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>
50 #include <net/netfilter/nf_conntrack.h>
51 #include <net/netfilter/nf_conntrack_l3proto.h>
52 #include <net/netfilter/nf_conntrack_l4proto.h>
53 #include <net/netfilter/nf_conntrack_expect.h>
54 #include <net/netfilter/nf_conntrack_helper.h>
55 #include <net/netfilter/nf_conntrack_core.h>
57 #define NF_CONNTRACK_VERSION "0.5.0"
62 #define DEBUGP(format, args...)
65 DEFINE_RWLOCK(nf_conntrack_lock
);
67 /* nf_conntrack_standalone needs this */
68 atomic_t nf_conntrack_count
= ATOMIC_INIT(0);
69 EXPORT_SYMBOL_GPL(nf_conntrack_count
);
71 void (*nf_conntrack_destroyed
)(struct nf_conn
*conntrack
) = NULL
;
72 unsigned int nf_conntrack_htable_size __read_mostly
;
73 int nf_conntrack_max __read_mostly
;
74 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
75 struct list_head
*nf_conntrack_hash __read_mostly
;
76 struct nf_conn nf_conntrack_untracked __read_mostly
;
77 unsigned int nf_ct_log_invalid __read_mostly
;
78 LIST_HEAD(unconfirmed
);
79 static int nf_conntrack_vmalloc __read_mostly
;
81 static unsigned int nf_conntrack_next_id
;
83 DEFINE_PER_CPU(struct ip_conntrack_stat
, nf_conntrack_stat
);
84 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat
);
87 * This scheme offers various size of "struct nf_conn" dependent on
88 * features(helper, nat, ...)
91 #define NF_CT_FEATURES_NAMELEN 256
93 /* name of slab cache. printed in /proc/slabinfo */
96 /* size of slab cache */
99 /* slab cache pointer */
100 kmem_cache_t
*cachep
;
102 /* allocated slab cache + modules which uses this slab cache */
105 } nf_ct_cache
[NF_CT_F_NUM
];
107 /* protect members of nf_ct_cache except of "use" */
108 DEFINE_RWLOCK(nf_ct_cache_lock
);
110 /* This avoids calling kmem_cache_create() with same name simultaneously */
111 static DEFINE_MUTEX(nf_ct_cache_mutex
);
113 static int nf_conntrack_hash_rnd_initted
;
114 static unsigned int nf_conntrack_hash_rnd
;
116 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
117 unsigned int size
, unsigned int rnd
)
120 a
= jhash((void *)tuple
->src
.u3
.all
, sizeof(tuple
->src
.u3
.all
),
121 ((tuple
->src
.l3num
) << 16) | tuple
->dst
.protonum
);
122 b
= jhash((void *)tuple
->dst
.u3
.all
, sizeof(tuple
->dst
.u3
.all
),
123 (tuple
->src
.u
.all
<< 16) | tuple
->dst
.u
.all
);
125 return jhash_2words(a
, b
, rnd
) % size
;
128 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
130 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
131 nf_conntrack_hash_rnd
);
134 int nf_conntrack_register_cache(u_int32_t features
, const char *name
,
139 kmem_cache_t
*cachep
;
141 DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n",
142 features
, name
, size
);
144 if (features
< NF_CT_F_BASIC
|| features
>= NF_CT_F_NUM
) {
145 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n",
150 mutex_lock(&nf_ct_cache_mutex
);
152 write_lock_bh(&nf_ct_cache_lock
);
153 /* e.g: multiple helpers are loaded */
154 if (nf_ct_cache
[features
].use
> 0) {
155 DEBUGP("nf_conntrack_register_cache: already resisterd.\n");
156 if ((!strncmp(nf_ct_cache
[features
].name
, name
,
157 NF_CT_FEATURES_NAMELEN
))
158 && nf_ct_cache
[features
].size
== size
) {
159 DEBUGP("nf_conntrack_register_cache: reusing.\n");
160 nf_ct_cache
[features
].use
++;
165 write_unlock_bh(&nf_ct_cache_lock
);
166 mutex_unlock(&nf_ct_cache_mutex
);
169 write_unlock_bh(&nf_ct_cache_lock
);
172 * The memory space for name of slab cache must be alive until
173 * cache is destroyed.
