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_conntrack_zones.h>
46 #include <net/netfilter/nf_nat.h>
47 #include <net/netfilter/nf_nat_core.h>
49 #define NF_CONNTRACK_VERSION "0.5.0"
51 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
52 enum nf_nat_manip_type manip
,
53 const struct nlattr
*attr
) __read_mostly
;
54 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
56 DEFINE_SPINLOCK(nf_conntrack_lock
);
57 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
59 unsigned int nf_conntrack_htable_size __read_mostly
;
60 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
62 unsigned int nf_conntrack_max __read_mostly
;
63 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
65 struct nf_conn nf_conntrack_untracked __read_mostly
;
66 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
68 static int nf_conntrack_hash_rnd_initted
;
69 static unsigned int nf_conntrack_hash_rnd
;
71 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
72 u16 zone
, unsigned int size
, unsigned int rnd
)
77 /* The direction must be ignored, so we hash everything up to the
78 * destination ports (which is a multiple of 4) and treat the last
79 * three bytes manually.
81 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
82 h
= jhash2((u32
*)tuple
, n
,
83 zone
^ rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
84 tuple
->dst
.protonum
));
86 return ((u64
)h
* size
) >> 32;
89 static inline u_int32_t
hash_conntrack(const struct net
*net
, u16 zone
,
90 const struct nf_conntrack_tuple
*tuple
)
92 return __hash_conntrack(tuple
, zone
, net
->ct
.htable_size
,
93 nf_conntrack_hash_rnd
);
97 nf_ct_get_tuple(const struct sk_buff
*skb
,
102 struct nf_conntrack_tuple
*tuple
,
103 const struct nf_conntrack_l3proto
*l3proto
,
104 const struct nf_conntrack_l4proto
*l4proto
)
106 memset(tuple
, 0, sizeof(*tuple
));
108 tuple
->src
.l3num
= l3num
;
109 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
112 tuple
->dst
.protonum
= protonum
;
113 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
115 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
117 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
119 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
120 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
122 struct nf_conntrack_l3proto
*l3proto
;
123 struct nf_conntrack_l4proto
*l4proto
;
124 unsigned int protoff
;
130 l3proto
= __nf_ct_l3proto_find(l3num
);
131 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
132 if (ret
!= NF_ACCEPT
) {
137 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
139 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
145 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
148 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
149 const struct nf_conntrack_tuple
*orig
,
150 const struct nf_conntrack_l3proto
*l3proto
,
151 const struct nf_conntrack_l4proto
*l4proto
)
153 memset(inverse
, 0, sizeof(*inverse
));
155 inverse
->src
.l3num
= orig
->src
.l3num
;
156 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
159 inverse
->dst
.dir
= !orig
->dst
.dir
;
161 inverse
->dst
.protonum
= orig
->dst
.protonum
;
162 return l4proto
->invert_tuple(inverse
, orig
);
164 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
167 clean_from_lists(struct nf_conn
*ct
)
169 pr_debug("clean_from_lists(%p)\n", ct
);
170 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
171 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
173 /* Destroy all pending expectations */
174 nf_ct_remove_expectations(ct
);
178 destroy_conntrack(struct nf_conntrack
*nfct
)
180 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
181 struct net
*net
= nf_ct_net(ct
);
182 struct nf_conntrack_l4proto
*l4proto
;
184 pr_debug("destroy_conntrack(%p)\n", ct
);
185 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
186 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
188 /* To make sure we don't get any weird locking issues here:
189 * destroy_conntrack() MUST NOT be called with a write lock
190 * to nf_conntrack_lock!!! -HW */
192 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
193 if (l4proto
&& l4proto
->destroy
)
194 l4proto
->destroy(ct
);
198 spin_lock_bh(&nf_conntrack_lock
);
199 /* Expectations will have been removed in clean_from_lists,
200 * except TFTP can create an expectation on the first packet,
201 * before connection is in the list, so we need to clean here,
203 nf_ct_remove_expectations(ct
);
205 /* We overload first tuple to link into unconfirmed list. */
206 if (!nf_ct_is_confirmed(ct
)) {
207 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
208 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
211 NF_CT_STAT_INC(net
, delete);
212 spin_unlock_bh(&nf_conntrack_lock
);
215 nf_ct_put(ct
->master
);
217 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
218 nf_conntrack_free(ct
);
221 void nf_ct_delete_from_lists(struct nf_conn
*ct
)
223 struct net
*net
= nf_ct_net(ct
);
225 nf_ct_helper_destroy(ct
);
226 spin_lock_bh(&nf_conntrack_lock
);
227 /* Inside lock so preempt is disabled on module removal path.
