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/sched.h>
18 #include <linux/skbuff.h>
19 #include <linux/proc_fs.h>
20 #include <linux/vmalloc.h>
21 #include <linux/stddef.h>
22 #include <linux/slab.h>
23 #include <linux/random.h>
24 #include <linux/jhash.h>
25 #include <linux/err.h>
26 #include <linux/percpu.h>
27 #include <linux/moduleparam.h>
28 #include <linux/notifier.h>
29 #include <linux/kernel.h>
30 #include <linux/netdevice.h>
31 #include <linux/socket.h>
33 #include <linux/nsproxy.h>
34 #include <linux/rculist_nulls.h>
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>
42 #include <net/netfilter/nf_conntrack_extend.h>
43 #include <net/netfilter/nf_conntrack_acct.h>
44 #include <net/netfilter/nf_conntrack_ecache.h>
45 #include <net/netfilter/nf_nat.h>
46 #include <net/netfilter/nf_nat_core.h>
48 #define NF_CONNTRACK_VERSION "0.5.0"
50 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
51 enum nf_nat_manip_type manip
,
52 const struct nlattr
*attr
) __read_mostly
;
53 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
55 DEFINE_SPINLOCK(nf_conntrack_lock
);
56 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
58 unsigned int nf_conntrack_htable_size __read_mostly
;
59 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
61 unsigned int nf_conntrack_max __read_mostly
;
62 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
64 struct nf_conn nf_conntrack_untracked __read_mostly
;
65 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
67 static int nf_conntrack_hash_rnd_initted
;
68 static unsigned int nf_conntrack_hash_rnd
;
70 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
71 unsigned int size
, unsigned int rnd
)
76 /* The direction must be ignored, so we hash everything up to the
77 * destination ports (which is a multiple of 4) and treat the last
78 * three bytes manually.
80 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
81 h
= jhash2((u32
*)tuple
, n
,
82 rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
83 tuple
->dst
.protonum
));
85 return ((u64
)h
* size
) >> 32;
88 static inline u_int32_t
hash_conntrack(const struct net
*net
,
89 const struct nf_conntrack_tuple
*tuple
)
91 return __hash_conntrack(tuple
, net
->ct
.htable_size
,
92 nf_conntrack_hash_rnd
);
96 nf_ct_get_tuple(const struct sk_buff
*skb
,
101 struct nf_conntrack_tuple
*tuple
,
102 const struct nf_conntrack_l3proto
*l3proto
,
103 const struct nf_conntrack_l4proto
*l4proto
)
105 memset(tuple
, 0, sizeof(*tuple
));
107 tuple
->src
.l3num
= l3num
;
108 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
111 tuple
->dst
.protonum
= protonum
;
112 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
114 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
116 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
118 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
119 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
121 struct nf_conntrack_l3proto
*l3proto
;
122 struct nf_conntrack_l4proto
*l4proto
;
123 unsigned int protoff
;
129 l3proto
= __nf_ct_l3proto_find(l3num
);
130 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
131 if (ret
!= NF_ACCEPT
) {
136 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
138 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
144 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
147 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
148 const struct nf_conntrack_tuple
*orig
,
149 const struct nf_conntrack_l3proto
*l3proto
,
150 const struct nf_conntrack_l4proto
*l4proto
)
152 memset(inverse
, 0, sizeof(*inverse
));
154 inverse
->src
.l3num
= orig
->src
.l3num
;
155 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
158 inverse
->dst
.dir
= !orig
->dst
.dir
;
160 inverse
->dst
.protonum
= orig
->dst
.protonum
;
161 return l4proto
->invert_tuple(inverse
, orig
);
163 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
166 clean_from_lists(struct nf_conn
*ct
)
168 pr_debug("clean_from_lists(%p)\n", ct
);
169 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
170 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
172 /* Destroy all pending expectations */
173 nf_ct_remove_expectations(ct
);
177 destroy_conntrack(struct nf_conntrack
*nfct
)
179 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
180 struct net
*net
= nf_ct_net(ct
);
181 struct nf_conntrack_l4proto
*l4proto
;
183 pr_debug("destroy_conntrack(%p)\n", ct
);
184 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
185 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
187 /* To make sure we don't get any weird locking issues here:
188 * destroy_conntrack() MUST NOT be called with a write lock
189 * to nf_conntrack_lock!!! -HW */
191 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
192 if (l4proto
&& l4proto
->destroy
)
193 l4proto
->destroy(ct
);
197 spin_lock_bh(&nf_conntrack_lock
);
198 /* Expectations will have been removed in clean_from_lists,
199 * except TFTP can create an expectation on the first packet,
200 * before connection is in the list, so we need to clean here,
202 nf_ct_remove_expectations(ct
);
204 /* We overload first tuple to link into unconfirmed list. */
205 if (!nf_ct_is_confirmed(ct
)) {
206 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
207 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
210 NF_CT_STAT_INC(net
, delete);
211 spin_unlock_bh(&nf_conntrack_lock
);
214 nf_ct_put(ct
->master
);
216 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
217 nf_conntrack_free(ct
);
220 void nf_ct_delete_from_lists(struct nf_conn
*ct
)
222 struct net
*net
= nf_ct_net(ct
);
224 nf_ct_helper_destroy(ct
);
225 spin_lock_bh(&nf_conntrack_lock
);
226 /* Inside lock so preempt is disabled on module removal path.