175 cache_name
= kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN
, GFP_ATOMIC
);
176 if (cache_name
== NULL
) {
177 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n");
182 if (strlcpy(cache_name
, name
, NF_CT_FEATURES_NAMELEN
)
183 >= NF_CT_FEATURES_NAMELEN
) {
184 printk("nf_conntrack_register_cache: name too long\n");
189 cachep
= kmem_cache_create(cache_name
, size
, 0, 0,
192 printk("nf_conntrack_register_cache: Can't create slab cache "
193 "for the features = 0x%x\n", features
);
198 write_lock_bh(&nf_ct_cache_lock
);
199 nf_ct_cache
[features
].use
= 1;
200 nf_ct_cache
[features
].size
= size
;
201 nf_ct_cache
[features
].cachep
= cachep
;
202 nf_ct_cache
[features
].name
= cache_name
;
203 write_unlock_bh(&nf_ct_cache_lock
);
210 mutex_unlock(&nf_ct_cache_mutex
);
214 /* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */
215 void nf_conntrack_unregister_cache(u_int32_t features
)
217 kmem_cache_t
*cachep
;
221 * This assures that kmem_cache_create() isn't called before destroying
224 DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features
);
225 mutex_lock(&nf_ct_cache_mutex
);
227 write_lock_bh(&nf_ct_cache_lock
);
228 if (--nf_ct_cache
[features
].use
> 0) {
229 write_unlock_bh(&nf_ct_cache_lock
);
230 mutex_unlock(&nf_ct_cache_mutex
);
233 cachep
= nf_ct_cache
[features
].cachep
;
234 name
= nf_ct_cache
[features
].name
;
235 nf_ct_cache
[features
].cachep
= NULL
;
236 nf_ct_cache
[features
].name
= NULL
;
237 nf_ct_cache
[features
].size
= 0;
238 write_unlock_bh(&nf_ct_cache_lock
);
242 kmem_cache_destroy(cachep
);
245 mutex_unlock(&nf_ct_cache_mutex
);
249 nf_ct_get_tuple(const struct sk_buff
*skb
,
251 unsigned int dataoff
,
254 struct nf_conntrack_tuple
*tuple
,
255 const struct nf_conntrack_l3proto
*l3proto
,
256 const struct nf_conntrack_l4proto
*l4proto
)
258 NF_CT_TUPLE_U_BLANK(tuple
);
260 tuple
->src
.l3num
= l3num
;
261 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
264 tuple
->dst
.protonum
= protonum
;
265 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
267 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
271 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
272 const struct nf_conntrack_tuple
*orig
,
273 const struct nf_conntrack_l3proto
*l3proto
,
274 const struct nf_conntrack_l4proto
*l4proto
)
276 NF_CT_TUPLE_U_BLANK(inverse
);
278 inverse
->src
.l3num
= orig
->src
.l3num
;
279 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
282 inverse
->dst
.dir
= !orig
->dst
.dir
;
284 inverse
->dst
.protonum
= orig
->dst
.protonum
;
285 return l4proto
->invert_tuple(inverse
, orig
);
289 clean_from_lists(struct nf_conn
*ct
)
291 DEBUGP("clean_from_lists(%p)\n", ct
);
292 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
293 list_del(&ct
->tuplehash
[IP_CT_DIR_REPLY
].list
);
295 /* Destroy all pending expectations */
296 nf_ct_remove_expectations(ct
);
300 destroy_conntrack(struct nf_conntrack
*nfct
)
302 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
303 struct nf_conntrack_l3proto
*l3proto
;
304 struct nf_conntrack_l4proto
*l4proto
;
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 /* To make sure we don't get any weird locking issues here:
314 * destroy_conntrack() MUST NOT be called with a write lock
315 * to nf_conntrack_lock!!! -HW */
316 l3proto
= __nf_ct_l3proto_find(ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.src
.l3num
);
317 if (l3proto
&& l3proto
->destroy
)
318 l3proto
->destroy(ct
);
320 l4proto
= __nf_ct_l4proto_find(ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.src
.l3num
, ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.dst
.protonum
);
321 if (l4proto
&& l4proto
->destroy
)
322 l4proto
->destroy(ct
);
324 if (nf_conntrack_destroyed
)
325 nf_conntrack_destroyed(ct
);
327 write_lock_bh(&nf_conntrack_lock
);
328 /* Expectations will have been removed in clean_from_lists,
329 * except TFTP can create an expectation on the first packet,
330 * before connection is in the list, so we need to clean here,
332 nf_ct_remove_expectations(ct
);
334 /* We overload first tuple to link into unconfirmed list. */
335 if (!nf_ct_is_confirmed(ct
)) {
336 BUG_ON(list_empty(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
));
337 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
340 NF_CT_STAT_INC(delete);
341 write_unlock_bh(&nf_conntrack_lock
);
344 nf_ct_put(ct
->master
);
346 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct
);
347 nf_conntrack_free(ct
);
350 static void death_by_timeout(unsigned long ul_conntrack
)
352 struct nf_conn
*ct
= (void *)ul_conntrack
;
354 write_lock_bh(&nf_conntrack_lock
);
355 /* Inside lock so preempt is disabled on module removal path.