228 * Otherwise we can get spurious warnings. */
229 NF_CT_STAT_INC(net
, delete_list
);
230 clean_from_lists(ct
);
231 spin_unlock_bh(&nf_conntrack_lock
);
233 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists
);
235 static void death_by_event(unsigned long ul_conntrack
)
237 struct nf_conn
*ct
= (void *)ul_conntrack
;
238 struct net
*net
= nf_ct_net(ct
);
240 if (nf_conntrack_event(IPCT_DESTROY
, ct
) < 0) {
241 /* bad luck, let's retry again */
242 ct
->timeout
.expires
= jiffies
+
243 (random32() % net
->ct
.sysctl_events_retry_timeout
);
244 add_timer(&ct
->timeout
);
247 /* we've got the event delivered, now it's dying */
248 set_bit(IPS_DYING_BIT
, &ct
->status
);
249 spin_lock(&nf_conntrack_lock
);
250 hlist_nulls_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
251 spin_unlock(&nf_conntrack_lock
);
255 void nf_ct_insert_dying_list(struct nf_conn
*ct
)
257 struct net
*net
= nf_ct_net(ct
);
259 /* add this conntrack to the dying list */
260 spin_lock_bh(&nf_conntrack_lock
);
261 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
263 spin_unlock_bh(&nf_conntrack_lock
);
264 /* set a new timer to retry event delivery */
265 setup_timer(&ct
->timeout
, death_by_event
, (unsigned long)ct
);
266 ct
->timeout
.expires
= jiffies
+
267 (random32() % net
->ct
.sysctl_events_retry_timeout
);
268 add_timer(&ct
->timeout
);
270 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list
);
272 static void death_by_timeout(unsigned long ul_conntrack
)
274 struct nf_conn
*ct
= (void *)ul_conntrack
;
276 if (!test_bit(IPS_DYING_BIT
, &ct
->status
) &&
277 unlikely(nf_conntrack_event(IPCT_DESTROY
, ct
) < 0)) {
278 /* destroy event was not delivered */
279 nf_ct_delete_from_lists(ct
);
280 nf_ct_insert_dying_list(ct
);
283 set_bit(IPS_DYING_BIT
, &ct
->status
);
284 nf_ct_delete_from_lists(ct
);
290 * - Caller must take a reference on returned object
291 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
293 * - Caller must lock nf_conntrack_lock before calling this function
295 struct nf_conntrack_tuple_hash
*
296 __nf_conntrack_find(struct net
*net
, u16 zone
,
297 const struct nf_conntrack_tuple
*tuple
)
299 struct nf_conntrack_tuple_hash
*h
;
300 struct hlist_nulls_node
*n
;
301 unsigned int hash
= hash_conntrack(net
, zone
, tuple
);
303 /* Disable BHs the entire time since we normally need to disable them
304 * at least once for the stats anyway.
308 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
309 if (nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
310 nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)) == zone
) {
311 NF_CT_STAT_INC(net
, found
);
315 NF_CT_STAT_INC(net
, searched
);
318 * if the nulls value we got at the end of this lookup is
319 * not the expected one, we must restart lookup.
320 * We probably met an item that was moved to another chain.