227 * Otherwise we can get spurious warnings. */
228 NF_CT_STAT_INC(net
, delete_list
);
229 clean_from_lists(ct
);
230 spin_unlock_bh(&nf_conntrack_lock
);
232 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists
);
234 static void death_by_event(unsigned long ul_conntrack
)
236 struct nf_conn
*ct
= (void *)ul_conntrack
;
237 struct net
*net
= nf_ct_net(ct
);
239 if (nf_conntrack_event(IPCT_DESTROY
, ct
) < 0) {
240 /* bad luck, let's retry again */
241 ct
->timeout
.expires
= jiffies
+
242 (random32() % net
->ct
.sysctl_events_retry_timeout
);
243 add_timer(&ct
->timeout
);
246 /* we've got the event delivered, now it's dying */
247 set_bit(IPS_DYING_BIT
, &ct
->status
);
248 spin_lock(&nf_conntrack_lock
);
249 hlist_nulls_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
250 spin_unlock(&nf_conntrack_lock
);
254 void nf_ct_insert_dying_list(struct nf_conn
*ct
)
256 struct net
*net
= nf_ct_net(ct
);
258 /* add this conntrack to the dying list */
259 spin_lock_bh(&nf_conntrack_lock
);
260 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
262 spin_unlock_bh(&nf_conntrack_lock
);
263 /* set a new timer to retry event delivery */
264 setup_timer(&ct
->timeout
, death_by_event
, (unsigned long)ct
);
265 ct
->timeout
.expires
= jiffies
+
266 (random32() % net
->ct
.sysctl_events_retry_timeout
);
267 add_timer(&ct
->timeout
);
269 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list
);
271 static void death_by_timeout(unsigned long ul_conntrack
)
273 struct nf_conn
*ct
= (void *)ul_conntrack
;
275 if (!test_bit(IPS_DYING_BIT
, &ct
->status
) &&
276 unlikely(nf_conntrack_event(IPCT_DESTROY
, ct
) < 0)) {
277 /* destroy event was not delivered */
278 nf_ct_delete_from_lists(ct
);
279 nf_ct_insert_dying_list(ct
);
282 set_bit(IPS_DYING_BIT
, &ct
->status
);
283 nf_ct_delete_from_lists(ct
);
289 * - Caller must take a reference on returned object
290 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
292 * - Caller must lock nf_conntrack_lock before calling this function
294 struct nf_conntrack_tuple_hash
*
295 __nf_conntrack_find(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
297 struct nf_conntrack_tuple_hash
*h
;
298 struct hlist_nulls_node
*n
;
299 unsigned int hash
= hash_conntrack(net
, tuple
);
301 /* Disable BHs the entire time since we normally need to disable them
302 * at least once for the stats anyway.
306 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
307 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
308 NF_CT_STAT_INC(net
, found
);
312 NF_CT_STAT_INC(net
, searched
);
315 * if the nulls value we got at the end of this lookup is
316 * not the expected one, we must restart lookup.
317 * We probably met an item that was moved to another chain.