356 * Otherwise we can get spurious warnings. */
357 NF_CT_STAT_INC(delete_list
);
358 clean_from_lists(ct
);
359 write_unlock_bh(&nf_conntrack_lock
);
363 struct nf_conntrack_tuple_hash
*
364 __nf_conntrack_find(const struct nf_conntrack_tuple
*tuple
,
365 const struct nf_conn
*ignored_conntrack
)
367 struct nf_conntrack_tuple_hash
*h
;
368 unsigned int hash
= hash_conntrack(tuple
);
370 list_for_each_entry(h
, &nf_conntrack_hash
[hash
], list
) {
371 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
372 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
373 NF_CT_STAT_INC(found
);
376 NF_CT_STAT_INC(searched
);
382 /* Find a connection corresponding to a tuple. */
383 struct nf_conntrack_tuple_hash
*
384 nf_conntrack_find_get(const struct nf_conntrack_tuple
*tuple
,
385 const struct nf_conn
*ignored_conntrack
)
387 struct nf_conntrack_tuple_hash
*h
;
389 read_lock_bh(&nf_conntrack_lock
);
390 h
= __nf_conntrack_find(tuple
, ignored_conntrack
);
392 atomic_inc(&nf_ct_tuplehash_to_ctrack(h
)->ct_general
.use
);
393 read_unlock_bh(&nf_conntrack_lock
);
398 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
400 unsigned int repl_hash
)
402 ct
->id
= ++nf_conntrack_next_id
;
403 list_add(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
,
404 &nf_conntrack_hash
[hash
]);
405 list_add(&ct
->tuplehash
[IP_CT_DIR_REPLY
].list
,
406 &nf_conntrack_hash
[repl_hash
]);
409 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
411 unsigned int hash
, repl_hash
;
413 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
414 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
416 write_lock_bh(&nf_conntrack_lock
);
417 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
418 write_unlock_bh(&nf_conntrack_lock
);
421 /* Confirm a connection given skb; places it in hash table */
423 __nf_conntrack_confirm(struct sk_buff
**pskb
)
425 unsigned int hash
, repl_hash
;
426 struct nf_conntrack_tuple_hash
*h
;
428 struct nf_conn_help
*help
;
429 enum ip_conntrack_info ctinfo
;
431 ct
= nf_ct_get(*pskb
, &ctinfo
);
433 /* ipt_REJECT uses nf_conntrack_attach to attach related
434 ICMP/TCP RST packets in other direction. Actual packet
435 which created connection will be IP_CT_NEW or for an
436 expected connection, IP_CT_RELATED. */
437 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
440 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
441 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
443 /* We're not in hash table, and we refuse to set up related
444 connections for unconfirmed conns. But packet copies and
445 REJECT will give spurious warnings here. */
446 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
448 /* No external references means noone else could have
450 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
451 DEBUGP("Confirming conntrack %p\n", ct
);
453 write_lock_bh(&nf_conntrack_lock
);
455 /* See if there's one in the list already, including reverse:
456 NAT could have grabbed it without realizing, since we're
457 not in the hash. If there is, we lost race. */
458 list_for_each_entry(h
, &nf_conntrack_hash
[hash
], list
)
459 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
462 list_for_each_entry(h
, &nf_conntrack_hash
[repl_hash
], list
)
463 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
467 /* Remove from unconfirmed list */
468 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
470 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
471 /* Timer relative to confirmation time, not original
472 setting time, otherwise we'd get timer wrap in
473 weird delay cases. */
474 ct
->timeout
.expires
+= jiffies
;
475 add_timer(&ct
->timeout
);
476 atomic_inc(&ct
->ct_general
.use
);
477 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
478 NF_CT_STAT_INC(insert
);
479 write_unlock_bh(&nf_conntrack_lock
);
480 help
= nfct_help(ct
);
481 if (help
&& help
->helper
)
482 nf_conntrack_event_cache(IPCT_HELPER
, *pskb
);
483 #ifdef CONFIG_NF_NAT_NEEDED
484 if (test_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
) ||
485 test_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
))
486 nf_conntrack_event_cache(IPCT_NATINFO
, *pskb
);
488 nf_conntrack_event_cache(master_ct(ct
) ?
