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>
32 //#ifdef CONFIG_BCM_NAT
36 #include <net/netfilter/nf_conntrack.h>
37 #include <net/netfilter/nf_conntrack_l3proto.h>
38 #include <net/netfilter/nf_conntrack_l4proto.h>
39 #include <net/netfilter/nf_conntrack_expect.h>
40 #include <net/netfilter/nf_conntrack_helper.h>
41 #include <net/netfilter/nf_conntrack_core.h>
43 #define NF_CONNTRACK_VERSION "0.5.0"
48 #define DEBUGP(format, args...)
51 DEFINE_RWLOCK(nf_conntrack_lock
);
52 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
54 /* nf_conntrack_standalone needs this */
55 atomic_t nf_conntrack_count
= ATOMIC_INIT(0);
56 EXPORT_SYMBOL_GPL(nf_conntrack_count
);
58 void (*nf_conntrack_destroyed
)(struct nf_conn
*conntrack
);
59 EXPORT_SYMBOL_GPL(nf_conntrack_destroyed
);
61 unsigned int nf_conntrack_htable_size __read_mostly
;
62 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
64 int nf_conntrack_max __read_mostly
;
65 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
67 struct list_head
*nf_conntrack_hash __read_mostly
;
68 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
70 struct nf_conn nf_conntrack_untracked __read_mostly
;
71 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
73 unsigned int nf_ct_log_invalid __read_mostly
;
74 LIST_HEAD(unconfirmed
);
75 static int nf_conntrack_vmalloc __read_mostly
;
77 static unsigned int nf_conntrack_next_id
;
79 DEFINE_PER_CPU(struct ip_conntrack_stat
, nf_conntrack_stat
);
80 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat
);
83 extern int ip_conntrack_ipct_delete(struct nf_conn
*ct
, int ct_timeout
);
86 #if defined(CONFIG_BCM_NAT) || defined(CONFIG_BCM_NAT_MODULE)
87 #define BCM_FASTNAT_DENY 1
88 extern int ipv4_conntrack_fastnat
;
90 typedef int (*bcmNatBindHook
)(struct nf_conn
*ct
,
91 enum ip_conntrack_info ctinfo
,
92 struct sk_buff
**pskb
,
93 struct nf_conntrack_l3proto
*l3proto
,
94 struct nf_conntrack_l4proto
*l4proto
);
96 static bcmNatBindHook bcm_nat_bind_hook
= NULL
;
97 int bcm_nat_bind_hook_func(bcmNatBindHook hook_func
) {
98 bcm_nat_bind_hook
= hook_func
;
102 typedef int (*bcmNatHitHook
)(struct sk_buff
*skb
);
103 bcmNatHitHook bcm_nat_hit_hook
= NULL
;
104 int bcm_nat_hit_hook_func(bcmNatHitHook hook_func
) {
105 bcm_nat_hit_hook
= hook_func
;
111 * This scheme offers various size of "struct nf_conn" dependent on
112 * features(helper, nat, ...)
115 #define NF_CT_FEATURES_NAMELEN 256
117 /* name of slab cache. printed in /proc/slabinfo */
120 /* size of slab cache */
123 /* slab cache pointer */
124 struct kmem_cache
*cachep
;
126 /* allocated slab cache + modules which uses this slab cache */
129 } nf_ct_cache
[NF_CT_F_NUM
];
131 /* protect members of nf_ct_cache except of "use" */
132 DEFINE_RWLOCK(nf_ct_cache_lock
);
134 /* This avoids calling kmem_cache_create() with same name simultaneously */
135 static DEFINE_MUTEX(nf_ct_cache_mutex
);
137 /*--SZ angela 09.03 { */
138 #define IP_TRACK_SMALL 0x01
139 #define IP_TRACK_PORT 0x02
140 #define IP_TRACK_DATA 0x04
141 #define IP_TRACK_UDP 0x08
143 extern int track_flag
;
145 u_int8_t port_num_udp
[65536];
147 extern unsigned char mbss_prio_1
; /*SZ-Angela Add for Multiple SSID */
148 extern unsigned char mbss_prio_2
;
149 extern unsigned char mbss_prio_3
;
151 //--SZ angela 09.03 }
153 static int nf_conntrack_hash_rnd_initted
;
154 static unsigned int nf_conntrack_hash_rnd
;
157 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
158 unsigned int size
, unsigned int rnd
)
162 a
= jhash2(tuple
->src
.u3
.all
, ARRAY_SIZE(tuple
->src
.u3
.all
),
163 (tuple
->src
.l3num
<< 16) | tuple
->dst
.protonum
);
164 b
= jhash2(tuple
->dst
.u3
.all
, ARRAY_SIZE(tuple
->dst
.u3
.all
),
165 (tuple
->src
.u
.all
<< 16) | tuple
->dst
.u
.all
);
167 return jhash_2words(a
, b
, rnd
) % size
;
170 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
172 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
173 nf_conntrack_hash_rnd
);
176 int nf_conntrack_register_cache(u_int32_t features
, const char *name
,
181 struct kmem_cache
*cachep
;
183 DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n",
184 features
, name
, size
);
186 if (features
< NF_CT_F_BASIC
|| features
>= NF_CT_F_NUM
) {
187 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n",
192 mutex_lock(&nf_ct_cache_mutex
);
194 write_lock_bh(&nf_ct_cache_lock
);
195 /* e.g: multiple helpers are loaded */
196 if (nf_ct_cache
[features
].use
> 0) {
197 DEBUGP("nf_conntrack_register_cache: already resisterd.\n");
198 if ((!strncmp(nf_ct_cache
[features
].name
, name
,
199 NF_CT_FEATURES_NAMELEN
))
200 && nf_ct_cache
[features
].size
== size
) {
201 DEBUGP("nf_conntrack_register_cache: reusing.\n");
202 nf_ct_cache
[features
].use
++;
207 write_unlock_bh(&nf_ct_cache_lock
);
208 mutex_unlock(&nf_ct_cache_mutex
);
211 write_unlock_bh(&nf_ct_cache_lock
);
214 * The memory space for name of slab cache must be alive until
215 * cache is destroyed.