319 if (get_nulls_value(n
) != hash
)
325 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
327 /* Find a connection corresponding to a tuple. */
328 struct nf_conntrack_tuple_hash
*
329 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
331 struct nf_conntrack_tuple_hash
*h
;
336 h
= __nf_conntrack_find(net
, tuple
);
338 ct
= nf_ct_tuplehash_to_ctrack(h
);
339 if (unlikely(nf_ct_is_dying(ct
) ||
340 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
343 if (unlikely(!nf_ct_tuple_equal(tuple
, &h
->tuple
))) {
353 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
355 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
357 unsigned int repl_hash
)
359 struct net
*net
= nf_ct_net(ct
);
361 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
362 &net
->ct
.hash
[hash
]);
363 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
364 &net
->ct
.hash
[repl_hash
]);
367 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
369 struct net
*net
= nf_ct_net(ct
);
370 unsigned int hash
, repl_hash
;
372 hash
= hash_conntrack(net
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
373 repl_hash
= hash_conntrack(net
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
375 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
377 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
379 /* Confirm a connection given skb; places it in hash table */
381 __nf_conntrack_confirm(struct sk_buff
*skb
)
383 unsigned int hash
, repl_hash
;
384 struct nf_conntrack_tuple_hash
*h
;
386 struct nf_conn_help
*help
;
387 struct hlist_nulls_node
*n
;
388 enum ip_conntrack_info ctinfo
;
391 ct
= nf_ct_get(skb
, &ctinfo
);
394 /* ipt_REJECT uses nf_conntrack_attach to attach related
395 ICMP/TCP RST packets in other direction. Actual packet
396 which created connection will be IP_CT_NEW or for an
397 expected connection, IP_CT_RELATED. */
398 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
401 hash
= hash_conntrack(net
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
402 repl_hash
= hash_conntrack(net
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
404 /* We're not in hash table, and we refuse to set up related
405 connections for unconfirmed conns. But packet copies and
406 REJECT will give spurious warnings here. */
407 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
409 /* No external references means noone else could have
411 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
412 pr_debug("Confirming conntrack %p\n", ct
);
414 spin_lock_bh(&nf_conntrack_lock
);
416 /* See if there's one in the list already, including reverse:
417 NAT could have grabbed it without realizing, since we're
418 not in the hash. If there is, we lost race. */
419 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
420 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
423 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnnode
)
424 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
428 /* Remove from unconfirmed list */
429 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
431 /* Timer relative to confirmation time, not original
432 setting time, otherwise we'd get timer wrap in
433 weird delay cases. */
434 ct
->timeout
.expires
+= jiffies
;
435 add_timer(&ct
->timeout
);
436 atomic_inc(&ct
->ct_general
.use
);
437 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
439 /* Since the lookup is lockless, hash insertion must be done after
440 * starting the timer and setting the CONFIRMED bit. The RCU barriers
441 * guarantee that no other CPU can find the conntrack before the above
442 * stores are visible.
444 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
445 NF_CT_STAT_INC(net
, insert
);
446 spin_unlock_bh(&nf_conntrack_lock
);
448 help
= nfct_help(ct
);
449 if (help
&& help
->helper
)
450 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
452 nf_conntrack_event_cache(master_ct(ct
) ?
453 IPCT_RELATED
: IPCT_NEW
, ct
);
457 NF_CT_STAT_INC(net
, insert_failed
);
458 spin_unlock_bh(&nf_conntrack_lock
);
461 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
463 /* Returns true if a connection correspondings to the tuple (required
466 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
467 const struct nf_conn
*ignored_conntrack
)
469 struct net
*net
= nf_ct_net(ignored_conntrack
);
470 struct nf_conntrack_tuple_hash
*h
;
471 struct hlist_nulls_node
*n
;
472 unsigned int hash
= hash_conntrack(net
, tuple
);
474 /* Disable BHs the entire time since we need to disable them at
475 * least once for the stats anyway.