489 IPCT_RELATED
: IPCT_NEW
, *pskb
);
493 NF_CT_STAT_INC(insert_failed
);
494 write_unlock_bh(&nf_conntrack_lock
);
498 /* Returns true if a connection correspondings to the tuple (required
501 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
502 const struct nf_conn
*ignored_conntrack
)
504 struct nf_conntrack_tuple_hash
*h
;
506 read_lock_bh(&nf_conntrack_lock
);
507 h
= __nf_conntrack_find(tuple
, ignored_conntrack
);
508 read_unlock_bh(&nf_conntrack_lock
);
513 /* There's a small race here where we may free a just-assured
514 connection. Too bad: we're in trouble anyway. */
515 static int early_drop(struct list_head
*chain
)
517 /* Traverse backwards: gives us oldest, which is roughly LRU */
518 struct nf_conntrack_tuple_hash
*h
;
519 struct nf_conn
*ct
= NULL
, *tmp
;
522 read_lock_bh(&nf_conntrack_lock
);
523 list_for_each_entry_reverse(h
, chain
, list
) {
524 tmp
= nf_ct_tuplehash_to_ctrack(h
);
525 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
)) {
527 atomic_inc(&ct
->ct_general
.use
);
531 read_unlock_bh(&nf_conntrack_lock
);
536 if (del_timer(&ct
->timeout
)) {
537 death_by_timeout((unsigned long)ct
);
539 NF_CT_STAT_INC(early_drop
);
545 static struct nf_conn
*
546 __nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
547 const struct nf_conntrack_tuple
*repl
,
548 const struct nf_conntrack_l3proto
*l3proto
)
550 struct nf_conn
*conntrack
= NULL
;
551 u_int32_t features
= 0;
552 struct nf_conntrack_helper
*helper
;
554 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
555 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
556 nf_conntrack_hash_rnd_initted
= 1;
559 /* We don't want any race condition at early drop stage */
560 atomic_inc(&nf_conntrack_count
);
563 && atomic_read(&nf_conntrack_count
) > nf_conntrack_max
) {
564 unsigned int hash
= hash_conntrack(orig
);
565 /* Try dropping from this hash chain. */
566 if (!early_drop(&nf_conntrack_hash
[hash
])) {
567 atomic_dec(&nf_conntrack_count
);
570 "nf_conntrack: table full, dropping"
572 return ERR_PTR(-ENOMEM
);
576 /* find features needed by this conntrack. */
577 features
= l3proto
->get_features(orig
);
579 /* FIXME: protect helper list per RCU */
580 read_lock_bh(&nf_conntrack_lock
);
581 helper
= __nf_ct_helper_find(repl
);
583 features
|= NF_CT_F_HELP
;
584 read_unlock_bh(&nf_conntrack_lock
);
586 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features
);
588 read_lock_bh(&nf_ct_cache_lock
);
590 if (unlikely(!nf_ct_cache
[features
].use
)) {
591 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n",
596 conntrack
= kmem_cache_alloc(nf_ct_cache
[features
].cachep
, GFP_ATOMIC
);
597 if (conntrack
== NULL
) {
598 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n");
602 memset(conntrack
, 0, nf_ct_cache
[features
].size
);
603 conntrack
->features
= features
;
604 atomic_set(&conntrack
->ct_general
.use
, 1);
605 conntrack
->ct_general
.destroy
= destroy_conntrack
;
606 conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
607 conntrack
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
608 /* Don't set timer yet: wait for confirmation */
609 init_timer(&conntrack
->timeout
);
610 conntrack
->timeout
.data
= (unsigned long)conntrack
;
611 conntrack
->timeout
.function
= death_by_timeout
;
612 read_unlock_bh(&nf_ct_cache_lock
);
616 read_unlock_bh(&nf_ct_cache_lock
);
617 atomic_dec(&nf_conntrack_count
);
621 struct nf_conn
*nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
622 const struct nf_conntrack_tuple
*repl
)
624 struct nf_conntrack_l3proto
*l3proto
;
626 l3proto
= __nf_ct_l3proto_find(orig
->src
.l3num
);
627 return __nf_conntrack_alloc(orig
, repl
, l3proto
);
630 void nf_conntrack_free(struct nf_conn
*conntrack
)
632 u_int32_t features
= conntrack
->features
;
633 NF_CT_ASSERT(features
>= NF_CT_F_BASIC
&& features
< NF_CT_F_NUM
);
634 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features
,
636 kmem_cache_free(nf_ct_cache
[features
].cachep
, conntrack
);
637 atomic_dec(&nf_conntrack_count
);