217 cache_name
= kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN
, GFP_ATOMIC
);
218 if (cache_name
== NULL
) {
219 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n");
224 if (strlcpy(cache_name
, name
, NF_CT_FEATURES_NAMELEN
)
225 >= NF_CT_FEATURES_NAMELEN
) {
226 printk("nf_conntrack_register_cache: name too long\n");
231 cachep
= kmem_cache_create(cache_name
, size
, 0, 0,
234 printk("nf_conntrack_register_cache: Can't create slab cache "
235 "for the features = 0x%x\n", features
);
240 write_lock_bh(&nf_ct_cache_lock
);
241 nf_ct_cache
[features
].use
= 1;
242 nf_ct_cache
[features
].size
= size
;
243 nf_ct_cache
[features
].cachep
= cachep
;
244 nf_ct_cache
[features
].name
= cache_name
;
245 write_unlock_bh(&nf_ct_cache_lock
);
252 mutex_unlock(&nf_ct_cache_mutex
);
255 EXPORT_SYMBOL_GPL(nf_conntrack_register_cache
);
257 /* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */
258 void nf_conntrack_unregister_cache(u_int32_t features
)
260 struct kmem_cache
*cachep
;
264 * This assures that kmem_cache_create() isn't called before destroying
267 DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features
);
268 mutex_lock(&nf_ct_cache_mutex
);
270 write_lock_bh(&nf_ct_cache_lock
);
271 if (--nf_ct_cache
[features
].use
> 0) {
272 write_unlock_bh(&nf_ct_cache_lock
);
273 mutex_unlock(&nf_ct_cache_mutex
);
276 cachep
= nf_ct_cache
[features
].cachep
;
277 name
= nf_ct_cache
[features
].name
;
278 nf_ct_cache
[features
].cachep
= NULL
;
279 nf_ct_cache
[features
].name
= NULL
;
280 nf_ct_cache
[features
].size
= 0;
281 write_unlock_bh(&nf_ct_cache_lock
);
285 kmem_cache_destroy(cachep
);
288 mutex_unlock(&nf_ct_cache_mutex
);
290 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_cache
);
293 nf_ct_get_tuple(const struct sk_buff
*skb
,
295 unsigned int dataoff
,
298 struct nf_conntrack_tuple
*tuple
,
299 const struct nf_conntrack_l3proto
*l3proto
,
300 const struct nf_conntrack_l4proto
*l4proto
)
302 NF_CT_TUPLE_U_BLANK(tuple
);
304 tuple
->src
.l3num
= l3num
;
305 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
307 tuple
->dst
.protonum
= protonum
;
308 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
310 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
312 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
315 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
316 const struct nf_conntrack_tuple
*orig
,
317 const struct nf_conntrack_l3proto
*l3proto
,
318 const struct nf_conntrack_l4proto
*l4proto
)
320 NF_CT_TUPLE_U_BLANK(inverse
);
322 inverse
->src
.l3num
= orig
->src
.l3num
;
323 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
326 inverse
->dst
.dir
= !orig
->dst
.dir
;
328 inverse
->dst
.protonum
= orig
->dst
.protonum
;
329 return l4proto
->invert_tuple(inverse
, orig
);
331 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
334 clean_from_lists(struct nf_conn
*ct
)
336 DEBUGP("clean_from_lists(%p)\n", ct
);
337 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
338 list_del(&ct
->tuplehash
[IP_CT_DIR_REPLY
].list
);
340 /* Destroy all pending expectations */
341 nf_ct_remove_expectations(ct
);
345 destroy_conntrack(struct nf_conntrack
*nfct
)
347 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
348 struct nf_conntrack_l4proto
*l4proto
;
349 typeof(nf_conntrack_destroyed
) destroyed
;
351 DEBUGP("destroy_conntrack(%p)\n", ct
);
352 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
353 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
356 ip_conntrack_ipct_delete(ct
, 0);
359 nf_conntrack_event(IPCT_DESTROY
, ct
);
360 set_bit(IPS_DYING_BIT
, &ct
->status
);
362 /* To make sure we don't get any weird locking issues here:
363 * destroy_conntrack() MUST NOT be called with a write lock
364 * to nf_conntrack_lock!!! -HW */
366 l4proto
= __nf_ct_l4proto_find(ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.src
.l3num
,
367 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
.dst
.protonum
);
368 if (l4proto
&& l4proto
->destroy
)
369 l4proto
->destroy(ct
);
371 destroyed
= rcu_dereference(nf_conntrack_destroyed
);
377 write_lock_bh(&nf_conntrack_lock
);
378 /* Expectations will have been removed in clean_from_lists,
379 * except TFTP can create an expectation on the first packet,
380 * before connection is in the list, so we need to clean here,
382 nf_ct_remove_expectations(ct
);
384 #if defined(CONFIG_NETFILTER_XT_MATCH_LAYER7) || defined(CONFIG_NETFILTER_XT_MATCH_LAYER7_MODULE)
385 if(ct
->layer7
.app_proto
)
386 kfree(ct
->layer7
.app_proto
);
387 if(ct
->layer7
.app_data
)
388 kfree(ct
->layer7
.app_data
);
392 /* We overload first tuple to link into unconfirmed list. */
393 if (!nf_ct_is_confirmed(ct
)) {
394 BUG_ON(list_empty(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
));
395 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
398 NF_CT_STAT_INC(delete);
399 write_unlock_bh(&nf_conntrack_lock
);
402 nf_ct_put(ct
->master
);
404 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct
);
405 nf_conntrack_free(ct
);
408 static void death_by_timeout(unsigned long ul_conntrack
)
410 struct nf_conn
*ct
= (void *)ul_conntrack
;
411 struct nf_conn_help
*help
= nfct_help(ct
);
412 struct nf_conntrack_helper
*helper
;
415 /* If negative error is returned it means the entry hasn't
418 if (ip_conntrack_ipct_delete(ct
, jiffies
>= ct
->timeout
.expires
? 1 : 0) != 0)
424 helper
= rcu_dereference(help
->helper
);
425 if (helper
&& helper
->destroy
)
430 write_lock_bh(&nf_conntrack_lock
);
431 /* Inside lock so preempt is disabled on module removal path.