478 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
479 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
480 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
481 NF_CT_STAT_INC(net
, found
);
482 rcu_read_unlock_bh();
485 NF_CT_STAT_INC(net
, searched
);
487 rcu_read_unlock_bh();
491 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
493 #define NF_CT_EVICTION_RANGE 8
495 /* There's a small race here where we may free a just-assured
496 connection. Too bad: we're in trouble anyway. */
497 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
499 /* Use oldest entry, which is roughly LRU */
500 struct nf_conntrack_tuple_hash
*h
;
501 struct nf_conn
*ct
= NULL
, *tmp
;
502 struct hlist_nulls_node
*n
;
503 unsigned int i
, cnt
= 0;
507 for (i
= 0; i
< net
->ct
.htable_size
; i
++) {
508 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
510 tmp
= nf_ct_tuplehash_to_ctrack(h
);
511 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
516 if (ct
&& unlikely(nf_ct_is_dying(ct
) ||
517 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
519 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
522 hash
= (hash
+ 1) % net
->ct
.htable_size
;
529 if (del_timer(&ct
->timeout
)) {
530 death_by_timeout((unsigned long)ct
);
532 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
538 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
539 const struct nf_conntrack_tuple
*orig
,
540 const struct nf_conntrack_tuple
*repl
,
545 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
546 get_random_bytes(&nf_conntrack_hash_rnd
,
547 sizeof(nf_conntrack_hash_rnd
));
548 nf_conntrack_hash_rnd_initted
= 1;
551 /* We don't want any race condition at early drop stage */
552 atomic_inc(&net
->ct
.count
);
554 if (nf_conntrack_max
&&
555 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
556 unsigned int hash
= hash_conntrack(net
, orig
);
557 if (!early_drop(net
, hash
)) {
558 atomic_dec(&net
->ct
.count
);
561 "nf_conntrack: table full, dropping"
563 return ERR_PTR(-ENOMEM
);
568 * Do not use kmem_cache_zalloc(), as this cache uses
569 * SLAB_DESTROY_BY_RCU.
571 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
573 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
574 atomic_dec(&net
->ct
.count
);
575 return ERR_PTR(-ENOMEM
);
578 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
579 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
581 memset(&ct
->tuplehash
[IP_CT_DIR_MAX
], 0,
582 sizeof(*ct
) - offsetof(struct nf_conn
, tuplehash
[IP_CT_DIR_MAX
]));
583 spin_lock_init(&ct
->lock
);
584 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
585 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
586 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
587 ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
= NULL
;
588 /* Don't set timer yet: wait for confirmation */
589 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
595 * changes to lookup keys must be done before setting refcnt to 1
598 atomic_set(&ct
->ct_general
.use
, 1);
601 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
603 void nf_conntrack_free(struct nf_conn
*ct
)
605 struct net
*net
= nf_ct_net(ct
);
607 nf_ct_ext_destroy(ct
);
608 atomic_dec(&net
->ct
.count
);
610 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
612 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
614 /* Allocate a new conntrack: we return -ENOMEM if classification
615 failed due to stress. Otherwise it really is unclassifiable. */
616 static struct nf_conntrack_tuple_hash
*
617 init_conntrack(struct net
*net
,
618 const struct nf_conntrack_tuple
*tuple
,
619 struct nf_conntrack_l3proto
*l3proto
,
620 struct nf_conntrack_l4proto
*l4proto
,
622 unsigned int dataoff
)
625 struct nf_conn_help
*help
;
626 struct nf_conntrack_tuple repl_tuple
;
627 struct nf_conntrack_expect
*exp
;
629 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
630 pr_debug("Can't invert tuple.\n");
634 ct
= nf_conntrack_alloc(net
, tuple
, &repl_tuple
, GFP_ATOMIC
);
636 pr_debug("Can't allocate conntrack.\n");
637 return (struct nf_conntrack_tuple_hash
*)ct
;
640 if (!l4proto
->new(ct
, skb
, dataoff
)) {
641 nf_conntrack_free(ct
);
642 pr_debug("init conntrack: can't track with proto module\n");
646 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
647 nf_ct_ecache_ext_add(ct
, GFP_ATOMIC
);
649 spin_lock_bh(&nf_conntrack_lock
);
650 exp
= nf_ct_find_expectation(net
, tuple
);
652 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
654 /* Welcome, Mr. Bond. We've been expecting you... */
655 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
656 ct
->master
= exp
->master
;
658 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
