd
, 4);
1194 write_lock_bh(&nf_conntrack_lock
);
1195 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1196 while (!list_empty(&nf_conntrack_hash
[i
])) {
1197 h
= list_entry(nf_conntrack_hash
[i
].next
,
1198 struct nf_conntrack_tuple_hash
, list
);
1200 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1201 list_add_tail(&h
->list
, &hash
[bucket
]);
1204 old_size
= nf_conntrack_htable_size
;
1205 old_vmalloced
= nf_conntrack_vmalloc
;
1206 old_hash
= nf_conntrack_hash
;
1208 nf_conntrack_htable_size
= hashsize
;
1209 nf_conntrack_vmalloc
= vmalloced
;
1210 nf_conntrack_hash
= hash
;
1211 nf_conntrack_hash_rnd
= rnd
;
1212 write_unlock_bh(&nf_conntrack_lock
);
1214 free_conntrack_hash(old_hash
, old_vmalloced
, old_size
);
1218 module_param_call(hashsize
, set_hashsize
, param_get_uint
,
1219 &nf_conntrack_htable_size
, 0600);
1221 int __init
nf_conntrack_init(void)
1226 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1227 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */
1228 if (!nf_conntrack_htable_size
) {
1229 nf_conntrack_htable_size
1230 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1231 / sizeof(struct list_head
));
1232 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1233 nf_conntrack_htable_size
= 8192;
1234 if (nf_conntrack_htable_size
< 16)
1235 nf_conntrack_htable_size
= 16;
1237 nf_conntrack_max
= 8 * nf_conntrack_htable_size
;
1239 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1240 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1243 nf_conntrack_hash
= alloc_hashtable(nf_conntrack_htable_size
,
1244 &nf_conntrack_vmalloc
);
1245 if (!nf_conntrack_hash
) {
1246 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1250 ret
= nf_conntrack_register_cache(NF_CT_F_BASIC
, "nf_conntrack:basic",
1251 sizeof(struct nf_conn
));
1253 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1257 nf_conntrack_expect_cachep
= kmem_cache_create("nf_conntrack_expect",
1258 sizeof(struct nf_conntrack_expect
),
1260 if (!nf_conntrack_expect_cachep
) {
1261 printk(KERN_ERR
"Unable to create nf_expect slab cache\n");
1262 goto err_free_conntrack_slab
;
1265 ret
= nf_conntrack_l4proto_register(&nf_conntrack_l4proto_generic
);
1267 goto out_free_expect_slab
;
1269 /* Don't NEED lock here, but good form anyway. */
1270 write_lock_bh(&nf_conntrack_lock
);
1271 for (i
= 0; i
< AF_MAX
; i
++)
1272 nf_ct_l3protos
[i
] = &nf_conntrack_l3proto_generic
;
1273 write_unlock_bh(&nf_conntrack_lock
);
1275 /* For use by REJECT target */
1276 rcu_assign_pointer(ip_ct_attach
, __nf_conntrack_attach
);
1278 /* Set up fake conntrack:
1279 - to never be deleted, not in any hashes */
1280 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1281 /* - and look it like as a confirmed connection */
1282 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1286 out_free_expect_slab
:
1287 kmem_cache_destroy(nf_conntrack_expect_cachep
);
1288 err_free_conntrack_slab
:
1289 nf_conntrack_unregister_cache(NF_CT_F_BASIC
);
1291 free_conntrack_hash(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1292 nf_conntrack_htable_size
);