432 * Otherwise we can get spurious warnings. */
433 NF_CT_STAT_INC(delete_list
);
434 clean_from_lists(ct
);
435 write_unlock_bh(&nf_conntrack_lock
);
439 struct nf_conntrack_tuple_hash
*
440 __nf_conntrack_find(const struct nf_conntrack_tuple
*tuple
,
441 const struct nf_conn
*ignored_conntrack
)
443 struct nf_conntrack_tuple_hash
*h
;
444 unsigned int hash
= hash_conntrack(tuple
);
446 list_for_each_entry(h
, &nf_conntrack_hash
[hash
], list
) {
447 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
448 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
449 NF_CT_STAT_INC(found
);
452 NF_CT_STAT_INC(searched
);
457 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
459 /* Find a connection corresponding to a tuple. */
460 struct nf_conntrack_tuple_hash
*
461 nf_conntrack_find_get(const struct nf_conntrack_tuple
*tuple
,
462 const struct nf_conn
*ignored_conntrack
)
464 struct nf_conntrack_tuple_hash
*h
;
466 read_lock_bh(&nf_conntrack_lock
);
467 h
= __nf_conntrack_find(tuple
, ignored_conntrack
);
469 atomic_inc(&nf_ct_tuplehash_to_ctrack(h
)->ct_general
.use
);
470 read_unlock_bh(&nf_conntrack_lock
);
474 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
476 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
478 unsigned int repl_hash
)
480 ct
->id
= ++nf_conntrack_next_id
;
481 list_add(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
,
482 &nf_conntrack_hash
[hash
]);
483 list_add(&ct
->tuplehash
[IP_CT_DIR_REPLY
].list
,
484 &nf_conntrack_hash
[repl_hash
]);
487 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
489 unsigned int hash
, repl_hash
;
491 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
492 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
494 write_lock_bh(&nf_conntrack_lock
);
495 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
496 write_unlock_bh(&nf_conntrack_lock
);
498 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
500 /* Confirm a connection given skb; places it in hash table */
502 __nf_conntrack_confirm(struct sk_buff
**pskb
)
504 unsigned int hash
, repl_hash
;
505 struct nf_conntrack_tuple_hash
*h
;
507 struct nf_conn_help
*help
;
508 enum ip_conntrack_info ctinfo
;
510 ct
= nf_ct_get(*pskb
, &ctinfo
);
512 /* ipt_REJECT uses nf_conntrack_attach to attach related
513 ICMP/TCP RST packets in other direction. Actual packet
514 which created connection will be IP_CT_NEW or for an
515 expected connection, IP_CT_RELATED. */
516 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
519 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
520 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
522 /* We're not in hash table, and we refuse to set up related
523 connections for unconfirmed conns. But packet copies and
524 REJECT will give spurious warnings here. */
525 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
527 /* No external references means noone else could have
529 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
530 DEBUGP("Confirming conntrack %p\n", ct
);
532 write_lock_bh(&nf_conntrack_lock
);
534 /* See if there's one in the list already, including reverse:
535 NAT could have grabbed it without realizing, since we're
536 not in the hash. If there is, we lost race. */
537 list_for_each_entry(h
, &nf_conntrack_hash
[hash
], list
)
538 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
541 list_for_each_entry(h
, &nf_conntrack_hash
[repl_hash
], list
)
542 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
546 /* Remove from unconfirmed list */
547 list_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
);
549 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
550 /* Timer relative to confirmation time, not original
551 setting time, otherwise we'd get timer wrap in
552 weird delay cases. */
553 ct
->timeout
.expires
+= jiffies
;
554 add_timer(&ct
->timeout
);
555 atomic_inc(&ct
->ct_general
.use
);
556 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
557 NF_CT_STAT_INC(insert
);
558 write_unlock_bh(&nf_conntrack_lock
);
559 help
= nfct_help(ct
);
560 if (help
&& help
->helper
)
561 nf_conntrack_event_cache(IPCT_HELPER
, *pskb
);
562 #ifdef CONFIG_NF_NAT_NEEDED
563 if (test_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
) ||
564 test_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
))
565 nf_conntrack_event_cache(IPCT_NATINFO
, *pskb
);
567 nf_conntrack_event_cache(master_ct(ct
) ?