322 if (get_nulls_value(n
) != hash
) {
323 NF_CT_STAT_INC(net
, search_restart
);
330 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
332 /* Find a connection corresponding to a tuple. */
333 struct nf_conntrack_tuple_hash
*
334 nf_conntrack_find_get(struct net
*net
, u16 zone
,
335 const struct nf_conntrack_tuple
*tuple
)
337 struct nf_conntrack_tuple_hash
*h
;
342 h
= __nf_conntrack_find(net
, zone
, tuple
);
344 ct
= nf_ct_tuplehash_to_ctrack(h
);
345 if (unlikely(nf_ct_is_dying(ct
) ||
346 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
349 if (unlikely(!nf_ct_tuple_equal(tuple
, &h
->tuple
) ||
350 nf_ct_zone(ct
) != zone
)) {
360 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
362 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
364 unsigned int repl_hash
)
366 struct net
*net
= nf_ct_net(ct
);
368 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
369 &net
->ct
.hash
[hash
]);
370 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
371 &net
->ct
.hash
[repl_hash
]);
374 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
376 struct net
*net
= nf_ct_net(ct
);
377 unsigned int hash
, repl_hash
;
380 zone
= nf_ct_zone(ct
);
381 hash
= hash_conntrack(net
, zone
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
382 repl_hash
= hash_conntrack(net
, zone
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
384 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
386 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
388 /* Confirm a connection given skb; places it in hash table */
390 __nf_conntrack_confirm(struct sk_buff
*skb
)
392 unsigned int hash
, repl_hash
;
393 struct nf_conntrack_tuple_hash
*h
;
395 struct nf_conn_help
*help
;
396 struct hlist_nulls_node
*n
;
397 enum ip_conntrack_info ctinfo
;
401 ct
= nf_ct_get(skb
, &ctinfo
);
404 /* ipt_REJECT uses nf_conntrack_attach to attach related
405 ICMP/TCP RST packets in other direction. Actual packet
406 which created connection will be IP_CT_NEW or for an
407 expected connection, IP_CT_RELATED. */
408 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
411 zone
= nf_ct_zone(ct
);
412 hash
= hash_conntrack(net
, zone
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
413 repl_hash
= hash_conntrack(net
, zone
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
415 /* We're not in hash table, and we refuse to set up related
416 connections for unconfirmed conns. But packet copies and
417 REJECT will give spurious warnings here. */
418 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
420 /* No external references means noone else could have
422 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
423 pr_debug("Confirming conntrack %p\n", ct
);
425 spin_lock_bh(&nf_conntrack_lock
);
427 /* We have to check the DYING flag inside the lock to prevent
428 a race against nf_ct_get_next_corpse() possibly called from
429 user context, else we insert an already 'dead' hash, blocking
430 further use of that particular connection -JM */
432 if (unlikely(nf_ct_is_dying(ct
))) {
433 spin_unlock_bh(&nf_conntrack_lock
);
437 /* See if there's one in the list already, including reverse:
438 NAT could have grabbed it without realizing, since we're
439 not in the hash. If there is, we lost race. */
440 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
441 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
443 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
445 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnnode
)
446 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
448 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
451 /* Remove from unconfirmed list */
452 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
454 /* Timer relative to confirmation time, not original
455 setting time, otherwise we'd get timer wrap in
456 weird delay cases. */
457 ct
->timeout
.expires
+= jiffies
;
458 add_timer(&ct
->timeout
);
459 atomic_inc(&ct
->ct_general
.use
);
460 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
462 /* Since the lookup is lockless, hash insertion must be done after
463 * starting the timer and setting the CONFIRMED bit. The RCU barriers
464 * guarantee that no other CPU can find the conntrack before the above
465 * stores are visible.
467 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
468 NF_CT_STAT_INC(net
, insert
);
469 spin_unlock_bh(&nf_conntrack_lock
);
471 help
= nfct_help(ct
);
472 if (help
&& help
->helper
)
473 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
475 nf_conntrack_event_cache(master_ct(ct
) ?