640 /* Allocate a new conntrack: we return -ENOMEM if classification
641 failed due to stress. Otherwise it really is unclassifiable. */
642 static struct nf_conntrack_tuple_hash
*
643 init_conntrack(const struct nf_conntrack_tuple
*tuple
,
644 struct nf_conntrack_l3proto
*l3proto
,
645 struct nf_conntrack_l4proto
*l4proto
,
647 unsigned int dataoff
)
649 struct nf_conn
*conntrack
;
650 struct nf_conntrack_tuple repl_tuple
;
651 struct nf_conntrack_expect
*exp
;
653 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
654 DEBUGP("Can't invert tuple.\n");
658 conntrack
= __nf_conntrack_alloc(tuple
, &repl_tuple
, l3proto
);
659 if (conntrack
== NULL
|| IS_ERR(conntrack
)) {
660 DEBUGP("Can't allocate conntrack.\n");
661 return (struct nf_conntrack_tuple_hash
*)conntrack
;
664 if (!l4proto
->new(conntrack
, skb
, dataoff
)) {
665 nf_conntrack_free(conntrack
);
666 DEBUGP("init conntrack: can't track with proto module\n");
670 write_lock_bh(&nf_conntrack_lock
);
671 exp
= find_expectation(tuple
);
674 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
676 /* Welcome, Mr. Bond. We've been expecting you... */
677 __set_bit(IPS_EXPECTED_BIT
, &conntrack
->status
);
678 conntrack
->master
= exp
->master
;
679 #ifdef CONFIG_NF_CONNTRACK_MARK
680 conntrack
->mark
= exp
->master
->mark
;
682 #ifdef CONFIG_NF_CONNTRACK_SECMARK
683 conntrack
->secmark
= exp
->master
->secmark
;
685 nf_conntrack_get(&conntrack
->master
->ct_general
);
686 NF_CT_STAT_INC(expect_new
);
688 struct nf_conn_help
*help
= nfct_help(conntrack
);
691 help
->helper
= __nf_ct_helper_find(&repl_tuple
);
695 /* Overload tuple linked list to put us in unconfirmed list. */
696 list_add(&conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
, &unconfirmed
);
698 write_unlock_bh(&nf_conntrack_lock
);
702 exp
->expectfn(conntrack
, exp
);
703 nf_conntrack_expect_put(exp
);
706 return &conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
];
709 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
710 static inline struct nf_conn
*
711 resolve_normal_ct(struct sk_buff
*skb
,
712 unsigned int dataoff
,
715 struct nf_conntrack_l3proto
*l3proto
,
716 struct nf_conntrack_l4proto
*l4proto
,
718 enum ip_conntrack_info
*ctinfo
)
720 struct nf_conntrack_tuple tuple
;
721 struct nf_conntrack_tuple_hash
*h
;
724 if (!nf_ct_get_tuple(skb
, (unsigned int)(skb
->nh
.raw
- skb
->data
),
725 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
727 DEBUGP("resolve_normal_ct: Can't get tuple\n");
731 /* look for tuple match */
732 h
= nf_conntrack_find_get(&tuple
, NULL
);
734 h
= init_conntrack(&tuple
, l3proto
, l4proto
, skb
, dataoff
);
740 ct
= nf_ct_tuplehash_to_ctrack(h
);
742 /* It exists; we have (non-exclusive) reference. */
743 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
744 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
745 /* Please set reply bit if this packet OK */
748 /* Once we've had two way comms, always ESTABLISHED. */
749 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
750 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct
);
751 *ctinfo
= IP_CT_ESTABLISHED
;
752 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
753 DEBUGP("nf_conntrack_in: related packet for %p\n", ct
);
754 *ctinfo
= IP_CT_RELATED
;
756 DEBUGP("nf_conntrack_in: new packet for %p\n", ct
);
761 skb
->nfct
= &ct
->ct_general
;
762 skb
->nfctinfo
= *ctinfo
;
767 nf_conntrack_in(int pf
, unsigned int hooknum
, struct sk_buff
**pskb
)
770 enum ip_conntrack_info ctinfo
;
771 struct nf_conntrack_l3proto
*l3proto
;
772 struct nf_conntrack_l4proto
*l4proto
;
773 unsigned int dataoff
;
778 /* Previously seen (loopback or untracked)? Ignore. */
780 NF_CT_STAT_INC(ignore
);
784 l3proto
= __nf_ct_l3proto_find((u_int16_t
)pf
);
785 if ((ret
= l3proto
->prepare(pskb
, hooknum
, &dataoff
, &protonum
)) <= 0) {
786 DEBUGP("not prepared to track yet or error occured\n");
790 l4proto
= __nf_ct_l4proto_find((u_int16_t
)pf
, protonum
);
792 /* It may be an special packet, error, unclean...