568 IPCT_RELATED
: IPCT_NEW
, *pskb
);
572 NF_CT_STAT_INC(insert_failed
);
573 write_unlock_bh(&nf_conntrack_lock
);
576 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
578 /* Returns true if a connection correspondings to the tuple (required
581 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
582 const struct nf_conn
*ignored_conntrack
)
584 struct nf_conntrack_tuple_hash
*h
;
586 read_lock_bh(&nf_conntrack_lock
);
587 h
= __nf_conntrack_find(tuple
, ignored_conntrack
);
588 read_unlock_bh(&nf_conntrack_lock
);
592 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
594 /* There's a small race here where we may free a just-assured
595 connection. Too bad: we're in trouble anyway. */
596 static int early_drop(struct list_head
*chain
)
598 /* Traverse backwards: gives us oldest, which is roughly LRU */
599 struct nf_conntrack_tuple_hash
*h
;
600 struct nf_conn
*ct
= NULL
, *tmp
;
603 read_lock_bh(&nf_conntrack_lock
);
604 list_for_each_entry_reverse(h
, chain
, list
) {
605 tmp
= nf_ct_tuplehash_to_ctrack(h
);
606 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
)) {
608 atomic_inc(&ct
->ct_general
.use
);
612 read_unlock_bh(&nf_conntrack_lock
);
618 ip_conntrack_ipct_delete(ct
, 0);
621 if (del_timer(&ct
->timeout
)) {
622 death_by_timeout((unsigned long)ct
);
624 NF_CT_STAT_INC_ATOMIC(early_drop
);
630 static struct nf_conn
*
631 __nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
632 const struct nf_conntrack_tuple
*repl
,
633 const struct nf_conntrack_l3proto
*l3proto
,
636 struct nf_conn
*conntrack
= NULL
;
637 struct nf_conntrack_helper
*helper
;
639 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
640 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
641 nf_conntrack_hash_rnd_initted
= 1;
644 /* We don't want any race condition at early drop stage */
645 atomic_inc(&nf_conntrack_count
);
648 && atomic_read(&nf_conntrack_count
) > nf_conntrack_max
) {
649 unsigned int hash
= hash_conntrack(orig
);
650 /* Try dropping from this hash chain. */
651 if (!early_drop(&nf_conntrack_hash
[hash
])) {
652 atomic_dec(&nf_conntrack_count
);
655 "nf_conntrack: table full, dropping"
657 return ERR_PTR(-ENOMEM
);
661 /* find features needed by this conntrack. */
662 features
|= l3proto
->get_features(orig
);
664 /* FIXME: protect helper list per RCU */
665 read_lock_bh(&nf_conntrack_lock
);
666 helper
= __nf_ct_helper_find(repl
);
667 /* NAT might want to assign a helper later */
668 if (helper
|| features
& NF_CT_F_NAT
)
669 features
|= NF_CT_F_HELP
;
670 read_unlock_bh(&nf_conntrack_lock
);
672 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features
);
674 read_lock_bh(&nf_ct_cache_lock
);
676 if (unlikely(!nf_ct_cache
[features
].use
)) {
677 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n",
682 conntrack
= kmem_cache_alloc(nf_ct_cache
[features
].cachep
, GFP_ATOMIC
);
683 if (conntrack
== NULL
) {
684 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n");
688 memset(conntrack
, 0, nf_ct_cache
[features
].size
);
689 conntrack
->features
= features
;
690 atomic_set(&conntrack
->ct_general
.use
, 1);
691 conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
692 conntrack
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
693 /* Don't set timer yet: wait for confirmation */
694 setup_timer(&conntrack
->timeout
, death_by_timeout
,
695 (unsigned long)conntrack
);
696 read_unlock_bh(&nf_ct_cache_lock
);
700 read_unlock_bh(&nf_ct_cache_lock
);
701 atomic_dec(&nf_conntrack_count
);
705 struct nf_conn
*nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
706 const struct nf_conntrack_tuple
*repl
)
708 struct nf_conntrack_l3proto
*l3proto
;
712 l3proto
= __nf_ct_l3proto_find(orig
->src
.l3num
);
713 ct
= __nf_conntrack_alloc(orig
, repl
, l3proto
, 0);
718 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
720 void nf_conntrack_free(struct nf_conn
*conntrack
)
722 u_int32_t features
= conntrack
->features
;
723 NF_CT_ASSERT(features
>= NF_CT_F_BASIC
&& features
< NF_CT_F_NUM
);
724 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features
,
726 kmem_cache_free(nf_ct_cache
[features
].cachep
, conntrack
);
727 atomic_dec(&nf_conntrack_count
);
729 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
731 /* Allocate a new conntrack: we return -ENOMEM if classification
732 failed due to stress. Otherwise it really is unclassifiable. */
733 static struct nf_conntrack_tuple_hash
*
734 init_conntrack(const struct nf_conntrack_tuple
*tuple
,
735 struct nf_conntrack_l3proto
*l3proto
,
736 struct nf_conntrack_l4proto
*l4proto
,
738 unsigned int dataoff
)
740 struct nf_conn
*conntrack
;
741 struct nf_conn_help
*help
;
742 struct nf_conntrack_tuple repl_tuple
;
743 struct nf_conntrack_expect
*exp
;
744 u_int32_t features
= 0;
746 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
747 DEBUGP("Can't invert tuple.\n");
751 read_lock_bh(&nf_conntrack_lock
);
752 exp
= __nf_conntrack_expect_find(tuple
);
753 if (exp
&& exp
->helper
)
754 features
= NF_CT_F_HELP
;
755 read_unlock_bh(&nf_conntrack_lock
);
757 conntrack
= __nf_conntrack_alloc(tuple
, &repl_tuple
, l3proto
, features
);
758 if (conntrack
== NULL
|| IS_ERR(conntrack
)) {
759 DEBUGP("Can't allocate conntrack.\n");
760 return (struct nf_conntrack_tuple_hash
*)conntrack
;
763 if (!l4proto
->new(conntrack
, skb
, dataoff
)) {