660 rcu_assign_pointer(help
->helper
, exp
->helper
);
663 #ifdef CONFIG_NF_CONNTRACK_MARK
664 ct
->mark
= exp
->master
->mark
;
666 #ifdef CONFIG_NF_CONNTRACK_SECMARK
667 ct
->secmark
= exp
->master
->secmark
;
669 nf_conntrack_get(&ct
->master
->ct_general
);
670 NF_CT_STAT_INC(net
, expect_new
);
672 __nf_ct_try_assign_helper(ct
, GFP_ATOMIC
);
673 NF_CT_STAT_INC(net
, new);
676 /* Overload tuple linked list to put us in unconfirmed list. */
677 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
678 &net
->ct
.unconfirmed
);
680 spin_unlock_bh(&nf_conntrack_lock
);
684 exp
->expectfn(ct
, exp
);
685 nf_ct_expect_put(exp
);
688 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
691 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
692 static inline struct nf_conn
*
693 resolve_normal_ct(struct net
*net
,
695 unsigned int dataoff
,
698 struct nf_conntrack_l3proto
*l3proto
,
699 struct nf_conntrack_l4proto
*l4proto
,
701 enum ip_conntrack_info
*ctinfo
)
703 struct nf_conntrack_tuple tuple
;
704 struct nf_conntrack_tuple_hash
*h
;
707 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
708 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
710 pr_debug("resolve_normal_ct: Can't get tuple\n");
714 /* look for tuple match */
715 h
= nf_conntrack_find_get(net
, &tuple
);
717 h
= init_conntrack(net
, &tuple
, l3proto
, l4proto
, skb
, dataoff
);
723 ct
= nf_ct_tuplehash_to_ctrack(h
);
725 /* It exists; we have (non-exclusive) reference. */
726 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
727 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
728 /* Please set reply bit if this packet OK */
731 /* Once we've had two way comms, always ESTABLISHED. */
732 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
733 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
734 *ctinfo
= IP_CT_ESTABLISHED
;
735 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
736 pr_debug("nf_conntrack_in: related packet for %p\n",
738 *ctinfo
= IP_CT_RELATED
;
740 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
745 skb
->nfct
= &ct
->ct_general
;
746 skb
->nfctinfo
= *ctinfo
;
751 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
755 enum ip_conntrack_info ctinfo
;
756 struct nf_conntrack_l3proto
*l3proto
;
757 struct nf_conntrack_l4proto
*l4proto
;
758 unsigned int dataoff
;
763 /* Previously seen (loopback or untracked)? Ignore. */
765 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
769 /* rcu_read_lock()ed by nf_hook_slow */
770 l3proto
= __nf_ct_l3proto_find(pf
);
771 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
772 &dataoff
, &protonum
);
774 pr_debug("not prepared to track yet or error occured\n");
775 NF_CT_STAT_INC_ATOMIC(net
, error
);
776 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
780 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
782 /* It may be an special packet, error, unclean...
783 * inverse of the return code tells to the netfilter
784 * core what to do with the packet. */
785 if (l4proto
->error
!= NULL
) {
786 ret
= l4proto
->error(net
, skb
, dataoff
, &ctinfo
, pf
, hooknum
);
788 NF_CT_STAT_INC_ATOMIC(net
, error
);
789 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
794 ct
= resolve_normal_ct(net
, skb
, dataoff
, pf
, protonum
,
795 l3proto
, l4proto
, &set_reply
, &ctinfo
);
797 /* Not valid part of a connection */
798 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
803 /* Too stressed to deal. */
804 NF_CT_STAT_INC_ATOMIC(net
, drop
);
808 NF_CT_ASSERT(skb
->nfct
);
810 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
812 /* Invalid: inverse of the return code tells
813 * the netfilter core what to do */
814 pr_debug("nf_conntrack_in: Can't track with proto module\n");
815 nf_conntrack_put(skb
->nfct
);
817 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
819 NF_CT_STAT_INC_ATOMIC(net
, drop
);
823 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
824 nf_conntrack_event_cache(IPCT_STATUS
, ct
);
828 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
830 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
831 const struct nf_conntrack_tuple
*orig
)
836 ret
= nf_ct_invert_tuple(inverse
, orig
,
837 __nf_ct_l3proto_find(orig
->src
.l3num
),
838 __nf_ct_l4proto_find(orig
->src
.l3num
,
839 orig
->dst
.protonum
));
843 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
845 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
846 implicitly racy: see __nf_conntrack_confirm */
847 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
848 const struct nf_conntrack_tuple
*newreply
)
850 struct nf_conn_help
*help
= nfct_help(ct
);