793 * inverse of the return code tells to the netfilter
794 * core what to do with the packet. */
795 if (l4proto
->error
!= NULL
&&
796 (ret
= l4proto
->error(*pskb
, dataoff
, &ctinfo
, pf
, hooknum
)) <= 0) {
797 NF_CT_STAT_INC(error
);
798 NF_CT_STAT_INC(invalid
);
802 ct
= resolve_normal_ct(*pskb
, dataoff
, pf
, protonum
, l3proto
, l4proto
,
803 &set_reply
, &ctinfo
);
805 /* Not valid part of a connection */
806 NF_CT_STAT_INC(invalid
);
811 /* Too stressed to deal. */
812 NF_CT_STAT_INC(drop
);
816 NF_CT_ASSERT((*pskb
)->nfct
);
818 ret
= l4proto
->packet(ct
, *pskb
, dataoff
, ctinfo
, pf
, hooknum
);
820 /* Invalid: inverse of the return code tells
821 * the netfilter core what to do */
822 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
823 nf_conntrack_put((*pskb
)->nfct
);
824 (*pskb
)->nfct
= NULL
;
825 NF_CT_STAT_INC(invalid
);
829 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
830 nf_conntrack_event_cache(IPCT_STATUS
, *pskb
);
835 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
836 const struct nf_conntrack_tuple
*orig
)
838 return nf_ct_invert_tuple(inverse
, orig
,
839 __nf_ct_l3proto_find(orig
->src
.l3num
),
840 __nf_ct_l4proto_find(orig
->src
.l3num
,
841 orig
->dst
.protonum
));
844 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
845 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
846 enum ip_conntrack_info ctinfo
,
847 const struct sk_buff
*skb
,
848 unsigned long extra_jiffies
,
853 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
856 write_lock_bh(&nf_conntrack_lock
);
858 /* Only update if this is not a fixed timeout */
859 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
)) {
860 write_unlock_bh(&nf_conntrack_lock
);
864 /* If not in hash table, timer will not be active yet */
865 if (!nf_ct_is_confirmed(ct
)) {
866 ct
->timeout
.expires
= extra_jiffies
;
867 event
= IPCT_REFRESH
;
869 unsigned long newtime
= jiffies
+ extra_jiffies
;
871 /* Only update the timeout if the new timeout is at least
872 HZ jiffies from the old timeout. Need del_timer for race
873 avoidance (may already be dying). */
874 if (newtime
- ct
->timeout
.expires
>= HZ
875 && del_timer(&ct
->timeout
)) {
876 ct
->timeout
.expires
= newtime
;
877 add_timer(&ct
->timeout
);
878 event
= IPCT_REFRESH
;
882 #ifdef CONFIG_NF_CT_ACCT
884 ct
->counters
[CTINFO2DIR(ctinfo
)].packets
++;
885 ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
+=
886 skb
->len
- (unsigned int)(skb
->nh
.raw
- skb
->data
);
888 if ((ct
->counters
[CTINFO2DIR(ctinfo
)].packets
& 0x80000000)
889 || (ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
& 0x80000000))
890 event
|= IPCT_COUNTER_FILLING
;
894 write_unlock_bh(&nf_conntrack_lock
);
896 /* must be unlocked when calling event cache */
898 nf_conntrack_event_cache(event
, skb
);
901 #if defined(CONFIG_NF_CT_NETLINK) || \
902 defined(CONFIG_NF_CT_NETLINK_MODULE)
904 #include <linux/netfilter/nfnetlink.h>
905 #include <linux/netfilter/nfnetlink_conntrack.h>
906 #include <linux/mutex.h>
909 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
910 * in ip_conntrack_core, since we don't want the protocols to autoload
911 * or depend on ctnetlink */
912 int nf_ct_port_tuple_to_nfattr(struct sk_buff
*skb
,
913 const struct nf_conntrack_tuple
*tuple
)
915 NFA_PUT(skb
, CTA_PROTO_SRC_PORT
, sizeof(u_int16_t
),
916 &tuple
->src
.u
.tcp
.port
);
917 NFA_PUT(skb
, CTA_PROTO_DST_PORT
, sizeof(u_int16_t
),
918 &tuple
->dst
.u
.tcp
.port
);
925 static const size_t cta_min_proto
[CTA_PROTO_MAX
] = {
926 [CTA_PROTO_SRC_PORT
-1] = sizeof(u_int16_t
),
927 [CTA_PROTO_DST_PORT
-1] = sizeof(u_int16_t
)
930 int nf_ct_port_nfattr_to_tuple(struct nfattr
*tb
[],
931 struct nf_conntrack_tuple
*t
)
933 if (!tb