764 nf_conntrack_free(conntrack
);
765 DEBUGP("init conntrack: can't track with proto module\n");
769 write_lock_bh(&nf_conntrack_lock
);
770 //--SZ angela 09.03 {
771 /* if the qos enable and the layer 3 protocol is ipv4 */
772 if((track_flag
== 1) && (strcmp(l3proto
->name
, "ipv4") == 0)) {
773 conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].track
.flag
= 0;
774 conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].track
.number
=1;
776 if(strcmp(l4proto
->name
, "udp") == 0) {
777 if(conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.u3
.ip
== ipaddr
)
778 port_num_udp
[conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.u
.all
]++;
780 port_num_udp
[conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.u
.all
]++;
783 //--SZ angela 09.03 }
785 exp
= find_expectation(tuple
);
787 help
= nfct_help(conntrack
);
789 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
791 /* Welcome, Mr. Bond. We've been expecting you... */
792 __set_bit(IPS_EXPECTED_BIT
, &conntrack
->status
);
793 conntrack
->master
= exp
->master
;
795 rcu_assign_pointer(help
->helper
, exp
->helper
);
796 #ifdef CONFIG_NF_CONNTRACK_MARK
797 conntrack
->mark
= exp
->master
->mark
;
799 #ifdef CONFIG_NF_CONNTRACK_SECMARK
800 conntrack
->secmark
= exp
->master
->secmark
;
802 nf_conntrack_get(&conntrack
->master
->ct_general
);
803 NF_CT_STAT_INC(expect_new
);
806 /* not in hash table yet, so not strictly necessary */
807 rcu_assign_pointer(help
->helper
,
808 __nf_ct_helper_find(&repl_tuple
));
813 /* Overload tuple linked list to put us in unconfirmed list. */
814 list_add(&conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].list
, &unconfirmed
);
816 write_unlock_bh(&nf_conntrack_lock
);
820 exp
->expectfn(conntrack
, exp
);
821 nf_conntrack_expect_put(exp
);
824 return &conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
];
827 //--SZ angela 09.03 {
828 //On success, returns h->track.flags & IP_TRACK_MARK
829 inline int deal_track(struct nf_conntrack_tuple_hash
*h
, int len
)
831 struct nf_conntrack_tuple_hash
*rep_h
;
832 unsigned int port
=0, dport
=0;
833 struct nf_conn
*ct
= NULL
;
835 // Add the packet number of this connect track and record the length of the packet
837 // if((h->track.flag & IP_TRACK_FULL) != IP_TRACK_FULL)
840 h
->track
.large_packet
++;
843 // The download packet set the IP_TRACK_DOWN flag
844 if(ntohl(h
->tuple
.dst
.u3
.ip
) == ipaddr
)
846 port
= h
->tuple
.dst
.u
.all
;
847 dport
= h
->tuple
.src
.u
.all
;
851 port
= h
->tuple
.src
.u
.all
;
852 dport
= h
->tuple
.dst
.u
.all
;
855 // if the connect track is data connect ,we return IP_TRACK_DATA
856 if((h
->track
.flag
& IP_TRACK_DATA
) == IP_TRACK_DATA
)
857 return IP_TRACK_DATA
;
859 // if the destination port of this connect track is one of 80,8080,443.We return IP_TRACK_PORT
860 if((h
->track
.flag
& IP_TRACK_PORT
) == IP_TRACK_PORT
)
861 return IP_TRACK_PORT
;
863 if((h
->track
.flag
& IP_TRACK_SMALL
) == IP_TRACK_SMALL
)
864 return IP_TRACK_SMALL
;
866 if((h
->track
.flag
& IP_TRACK_UDP
) == IP_TRACK_UDP
)
869 ct
= nf_ct_tuplehash_to_ctrack(h
);
872 if(NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
)
873 rep_h
= &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
875 rep_h
= &ct
->tuplehash
[IP_CT_DIR_REPLY
];
879 if(ntohs(dport
) == 80 || ntohs(dport
) == 8080 || ntohs(dport
) == 443)
881 h
->track
.flag
|= IP_TRACK_PORT
;
882 rep_h
->track
.flag
|= IP_TRACK_PORT
;
883 return IP_TRACK_PORT
;
886 // if the port has connections more than 30, we mark it and return IP_TRACK_UDP
887 // h->tuple.dst.protonum == 17 &&
888 if((port_num_udp
[port
] > 30 || port_num_udp
[dport
] >30))
890 h
->track
.flag
|= IP_TRACK_UDP
;
891 rep_h
->track
.flag
|= IP_TRACK_UDP
;
895 if(h
->track
.number
== 250)
897 if(h
->track
.large_packet
<70)
899 if((rep_h
->track
.flag
& IP_TRACK_DATA
) == IP_TRACK_DATA
)
901 h
->track
.flag
|= IP_TRACK_DATA
;
902 return IP_TRACK_DATA
;
904 h
->track
.flag
|= IP_TRACK_SMALL
;
905 return IP_TRACK_SMALL
;
908 if((rep_h
->track
.flag
& IP_TRACK_SMALL
) == IP_TRACK_SMALL
)
910 rep_h
->track
.flag
|= IP_TRACK_DATA
;
911 rep_h
->track
.flag
&= ~IP_TRACK_SMALL
;
913 h
->track
.flag
|= IP_TRACK_DATA
;
914 return IP_TRACK_DATA
;
919 //--SZ angela 09.03 }
921 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
922 static inline struct nf_conn
*
923 resolve_normal_ct(struct sk_buff
*skb
,
924 unsigned int dataoff
,
927 struct nf_conntrack_l3proto
*l3proto
,
928 struct nf_conntrack_l4proto
*l4proto
,
930 enum ip_conntrack_info
*ctinfo
)
932 struct nf_conntrack_tuple tuple
;
933 struct nf_conntrack_tuple_hash
*h
;
936 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
937 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
939 DEBUGP("resolve_normal_ct: Can't get tuple\n");
943 /* look for tuple match */
944 h
= nf_conntrack_find_get(&tuple
, NULL
);
946 h
= init_conntrack(&tuple
, l3proto
, l4proto
, skb
, dataoff
);
952 /*/--SZ angela 09.03 {*/
953 else if((track_flag
== 1) && (strcmp(l3proto
->name
, "ipv4") == 0)) {
954 write_lock_bh(&nf_conntrack_lock
);
956 switch(deal_track(h
, ntohs(ip_hdr(skb
)->tot_len
))) {
958 if(ntohl(h
->tuple
.dst
.u3
.ip
) == ipaddr
)
964 if(ntohl(h
->tuple
.dst
.u3
.ip
) == ipaddr
)
970 if(ntohl(h
->tuple
.dst
.u3
.ip
) == ipaddr
)
976 if(ntohl(h
->tuple
.dst
.u3
.ip
) == ipaddr
)