852 /* Should be unconfirmed, so not in hash table yet */
853 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
855 pr_debug("Altering reply tuple of %p to ", ct
);
856 nf_ct_dump_tuple(newreply
);
858 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
859 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
863 __nf_ct_try_assign_helper(ct
, GFP_ATOMIC
);
866 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
868 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
869 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
870 enum ip_conntrack_info ctinfo
,
871 const struct sk_buff
*skb
,
872 unsigned long extra_jiffies
,
875 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
878 /* Only update if this is not a fixed timeout */
879 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
882 /* If not in hash table, timer will not be active yet */
883 if (!nf_ct_is_confirmed(ct
)) {
884 ct
->timeout
.expires
= extra_jiffies
;
886 unsigned long newtime
= jiffies
+ extra_jiffies
;
888 /* Only update the timeout if the new timeout is at least
889 HZ jiffies from the old timeout. Need del_timer for race
890 avoidance (may already be dying). */
891 if (newtime
- ct
->timeout
.expires
>= HZ
)
892 mod_timer_pending(&ct
->timeout
, newtime
);
897 struct nf_conn_counter
*acct
;
899 acct
= nf_conn_acct_find(ct
);
901 spin_lock_bh(&ct
->lock
);
902 acct
[CTINFO2DIR(ctinfo
)].packets
++;
903 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
904 skb
->len
- skb_network_offset(skb
);
905 spin_unlock_bh(&ct
->lock
);
909 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
911 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
912 enum ip_conntrack_info ctinfo
,
913 const struct sk_buff
*skb
,
917 struct nf_conn_counter
*acct
;
919 acct
= nf_conn_acct_find(ct
);
921 spin_lock_bh(&ct
->lock
);
922 acct
[CTINFO2DIR(ctinfo
)].packets
++;
923 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
924 skb
->len
- skb_network_offset(skb
);
925 spin_unlock_bh(&ct
->lock
);
929 if (del_timer(&ct
->timeout
)) {
930 ct
->timeout
.function((unsigned long)ct
);
935 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
937 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
939 #include <linux/netfilter/nfnetlink.h>
940 #include <linux/netfilter/nfnetlink_conntrack.h>
941 #include <linux/mutex.h>
943 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
944 * in ip_conntrack_core, since we don't want the protocols to autoload
945 * or depend on ctnetlink */
946 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
947 const struct nf_conntrack_tuple
*tuple
)
949 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
950 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
956 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
958 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
959 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
960 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
962 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
964 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
965 struct nf_conntrack_tuple
*t
)
967 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
970 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
971 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
975 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
977 int nf_ct_port_nlattr_tuple_size(void)
979 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
981 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
984 /* Used by ipt_REJECT and ip6t_REJECT. */
985 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
988 enum ip_conntrack_info ctinfo
;
990 /* This ICMP is in reverse direction to the packet which caused it */
991 ct
= nf_ct_get(skb
, &ctinfo
);
992 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
993 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
995 ctinfo
= IP_CT_RELATED
;
997 /* Attach to new skbuff, and increment count */
998 nskb
->nfct
= &ct
->ct_general
;
999 nskb
->nfctinfo
= ctinfo
;
1000 nf_conntrack_get(nskb
->nfct
);
1003 /* Bring out ya dead! */
1004 static struct nf_conn
*
1005 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1006 void *data
, unsigned int *bucket
)
1008 struct nf_conntrack_tuple_hash
*h
;
1010 struct hlist_nulls_node
*n
;
1012 spin_lock_bh(&nf_conntrack_lock
);
1013 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1014 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1015 ct
= nf_ct_tuplehash_to_ctrack(h
);
1020 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnnode
) {
1021 ct
= nf_ct_tuplehash_to_ctrack(h
);
1023 set_bit(IPS_DYING_BIT
, &ct
->status
);
1025 spin_unlock_bh(&nf_conntrack_lock
);
1028 atomic_inc(&ct