[CTA_PROTO_SRC_PORT
-1] || !tb
[CTA_PROTO_DST_PORT
-1])
936 if (nfattr_bad_size(tb
, CTA_PROTO_MAX
, cta_min_proto
))
939 t
->src
.u
.tcp
.port
= *(__be16
*)NFA_DATA(tb
[CTA_PROTO_SRC_PORT
-1]);
940 t
->dst
.u
.tcp
.port
= *(__be16
*)NFA_DATA(tb
[CTA_PROTO_DST_PORT
-1]);
946 /* Used by ipt_REJECT and ip6t_REJECT. */
947 void __nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
950 enum ip_conntrack_info ctinfo
;
952 /* This ICMP is in reverse direction to the packet which caused it */
953 ct
= nf_ct_get(skb
, &ctinfo
);
954 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
955 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
957 ctinfo
= IP_CT_RELATED
;
959 /* Attach to new skbuff, and increment count */
960 nskb
->nfct
= &ct
->ct_general
;
961 nskb
->nfctinfo
= ctinfo
;
962 nf_conntrack_get(nskb
->nfct
);
966 do_iter(const struct nf_conntrack_tuple_hash
*i
,
967 int (*iter
)(struct nf_conn
*i
, void *data
),
970 return iter(nf_ct_tuplehash_to_ctrack(i
), data
);
973 /* Bring out ya dead! */
974 static struct nf_conn
*
975 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
976 void *data
, unsigned int *bucket
)
978 struct nf_conntrack_tuple_hash
*h
;
981 write_lock_bh(&nf_conntrack_lock
);
982 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
983 list_for_each_entry(h
, &nf_conntrack_hash
[*bucket
], list
) {
984 ct
= nf_ct_tuplehash_to_ctrack(h
);
989 list_for_each_entry(h
, &unconfirmed
, list
) {
990 ct
= nf_ct_tuplehash_to_ctrack(h
);
994 write_unlock_bh(&nf_conntrack_lock
);
997 atomic_inc(&ct
->ct_general
.use
);
998 write_unlock_bh(&nf_conntrack_lock
);
1003 nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
1006 unsigned int bucket
= 0;
1008 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
1009 /* Time to push up daises... */
1010 if (del_timer(&ct
->timeout
))
1011 death_by_timeout((unsigned long)ct
);
1012 /* ... else the timer will get him soon. */
1018 static int kill_all(struct nf_conn
*i
, void *data
)
1023 static void free_conntrack_hash(struct list_head
*hash
, int vmalloced
, int size
)
1028 free_pages((unsigned long)hash
,
1029 get_order(sizeof(struct list_head
) * size
));
1032 void nf_conntrack_flush()
1034 nf_ct_iterate_cleanup(kill_all
, NULL
);
1037 /* Mishearing the voices in his head, our hero wonders how he's
1038 supposed to kill the mall. */
1039 void nf_conntrack_cleanup(void)
1043 ip_ct_attach
= NULL
;
1045 /* This makes sure all current packets have passed through
1046 netfilter framework. Roll on, two-stage module
1050 nf_ct_event_cache_flush();
1052 nf_conntrack_flush();
1053 if (atomic_read(&nf_conntrack_count
) != 0) {
1055 goto i_see_dead_people
;
1057 /* wait until all references to nf_conntrack_untracked are dropped */
1058 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1061 for (i
= 0; i
< NF_CT_F_NUM
; i
++) {
1062 if (nf_ct_cache
[i
].use
== 0)
1065 NF_CT_ASSERT(nf_ct_cache
[i
].use
== 1);
1066 nf_ct_cache
[i
].use
= 1;
1067 nf_conntrack_unregister_cache(i
);
1069 kmem_cache_destroy(nf_conntrack_expect_cachep
);
1070 free_conntrack_hash(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1071 nf_conntrack_htable_size
);
1073 nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_generic
);
1075 /* free l3proto protocol tables */
1076 for (i
= 0; i
< PF_MAX
; i
++)
1077 if (nf_ct_protos
[i
]) {
1078 kfree(nf_ct_protos
[i
]);
1079 nf_ct_protos
[i
] = NULL
;
1083 static struct list_head
*alloc_hashtable(int size
, int *vmalloced
)
1085 struct list_head
*hash
;
1089 hash
= (void*)__get_free_pages(GFP_KERNEL
,
1090 get_order(sizeof(struct list_head
)
1094 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1095 hash
= vmalloc(sizeof(struct list_head
) * size
);
1099 for (i