982 if(ntohl(h
->tuple
.dst
.u3
.ip
) == ipaddr
)
990 switch(skb
->wl_idx
) /*/SZ-Angela Add for Multiple SSID*/
1001 if(mbss_prio_3
== 0)
1008 write_unlock_bh(&nf_conntrack_lock
);
1009 } //--SZ angela 09.03 }
1010 ct
= nf_ct_tuplehash_to_ctrack(h
);
1012 /* It exists; we have (non-exclusive) reference. */
1013 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1014 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
1015 /* Please set reply bit if this packet OK */
1018 /* Once we've had two way comms, always ESTABLISHED. */
1019 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1020 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct
);
1021 *ctinfo
= IP_CT_ESTABLISHED
;
1022 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1023 DEBUGP("nf_conntrack_in: related packet for %p\n", ct
);
1024 *ctinfo
= IP_CT_RELATED
;
1026 DEBUGP("nf_conntrack_in: new packet for %p\n", ct
);
1027 *ctinfo
= IP_CT_NEW
;
1031 skb
->nfct
= &ct
->ct_general
;
1032 skb
->nfctinfo
= *ctinfo
;
1036 #if defined(CONFIG_BCM_NAT) || defined(CONFIG_BCM_NAT_MODULE)
1037 extern struct sk_buff
* nf_ct_ipv4_gather_frags(struct sk_buff
*skb
, u_int32_t user
);
1041 nf_conntrack_in(int pf
, unsigned int hooknum
, struct sk_buff
**pskb
)
1044 enum ip_conntrack_info ctinfo
;
1045 struct nf_conntrack_l3proto
*l3proto
;
1046 struct nf_conntrack_l4proto
*l4proto
;
1047 unsigned int dataoff
;
1051 #if defined(CONFIG_BCM_NAT) || defined(CONFIG_BCM_NAT_MODULE)
1052 struct nf_conn_nat
*nat
;
1053 struct nf_conn_help
*help
;
1056 /* Previously seen (loopback or untracked)? Ignore. */
1057 if ((*pskb
)->nfct
) {
1058 NF_CT_STAT_INC_ATOMIC(ignore
);
1062 /* rcu_read_lock()ed by nf_hook_slow */
1063 l3proto
= __nf_ct_l3proto_find((u_int16_t
)pf
);
1065 if ((ret
= l3proto
->prepare(pskb
, hooknum
, &dataoff
, &protonum
)) <= 0) {
1066 DEBUGP("not prepared to track yet or error occured\n");
1070 #if defined(CONFIG_BCM_NAT) || defined(CONFIG_BCM_NAT_MODULE)
1071 if (pf
== PF_INET
) {
1072 /* Gather fragments. */
1073 if (ip_hdr(*pskb
)->frag_off
& htons(IP_MF
| IP_OFFSET
)) {
1074 *pskb
= nf_ct_ipv4_gather_frags(*pskb
,
1075 hooknum
== NF_IP_PRE_ROUTING
?
1076 IP_DEFRAG_CONNTRACK_IN
:
1077 IP_DEFRAG_CONNTRACK_OUT
);
1084 l4proto
= __nf_ct_l4proto_find((u_int16_t
)pf
, protonum
);
1086 /* It may be an special packet, error, unclean...
1087 * inverse of the return code tells to the netfilter
1088 * core what to do with the packet. */
1089 if (l4proto
->error
!= NULL
&&
1090 (ret
= l4proto
->error(*pskb
, dataoff
, &ctinfo
, pf
, hooknum
)) <= 0) {
1091 NF_CT_STAT_INC_ATOMIC(error
);
1092 NF_CT_STAT_INC_ATOMIC(invalid
);
1096 ct
= resolve_normal_ct(*pskb
, dataoff
, pf
, protonum
, l3proto
, l4proto
,
1097 &set_reply
, &ctinfo
);
1099 /* Not valid part of a connection */
1100 NF_CT_STAT_INC_ATOMIC(invalid
);
1105 /* Too stressed to deal. */
1106 NF_CT_STAT_INC_ATOMIC(drop
);
1110 NF_CT_ASSERT((*pskb
)->nfct
);
1112 ret
= l4proto
->packet(ct
, *pskb
, dataoff
, ctinfo
, pf
, hooknum
);
1114 /* Invalid: inverse of the return code tells
1115 * the netfilter core what to do */
1116 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
1117 nf_conntrack_put((*pskb
)->nfct
);
1118 (*pskb
)->nfct
= NULL
;
1119 NF_CT_STAT_INC_ATOMIC(invalid
);
1123 #if defined(CONFIG_BCM_NAT) || defined(CONFIG_BCM_NAT_MODULE)
1124 help
= nfct_help(ct
);
1126 if (pf
== PF_INET
&& ipv4_conntrack_fastnat
&& bcm_nat_bind_hook
1127 && !(nat
->info
.nat_type
& BCM_FASTNAT_DENY
)
1129 && (ctinfo
== IP_CT_ESTABLISHED
|| ctinfo
== IP_CT_IS_REPLY
)
1130 && (hooknum
== NF_IP_PRE_ROUTING
) &&
1131 (protonum
== IPPROTO_TCP
|| protonum
== IPPROTO_UDP
)) {
1133 struct nf_conntrack_tuple
*t1
, *t2
;
1134 t1
= &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
;
1135 t2
= &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
;
1136 if (!(t1
->dst
.u3
.ip
== t2
->src
.u3
.ip
&&
1137 t1
->src
.u3
.ip
== t2
->dst
.u3
.ip
&&
1138 t1
->dst
.u
.all
== t2
->src
.u
.all
&&
1139 t1
->src
.u
.all
== t2
->dst
.u
.all
)) {
1140 ret
= bcm_nat_bind_hook(ct
, ctinfo
, pskb
, l3proto
, l4proto
);
1145 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1146 #if defined(CONFIG_BCM_NAT) || defined(CONFIG_BCM_NAT_MODULE)
1147 if (pf
== PF_INET
&& hooknum
== NF_IP_LOCAL_OUT
)
1148 nat
->info
.nat_type
|= BCM_FASTNAT_DENY
;
1150 nf_conntrack_event_cache(IPCT_STATUS
, *pskb
);
1154 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1156 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1157 const struct nf_conntrack_tuple
*orig
)
1162 ret
= nf_ct_invert_tuple(inverse
, orig
,
1163 __nf_ct_l3proto_find(orig
->src
.l3num
),
1164 __nf_ct_l4proto_find(orig
->src
.l3num
,
1165 orig
->dst
.protonum
));
1169 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1171 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1172 implicitly racy: see __nf_conntrack_confirm */
1173 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1174 const struct nf_conntrack_tuple
*newreply
)
1176 struct nf_conn_help
*help
= nfct_help(ct
);
1178 write_lock_bh(&nf_conntrack_lock
);
1179 /* Should be unconfirmed, so not in hash table yet */
1180 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1182 DEBUGP("Altering reply tuple of %p to ", ct
);
1183 NF_CT_DUMP_TUPLE(newreply
);
1185 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1186 if (!ct