476 IPCT_RELATED
: IPCT_NEW
, ct
);
480 NF_CT_STAT_INC(net
, insert_failed
);
481 spin_unlock_bh(&nf_conntrack_lock
);
484 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
486 /* Returns true if a connection correspondings to the tuple (required
489 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
490 const struct nf_conn
*ignored_conntrack
)
492 struct net
*net
= nf_ct_net(ignored_conntrack
);
493 struct nf_conntrack_tuple_hash
*h
;
494 struct hlist_nulls_node
*n
;
496 u16 zone
= nf_ct_zone(ignored_conntrack
);
497 unsigned int hash
= hash_conntrack(net
, zone
, tuple
);
499 /* Disable BHs the entire time since we need to disable them at
500 * least once for the stats anyway.
503 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
504 ct
= nf_ct_tuplehash_to_ctrack(h
);
505 if (ct
!= ignored_conntrack
&&
506 nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
507 nf_ct_zone(ct
) == zone
) {
508 NF_CT_STAT_INC(net
, found
);
509 rcu_read_unlock_bh();
512 NF_CT_STAT_INC(net
, searched
);
514 rcu_read_unlock_bh();
518 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
520 #define NF_CT_EVICTION_RANGE 8
522 /* There's a small race here where we may free a just-assured
523 connection. Too bad: we're in trouble anyway. */
524 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
526 /* Use oldest entry, which is roughly LRU */
527 struct nf_conntrack_tuple_hash
*h
;
528 struct nf_conn
*ct
= NULL
, *tmp
;
529 struct hlist_nulls_node
*n
;
530 unsigned int i
, cnt
= 0;
534 for (i
= 0; i
< net
->ct
.htable_size
; i
++) {
535 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
537 tmp
= nf_ct_tuplehash_to_ctrack(h
);
538 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
544 if (likely(!nf_ct_is_dying(ct
) &&
545 atomic_inc_not_zero(&ct
->ct_general
.use
)))
551 if (cnt
>= NF_CT_EVICTION_RANGE
)
554 hash
= (hash
+ 1) % net
->ct
.htable_size
;
561 if (del_timer(&ct
->timeout
)) {
562 death_by_timeout((unsigned long)ct
);
564 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
570 struct nf_conn
*nf_conntrack_alloc(struct net
*net
, u16 zone
,
571 const struct nf_conntrack_tuple
*orig
,
572 const struct nf_conntrack_tuple
*repl
,
577 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
578 get_random_bytes(&nf_conntrack_hash_rnd
,
579 sizeof(nf_conntrack_hash_rnd
));
580 nf_conntrack_hash_rnd_initted
= 1;
583 /* We don't want any race condition at early drop stage */
584 atomic_inc(&net
->ct
.count
);
586 if (nf_conntrack_max
&&
587 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
588 unsigned int hash
= hash_conntrack(net
, zone
, orig
);
589 if (!early_drop(net
, hash
)) {
590 atomic_dec(&net
->ct
.count
);
593 "nf_conntrack: table full, dropping"
595 return ERR_PTR(-ENOMEM
);
600 * Do not use kmem_cache_zalloc(), as this cache uses
601 * SLAB_DESTROY_BY_RCU.