= 0; i
< size
; i
++)
1100 INIT_LIST_HEAD(&hash
[i
]);
1105 int set_hashsize(const char *val
, struct kernel_param
*kp
)
1107 int i
, bucket
, hashsize
, vmalloced
;
1108 int old_vmalloced
, old_size
;
1110 struct list_head
*hash
, *old_hash
;
1111 struct nf_conntrack_tuple_hash
*h
;
1113 /* On boot, we can set this without any fancy locking. */
1114 if (!nf_conntrack_htable_size
)
1115 return param_set_uint(val
, kp
);
1117 hashsize
= simple_strtol(val
, NULL
, 0);
1121 hash
= alloc_hashtable(hashsize
, &vmalloced
);
1125 /* We have to rehahs for the new table anyway, so we also can
1126 * use a newrandom seed */
1127 get_random_bytes(&rnd
, 4);
1129 write_lock_bh(&nf_conntrack_lock
);
1130 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1131 while (!list_empty(&nf_conntrack_hash
[i
])) {
1132 h
= list_entry(nf_conntrack_hash
[i
].next
,
1133 struct nf_conntrack_tuple_hash
, list
);
1135 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1136 list_add_tail(&h
->list
, &hash
[bucket
]);
1139 old_size
= nf_conntrack_htable_size
;
1140 old_vmalloced
= nf_conntrack_vmalloc
;
1141 old_hash
= nf_conntrack_hash
;
1143 nf_conntrack_htable_size
= hashsize
;
1144 nf_conntrack_vmalloc
= vmalloced
;
1145 nf_conntrack_hash
= hash
;
1146 nf_conntrack_hash_rnd
= rnd
;
1147 write_unlock_bh(&nf_conntrack_lock
);
1149 free_conntrack_hash(old_hash
, old_vmalloced
, old_size
);
1153 module_param_call(hashsize
, set_hashsize
, param_get_uint
,
1154 &nf_conntrack_htable_size
, 0600);
1156 int __init
nf_conntrack_init(void)
1161 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1162 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */
1163 if (!nf_conntrack_htable_size
) {
1164 nf_conntrack_htable_size
1165 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1166 / sizeof(struct list_head
));
1167 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1168 nf_conntrack_htable_size
= 8192;
1169 if (nf_conntrack_htable_size
< 16)
1170 nf_conntrack_htable_size
= 16;
1172 nf_conntrack_max
= 8 * nf_conntrack_htable_size
;
1174 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1175 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1178 nf_conntrack_hash
= alloc_hashtable(nf_conntrack_htable_size
,
1179 &nf_conntrack_vmalloc
);
1180 if (!nf_conntrack_hash
) {
1181 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1185 ret
= nf_conntrack_register_cache(NF_CT_F_BASIC
, "nf_conntrack:basic",
1186 sizeof(struct nf_conn
));
1188 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1192 nf_conntrack_expect_cachep
= kmem_cache_create("nf_conntrack_expect",
1193 sizeof(struct nf_conntrack_expect
),
1195 if (!nf_conntrack_expect_cachep
) {
1196 printk(KERN_ERR
"Unable to create nf_expect slab cache\n");
1197 goto err_free_conntrack_slab
;
1200 ret
= nf_conntrack_l4proto_register(&nf_conntrack_l4proto_generic
);
1202 goto out_free_expect_slab
;
1204 /* Don't NEED lock here, but good form anyway. */
1205 write_lock_bh(&nf_conntrack_lock
);
1206 for (i
= 0; i
< AF_MAX
; i
++)
1207 nf_ct_l3protos
[i
] = &nf_conntrack_l3proto_generic
;
1208 write_unlock_bh(&nf_conntrack_lock
);
1210 /* For use by REJECT target */
1211 ip_ct_attach
= __nf_conntrack_attach
;
1213 /* Set up fake conntrack:
1214 - to never be deleted, not in any hashes */
1215 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1216 /* - and look it like as a confirmed connection */
1217 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1221 out_free_expect_slab
:
1222 kmem_cache_destroy(nf_conntrack_expect_cachep
);
1223 err_free_conntrack_slab
:
1224 nf_conntrack_unregister_cache(NF_CT_F_BASIC
);
1226 free_conntrack_hash(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1227 nf_conntrack_htable_size
);