->master
&& help
&& help
->expecting
== 0) {
1187 struct nf_conntrack_helper
*helper
;
1188 helper
= __nf_ct_helper_find(newreply
);
1190 memset(&help
->help
, 0, sizeof(help
->help
));
1191 /* not in hash table yet, so not strictly necessary */
1192 rcu_assign_pointer(help
->helper
, helper
);
1194 write_unlock_bh(&nf_conntrack_lock
);
1196 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1198 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1199 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1200 enum ip_conntrack_info ctinfo
,
1201 const struct sk_buff
*skb
,
1202 unsigned long extra_jiffies
,
1207 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1210 write_lock_bh(&nf_conntrack_lock
);
1212 /* Only update if this is not a fixed timeout */
1213 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
)) {
1214 write_unlock_bh(&nf_conntrack_lock
);
1218 /* If not in hash table, timer will not be active yet */
1219 if (!nf_ct_is_confirmed(ct
)) {
1221 ct
->expire_jiffies
= extra_jiffies
;
1223 ct
->timeout
.expires
= extra_jiffies
;
1224 event
= IPCT_REFRESH
;
1226 unsigned long newtime
= jiffies
+ extra_jiffies
;
1228 /* Only update the timeout if the new timeout is at least
1229 HZ jiffies from the old timeout. Need del_timer for race
1230 avoidance (may already be dying). */
1231 if (newtime
- ct
->timeout
.expires
>= HZ
1232 && del_timer(&ct
->timeout
)) {
1234 ct
->expire_jiffies
= extra_jiffies
;
1236 ct
->timeout
.expires
= newtime
;
1237 add_timer(&ct
->timeout
);
1238 event
= IPCT_REFRESH
;
1242 #ifdef CONFIG_NF_CT_ACCT
1244 ct
->counters
[CTINFO2DIR(ctinfo
)].packets
++;
1245 ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
+=
1246 skb
->len
- skb_network_offset(skb
);
1248 if ((ct
->counters
[CTINFO2DIR(ctinfo
)].packets
& 0x80000000)
1249 || (ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
& 0x80000000))
1250 event
|= IPCT_COUNTER_FILLING
;
1254 write_unlock_bh(&nf_conntrack_lock
);
1256 /* must be unlocked when calling event cache */
1258 nf_conntrack_event_cache(event
, skb
);
1260 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1262 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1264 #include <linux/netfilter/nfnetlink.h>
1265 #include <linux/netfilter/nfnetlink_conntrack.h>
1266 #include <linux/mutex.h>
1269 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1270 * in ip_conntrack_core, since we don't want the protocols to autoload
1271 * or depend on ctnetlink */
1272 int nf_ct_port_tuple_to_nfattr(struct sk_buff
*skb
,
1273 const struct nf_conntrack_tuple
*tuple
)
1275 NFA_PUT(skb
, CTA_PROTO_SRC_PORT
, sizeof(u_int16_t
),
1276 &tuple
->src
.u
.tcp
.port
);
1277 NFA_PUT(skb
, CTA_PROTO_DST_PORT
, sizeof(u_int16_t
),
1278 &tuple
->dst
.u
.tcp
.port
);
1284 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr
);
1286 static const size_t cta_min_proto
[CTA_PROTO_MAX
] = {
1287 [CTA_PROTO_SRC_PORT
-1] = sizeof(u_int16_t
),
1288 [CTA_PROTO_DST_PORT
-1] = sizeof(u_int16_t
)
1291 int nf_ct_port_nfattr_to_tuple(struct nfattr
*tb
[],
1292 struct nf_conntrack_tuple
*t
)
1294 if (!tb
[CTA_PROTO_SRC_PORT
-1] || !tb
[CTA_PROTO_DST_PORT
-1])
1297 if (nfattr_bad_size(tb
, CTA_PROTO_MAX
, cta_min_proto
))
1300 t
->src
.u
.tcp
.port
= *(__be16
*)NFA_DATA(tb
[CTA_PROTO_SRC_PORT
-1]);
1301 t
->dst
.u
.tcp
.port
= *(__be16
*)NFA_DATA(tb
[CTA_PROTO_DST_PORT
-1]);
1305 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple
);
1308 /* Used by ipt_REJECT and ip6t_REJECT. */
1309 void __nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
1312 enum ip_conntrack_info ctinfo
;
1314 /* This ICMP is in reverse direction to the packet which caused it */
1315 ct
= nf_ct_get(skb
, &ctinfo
);
1316 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1317 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
1319 ctinfo
= IP_CT_RELATED
;
1321 /* Attach to new skbuff, and increment count */
1322 nskb
->nfct
= &ct
->ct_general
;
1323 nskb
->nfctinfo
= ctinfo
;
1324 nf_conntrack_get(nskb
->nfct
);
1326 EXPORT_SYMBOL_GPL(__nf_conntrack_attach
);
1329 do_iter(const struct nf_conntrack_tuple_hash
*i
,
1330 int (*iter
)(struct nf_conn
*i
, void *data
),
1333 return iter(nf_ct_tuplehash_to_ctrack(i
), data
);
1336 /* Bring out ya dead! */
1337 static struct nf_conn
*
1338 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
1339 void *data
, unsigned int *bucket
)
1341 struct nf_conntrack_tuple_hash
*h
;
1344 write_lock_bh(&nf_conntrack_lock
);
1345 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1346 list_for_each_entry(h
, &nf_conntrack_hash
[*bucket
], list
) {
1347 ct
= nf_ct_tuplehash_to_ctrack(h
);
1352 list_for_each_entry(h
, &unconfirmed
, list
) {
1353 ct
= nf_ct_tuplehash_to_ctrack(h
);
1355 set_bit(IPS_DYING_BIT
, &ct
->status
);
1357 write_unlock_bh(&nf_conntrack_lock
);
1360 atomic_inc(&ct
->ct_general
.use
);
1361 write_unlock_bh(&nf_conntrack_lock
);
1366 nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
1369 unsigned int bucket
= 0;
1371 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
1373 ip_conntrack_ipct_delete(ct
, 0);
1375 /* Time to push up daises... */
1376 if (del_timer(&ct
->timeout
))
1377 death_by_timeout((unsigned long)ct
);
1378 /* ... else the timer will get him soon. */
1383 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1385 static int kill_all(struct nf_conn