603 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
605 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
606 atomic_dec(&net
->ct
.count
);
607 return ERR_PTR(-ENOMEM
);
610 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
611 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
613 memset(&ct
->tuplehash
[IP_CT_DIR_MAX
], 0,
614 sizeof(*ct
) - offsetof(struct nf_conn
, tuplehash
[IP_CT_DIR_MAX
]));
615 spin_lock_init(&ct
->lock
);
616 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
617 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
618 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
619 ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
= NULL
;
620 /* Don't set timer yet: wait for confirmation */
621 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
625 #ifdef CONFIG_NF_CONNTRACK_ZONES
627 struct nf_conntrack_zone
*nf_ct_zone
;
629 nf_ct_zone
= nf_ct_ext_add(ct
, NF_CT_EXT_ZONE
, GFP_ATOMIC
);
632 nf_ct_zone
->id
= zone
;
636 * changes to lookup keys must be done before setting refcnt to 1
639 atomic_set(&ct
->ct_general
.use
, 1);
642 #ifdef CONFIG_NF_CONNTRACK_ZONES
644 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
645 return ERR_PTR(-ENOMEM
);
648 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
650 void nf_conntrack_free(struct nf_conn
*ct
)
652 struct net
*net
= nf_ct_net(ct
);
654 nf_ct_ext_destroy(ct
);
655 atomic_dec(&net
->ct
.count
);
657 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
659 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
661 /* Allocate a new conntrack: we return -ENOMEM if classification
662 failed due to stress. Otherwise it really is unclassifiable. */
663 static struct nf_conntrack_tuple_hash
*
664 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
665 const struct nf_conntrack_tuple
*tuple
,
666 struct nf_conntrack_l3proto
*l3proto
,
667 struct nf_conntrack_l4proto
*l4proto
,
669 unsigned int dataoff
)
672 struct nf_conn_help
*help
;
673 struct nf_conntrack_tuple repl_tuple
;
674 struct nf_conntrack_ecache
*ecache
;
675 struct nf_conntrack_expect
*exp
;
676 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
678 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
679 pr_debug("Can't invert tuple.\n");
683 ct
= nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
);
685 pr_debug("Can't allocate conntrack.\n");
686 return (struct nf_conntrack_tuple_hash
*)ct
;
689 if (!l4proto
->new(ct
, skb
, dataoff
)) {
690 nf_conntrack_free(ct
);
691 pr_debug("init conntrack: can't track with proto module\n");
695 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
697 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
698 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
699 ecache
? ecache
->expmask
: 0,
702 spin_lock_bh(&nf_conntrack_lock
);
703 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
705 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
707 /* Welcome, Mr. Bond. We've been expecting you... */
708 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
709 ct
->master
= exp
->master
;
711 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
713 rcu_assign_pointer(help
->helper
, exp
->helper
);
716 #ifdef CONFIG_NF_CONNTRACK_MARK
717 ct
->mark
= exp
->master
->mark
;
719 #ifdef CONFIG_NF_CONNTRACK_SECMARK
720 ct
->secmark
= exp
->master
->secmark
;
722 nf_conntrack_get(&ct
->master
->ct_general
);
723 NF_CT_STAT_INC(net
, expect_new
);
725 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
726 NF_CT_STAT_INC(net
, new);
729 /* Overload tuple linked list to put us in unconfirmed list. */
730 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
731 &net
->ct
.unconfirmed
);
733 spin_unlock_bh(&nf_conntrack_lock
);
737 exp
->expectfn(ct
, exp
);
738 nf_ct_expect_put(exp
);
741 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
744 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
745 static inline struct nf_conn
*
746 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
748 unsigned int dataoff
,
751 struct nf_conntrack_l3proto
*l3proto
,
752 struct nf_conntrack_l4proto
*l4proto
,
754 enum ip_conntrack_info
*ctinfo
)
756 struct nf_conntrack_tuple tuple
;
757 struct nf_conntrack_tuple_hash
*h
;
759 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
761 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
762 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
764 pr_debug("resolve_normal_ct: Can't get tuple\n");
768 /* look for tuple match */
769 h
= nf_conntrack_find_get(net
, zone
, &tuple
);
771 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
778 ct
= nf_ct_tuplehash_to_ctrack(h
);
780 /* It exists; we have (non-exclusive) reference. */
781 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
782 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
783 /* Please set reply bit if this packet OK */
786 /* Once we've had two way comms, always ESTABLISHED. */
787 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
788 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
789 *ctinfo
= IP_CT_ESTABLISHED
;
790 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
791 pr_debug("nf_conntrack_in: related packet for %p\n",
793 *ctinfo
= IP_CT_RELATED
;
795 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
800 skb
->nfct
= &ct
->ct_general
;
801 skb
->nfctinfo
= *ctinfo
;
806 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
809 struct nf_conn
*ct
, *tmpl
= NULL
;
810 enum ip_conntrack_info ctinfo
;
811 struct nf_conntrack_l3proto
*l3proto
;
812 struct nf_conntrack_l4proto
*l4proto
;
813 unsigned int dataoff
;
819 /* Previously seen (loopback or untracked)? Ignore. */
820 tmpl
= (struct nf_conn
*)skb
->nfct
;
821 if (!nf_ct_is_template(tmpl
)) {
822 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
828 /* rcu_read_lock()ed by nf_hook_slow */
829 l3proto
= __nf_ct_l3proto_find(pf
);
830 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
831 &dataoff
, &protonum
);
833 pr_debug("not prepared to track yet or error occured\n");
834 NF_CT_STAT_INC_ATOMIC(net
, error
);
835 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
840 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
842 /* It may be an special packet, error, unclean...