->ct_general
.use
);
1029 spin_unlock_bh(&nf_conntrack_lock
);
1033 void nf_ct_iterate_cleanup(struct net
*net
,
1034 int (*iter
)(struct nf_conn
*i
, void *data
),
1038 unsigned int bucket
= 0;
1040 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1041 /* Time to push up daises... */
1042 if (del_timer(&ct
->timeout
))
1043 death_by_timeout((unsigned long)ct
);
1044 /* ... else the timer will get him soon. */
1049 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1051 struct __nf_ct_flush_report
{
1056 static int kill_report(struct nf_conn
*i
, void *data
)
1058 struct __nf_ct_flush_report
*fr
= (struct __nf_ct_flush_report
*)data
;
1060 /* If we fail to deliver the event, death_by_timeout() will retry */
1061 if (nf_conntrack_event_report(IPCT_DESTROY
, i
,
1062 fr
->pid
, fr
->report
) < 0)
1065 /* Avoid the delivery of the destroy event in death_by_timeout(). */
1066 set_bit(IPS_DYING_BIT
, &i
->status
);
1070 static int kill_all(struct nf_conn
*i
, void *data
)
1075 void nf_ct_free_hashtable(void *hash
, int vmalloced
, unsigned int size
)
1080 free_pages((unsigned long)hash
,
1081 get_order(sizeof(struct hlist_head
) * size
));
1083 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1085 void nf_conntrack_flush_report(struct net
*net
, u32 pid
, int report
)
1087 struct __nf_ct_flush_report fr
= {
1091 nf_ct_iterate_cleanup(net
, kill_report
, &fr
);
1093 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report
);
1095 static void nf_ct_release_dying_list(struct net
*net
)
1097 struct nf_conntrack_tuple_hash
*h
;
1099 struct hlist_nulls_node
*n
;
1101 spin_lock_bh(&nf_conntrack_lock
);
1102 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.dying
, hnnode
) {
1103 ct
= nf_ct_tuplehash_to_ctrack(h
);
1104 /* never fails to remove them, no listeners at this point */
1107 spin_unlock_bh(&nf_conntrack_lock
);
1110 static void nf_conntrack_cleanup_init_net(void)
1112 /* wait until all references to nf_conntrack_untracked are dropped */
1113 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1116 nf_conntrack_helper_fini();
1117 nf_conntrack_proto_fini();
1120 static void nf_conntrack_cleanup_net(struct net
*net
)
1123 nf_ct_iterate_cleanup(net
, kill_all
, NULL
);
1124 nf_ct_release_dying_list(net
);
1125 if (atomic_read(&net
->ct
.count
) != 0) {
1127 goto i_see_dead_people
;
1130 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1131 net
->ct
.htable_size
);
1132 nf_conntrack_ecache_fini(net
);
1133 nf_conntrack_acct_fini(net
);
1134 nf_conntrack_expect_fini(net
);
1135 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1136 kfree(net
->ct
.slabname
);
1137 free_percpu(net
->ct
.stat
);
1140 /* Mishearing the voices in his head, our hero wonders how he's
1141 supposed to kill the mall. */
1142 void nf_conntrack_cleanup(struct net
*net
)
1144 if (net_eq(net
, &init_net
))
1145 rcu_assign_pointer(ip_ct_attach
, NULL
);
1147 /* This makes sure all current packets have passed through
1148 netfilter framework. Roll on, two-stage module
1152 nf_conntrack_cleanup_net(net
);
1154 if (net_eq(net
, &init_net
)) {
1155 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1156 nf_conntrack_cleanup_init_net();
1160 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
, int nulls
)
1162 struct hlist_nulls_head
*hash
;
1163 unsigned int nr_slots
, i
;
1168 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1169 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1170 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1171 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1175 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1176 hash
= __vmalloc(sz
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
1181 for (i
= 0; i
< nr_slots
; i
++)
1182 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1186 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1188 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1190 int i
, bucket
, vmalloced
, old_vmalloced
;
1191 unsigned int hashsize
, old_size
;
1192 struct hlist_nulls_head
*hash
, *old_hash
;
1193 struct nf_conntrack_tuple_hash
*h
;
1195 if (current
->nsproxy
->net_ns
!= &init_net
)
1198 /* On boot, we can set this without any fancy locking. */
1199 if (!nf_conntrack_htable_size
)
1200 return param_set_uint(val
, kp
);
1202 hashsize
= simple_strtoul(val
, NULL
, 0);
1206 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
, 1);
1210 /* Lookups in the old hash might happen in parallel, which means we
1211 * might get false negatives during connection lookup. New connections
1212 * created because of a false negative won't make it into the hash
1213 * though since that required taking the lock.