*i
, void *data
)
1390 static void free_conntrack_hash(struct list_head
*hash
, int vmalloced
, int size
)
1395 free_pages((unsigned long)hash
,
1396 get_order(sizeof(struct list_head
) * size
));
1399 void nf_conntrack_flush(void)
1401 nf_ct_iterate_cleanup(kill_all
, NULL
);
1403 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
1405 /* Mishearing the voices in his head, our hero wonders how he's
1406 supposed to kill the mall. */
1407 void nf_conntrack_cleanup(void)
1411 rcu_assign_pointer(ip_ct_attach
, NULL
);
1413 /* This makes sure all current packets have passed through
1414 netfilter framework. Roll on, two-stage module
1418 nf_ct_event_cache_flush();
1420 nf_conntrack_flush();
1421 if (atomic_read(&nf_conntrack_count
) != 0) {
1423 goto i_see_dead_people
;
1425 /* wait until all references to nf_conntrack_untracked are dropped */
1426 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1429 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1431 for (i
= 0; i
< NF_CT_F_NUM
; i
++) {
1432 if (nf_ct_cache
[i
].use
== 0)
1435 NF_CT_ASSERT(nf_ct_cache
[i
].use
== 1);
1436 nf_ct_cache
[i
].use
= 1;
1437 nf_conntrack_unregister_cache(i
);
1439 kmem_cache_destroy(nf_conntrack_expect_cachep
);
1440 free_conntrack_hash(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1441 nf_conntrack_htable_size
);
1443 nf_conntrack_proto_fini();
1446 static struct list_head
*alloc_hashtable(int size
, int *vmalloced
)
1448 struct list_head
*hash
;
1452 hash
= (void*)__get_free_pages(GFP_KERNEL
,
1453 get_order(sizeof(struct list_head
)
1457 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1458 hash
= vmalloc(sizeof(struct list_head
) * size
);
1462 for (i
= 0; i
< size
; i
++)
1463 INIT_LIST_HEAD(&hash
[i
]);
1468 int set_hashsize(const char *val
, struct kernel_param
*kp
)
1470 int i
, bucket
, hashsize
, vmalloced
;
1471 int old_vmalloced
, old_size
;
1473 struct list_head
*hash
, *old_hash
;
1474 struct nf_conntrack_tuple_hash
*h
;
1476 /* On boot, we can set this without any fancy locking. */
1477 if (!nf_conntrack_htable_size
)
1478 return param_set_uint(val
, kp
);
1480 hashsize
= simple_strtol(val
, NULL
, 0);
1484 hash
= alloc_hashtable(hashsize
, &vmalloced
);
1488 /* We have to rehahs for the new table anyway, so we also can
1489 * use a newrandom seed */
1490 get_random_bytes(&rnd
, 4);
1492 write_lock_bh(&nf_conntrack_lock
);
1493 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1494 while (!list_empty(&nf_conntrack_hash
[i
])) {
1495 h
= list_entry(nf_conntrack_hash
[i
].next
,
1496 struct nf_conntrack_tuple_hash
, list
);
1498 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1499 list_add_tail(&h
->list
, &hash
[bucket
]);
1502 old_size
= nf_conntrack_htable_size
;
1503 old_vmalloced
= nf_conntrack_vmalloc
;
1504 old_hash
= nf_conntrack_hash
;
1506 nf_conntrack_htable_size
= hashsize
;
1507 nf_conntrack_vmalloc
= vmalloced
;
1508 nf_conntrack_hash
= hash
;
1509 nf_conntrack_hash_rnd
= rnd
;
1510 write_unlock_bh(&nf_conntrack_lock
);
1512 free_conntrack_hash(old_hash
, old_vmalloced
, old_size
);
1516 module_param_call(hashsize
, set_hashsize
, param_get_uint
,
1517 &nf_conntrack_htable_size
, 0600);
1519 int __init
nf_conntrack_init(void)
1523 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1524 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */
1525 if (!nf_conntrack_htable_size
) {
1526 nf_conntrack_htable_size
1527 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1528 / sizeof(struct list_head
));
1529 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1530 nf_conntrack_htable_size
= 8192;
1531 if (nf_conntrack_htable_size
< 16)
1532 nf_conntrack_htable_size
= 16;
1534 nf_conntrack_max
= 8 * nf_conntrack_htable_size
;
1536 for(ret
=0; ret
<65535; ret
++) //--SZ Angela 09.03 QOS Initialization
1537 port_num_udp
[ret
]=0;
1539 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1540 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1543 nf_conntrack_hash
= alloc_hashtable(nf_conntrack_htable_size
,
1544 &nf_conntrack_vmalloc
);
1545 if (!nf_conntrack_hash
) {
1546 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1550 ret
= nf_conntrack_register_cache(NF_CT_F_BASIC
, "nf_conntrack:basic",
1551 sizeof(struct nf_conn
));
1553 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1557 nf_conntrack_expect_cachep
= kmem_cache_create("nf_conntrack_expect",
1558 sizeof(struct nf_conntrack_expect
),
1560 if (!nf_conntrack_expect_cachep
) {
1561 printk(KERN_ERR
"Unable to create nf_expect slab cache\n");
1562 goto err_free_conntrack_slab
;
1565 ret
= nf_conntrack_proto_init();
1567 goto out_free_expect_slab
;
1569 /* For use by REJECT target */
1570 rcu_assign_pointer(ip_ct_attach
, __nf_conntrack_attach
);
1571 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1573 /* Set up fake conntrack:
1574 - to never be deleted, not in any hashes */
1575 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1576 /* - and look it like as a confirmed connection */
1577 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1581 out_free_expect_slab
:
1582 kmem_cache_destroy(nf_conntrack_expect_cachep
);
1583 err_free_conntrack_slab
:
1584 nf_conntrack_unregister_cache(NF_CT_F_BASIC
);
1586 free_conntrack_hash(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1587 nf_conntrack_htable_size
);