843 * inverse of the return code tells to the netfilter
844 * core what to do with the packet. */
845 if (l4proto
->error
!= NULL
) {
846 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
849 NF_CT_STAT_INC_ATOMIC(net
, error
);
850 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
856 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
857 l3proto
, l4proto
, &set_reply
, &ctinfo
);
859 /* Not valid part of a connection */
860 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
866 /* Too stressed to deal. */
867 NF_CT_STAT_INC_ATOMIC(net
, drop
);
872 NF_CT_ASSERT(skb
->nfct
);
874 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
876 /* Invalid: inverse of the return code tells
877 * the netfilter core what to do */
878 pr_debug("nf_conntrack_in: Can't track with proto module\n");
879 nf_conntrack_put(skb
->nfct
);
881 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
883 NF_CT_STAT_INC_ATOMIC(net
, drop
);
888 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
889 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
896 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
898 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
899 const struct nf_conntrack_tuple
*orig
)
904 ret
= nf_ct_invert_tuple(inverse
, orig
,
905 __nf_ct_l3proto_find(orig
->src
.l3num
),
906 __nf_ct_l4proto_find(orig
->src
.l3num
,
907 orig
->dst
.protonum
));
911 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
913 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
914 implicitly racy: see __nf_conntrack_confirm */
915 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
916 const struct nf_conntrack_tuple
*newreply
)
918 struct nf_conn_help
*help
= nfct_help(ct
);
920 /* Should be unconfirmed, so not in hash table yet */
921 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
923 pr_debug("Altering reply tuple of %p to ", ct
);
924 nf_ct_dump_tuple(newreply
);
926 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
927 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
931 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
934 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
936 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
937 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
938 enum ip_conntrack_info ctinfo
,
939 const struct sk_buff
*skb
,
940 unsigned long extra_jiffies
,
943 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
946 /* Only update if this is not a fixed timeout */
947 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
950 /* If not in hash table, timer will not be active yet */
951 if (!nf_ct_is_confirmed(ct
)) {
952 ct
->timeout
.expires
= extra_jiffies
;
954 unsigned long newtime
= jiffies
+ extra_jiffies
;
956 /* Only update the timeout if the new timeout is at least
957 HZ jiffies from the old timeout. Need del_timer for race
958 avoidance (may already be dying). */
959 if (newtime
- ct
->timeout
.expires
>= HZ
)
960 mod_timer_pending(&ct
->timeout
, newtime
);
965 struct nf_conn_counter
*acct
;
967 acct
= nf_conn_acct_find(ct
);
969 spin_lock_bh(&ct
->lock
);
970 acct
[CTINFO2DIR(ctinfo
)].packets
++;
971 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
972 skb
->len
- skb_network_offset(skb
);
973 spin_unlock_bh(&ct
->lock
);
977 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
979 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
980 enum ip_conntrack_info ctinfo
,
981 const struct sk_buff
*skb
,
985 struct nf_conn_counter
*acct
;
987 acct
= nf_conn_acct_find(ct
);
989 spin_lock_bh(&ct
->lock
);
990 acct
[CTINFO2DIR(ctinfo
)].packets
++;
991 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
992 skb
->len
- skb_network_offset(skb
);
993 spin_unlock_bh(&ct
->lock
);
997 if (del_timer(&ct
->timeout
)) {
998 ct
->timeout
.function((unsigned long)ct
);