1215 spin_lock_bh(&nf_conntrack_lock
);
1216 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1217 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1218 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1219 struct nf_conntrack_tuple_hash
, hnnode
);
1220 hlist_nulls_del_rcu(&h
->hnnode
);
1221 bucket
= __hash_conntrack(&h
->tuple
, hashsize
,
1222 nf_conntrack_hash_rnd
);
1223 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1226 old_size
= init_net
.ct
.htable_size
;
1227 old_vmalloced
= init_net
.ct
.hash_vmalloc
;
1228 old_hash
= init_net
.ct
.hash
;
1230 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1231 init_net
.ct
.hash_vmalloc
= vmalloced
;
1232 init_net
.ct
.hash
= hash
;
1233 spin_unlock_bh(&nf_conntrack_lock
);
1235 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1238 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1240 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1241 &nf_conntrack_htable_size
, 0600);
1243 static int nf_conntrack_init_init_net(void)
1248 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1249 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1250 if (!nf_conntrack_htable_size
) {
1251 nf_conntrack_htable_size
1252 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1253 / sizeof(struct hlist_head
));
1254 if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1255 nf_conntrack_htable_size
= 16384;
1256 if (nf_conntrack_htable_size
< 32)
1257 nf_conntrack_htable_size
= 32;
1259 /* Use a max. factor of four by default to get the same max as
1260 * with the old struct list_heads. When a table size is given
1261 * we use the old value of 8 to avoid reducing the max.
1265 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1267 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1268 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1271 ret
= nf_conntrack_proto_init();
1275 ret
= nf_conntrack_helper_init();
1279 /* Set up fake conntrack: to never be deleted, not in any hashes */
1280 #ifdef CONFIG_NET_NS
1281 nf_conntrack_untracked
.ct_net
= &init_net
;
1283 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1284 /* - and look it like as a confirmed connection */
1285 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1290 nf_conntrack_proto_fini();
1296 * We need to use special "null" values, not used in hash table
1298 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1299 #define DYING_NULLS_VAL ((1<<30)+1)
1301 static int nf_conntrack_init_net(struct net
*net
)
1305 atomic_set(&net
->ct
.count
, 0);
1306 INIT_HLIST_NULLS_HEAD(&net
->ct
.unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1307 INIT_HLIST_NULLS_HEAD(&net
->ct
.dying
, DYING_NULLS_VAL
);
1308 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1309 if (!net
->ct
.stat
) {
1314 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1315 if (!net
->ct
.slabname
) {
1320 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1321 sizeof(struct nf_conn
), 0,
1322 SLAB_DESTROY_BY_RCU
, NULL
);
1323 if (!net
->ct
.nf_conntrack_cachep
) {
1324 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1329 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1330 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
,
1331 &net
->ct
.hash_vmalloc
, 1);
1332 if (!net
->ct
.hash
) {
1334 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1337 ret
= nf_conntrack_expect_init(net
);
1340 ret
= nf_conntrack_acct_init(net
);
1343 ret
= nf_conntrack_ecache_init(net
);
1350 nf_conntrack_acct_fini(net
);
1352 nf_conntrack_expect_fini(net
);
1354 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1355 net
->ct
.htable_size
);
1357 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1359 kfree(net
->ct
.slabname
);
1361 free_percpu(net
->ct
.stat
);
1366 s16 (*nf_ct_nat_offset
)(const struct nf_conn
*ct
,
1367 enum ip_conntrack_dir dir
,
1369 EXPORT_SYMBOL_GPL(nf_ct_nat_offset
);
1371 int nf_conntrack_init(struct net
*net
)
1375 if (net_eq(net
, &init_net
)) {
1376 ret
= nf_conntrack_init_init_net();
1380 ret
= nf_conntrack_init_net(net
);
1384 if (net_eq(net
, &init_net
)) {
1385 /* For use by REJECT target */
1386 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1387 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1389 /* Howto get NAT offsets */
1390 rcu_assign_pointer(nf_ct_nat_offset
, NULL
);
1395 if (net_eq(net
, &init_net
))
1396 nf_conntrack_cleanup_init_net();