1003 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1005 #ifdef CONFIG_NF_CONNTRACK_ZONES
1006 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1007 .len
= sizeof(struct nf_conntrack_zone
),
1008 .align
= __alignof__(struct nf_conntrack_zone
),
1009 .id
= NF_CT_EXT_ZONE
,
1013 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1015 #include <linux/netfilter/nfnetlink.h>
1016 #include <linux/netfilter/nfnetlink_conntrack.h>
1017 #include <linux/mutex.h>
1019 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1020 * in ip_conntrack_core, since we don't want the protocols to autoload
1021 * or depend on ctnetlink */
1022 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1023 const struct nf_conntrack_tuple
*tuple
)
1025 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
1026 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
1032 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1034 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1035 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1036 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1038 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1040 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1041 struct nf_conntrack_tuple
*t
)
1043 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1046 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1047 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1051 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1053 int nf_ct_port_nlattr_tuple_size(void)
1055 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1057 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1060 /* Used by ipt_REJECT and ip6t_REJECT. */
1061 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
1064 enum ip_conntrack_info ctinfo
;
1066 /* This ICMP is in reverse direction to the packet which caused it */
1067 ct
= nf_ct_get(skb
, &ctinfo
);
1068 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1069 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
1071 ctinfo
= IP_CT_RELATED
;
1073 /* Attach to new skbuff, and increment count */
1074 nskb
->nfct
= &ct
->ct_general
;
1075 nskb
->nfctinfo
= ctinfo
;
1076 nf_conntrack_get(nskb
->nfct
);
1079 /* Bring out ya dead! */
1080 static struct nf_conn
*
1081 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1082 void *data
, unsigned int *bucket
)
1084 struct nf_conntrack_tuple_hash
*h
;
1086 struct hlist_nulls_node
*n
;
1088 spin_lock_bh(&nf_conntrack_lock
);
1089 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1090 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1091 ct
= nf_ct_tuplehash_to_ctrack(h
);
1096 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnnode
) {
1097 ct
= nf_ct_tuplehash_to_ctrack(h
);
1099 set_bit(IPS_DYING_BIT
, &ct
->status
);
1101 spin_unlock_bh(&nf_conntrack_lock
);
1104 atomic_inc(&ct
->ct_general
.use
);
1105 spin_unlock_bh(&nf_conntrack_lock
);
1109 void nf_ct_iterate_cleanup(struct net
*net
,
1110 int (*iter
)(struct nf_conn
*i
, void *data
),
1114 unsigned int bucket
= 0;
1116 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1117 /* Time to push up daises... */
1118 if (del_timer(&ct
->timeout
))
1119 death_by_timeout((unsigned long)ct
);
1120 /* ... else the timer will get him soon. */
1125 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1127 struct __nf_ct_flush_report
{
1132 static int kill_report(struct nf_conn
*i
, void *data
)
1134 struct __nf_ct_flush_report
*fr
= (struct __nf_ct_flush_report
*)data
;
1136 /* If we fail to deliver the event, death_by_timeout() will retry */
1137 if (nf_conntrack_event_report(IPCT_DESTROY
, i
,
1138 fr
->pid
, fr
->report
) < 0)
1141 /* Avoid the delivery of the destroy event in death_by_timeout(). */
1142 set_bit(IPS_DYING_BIT
, &i
->status
);
1146 static int kill_all(struct nf_conn
*i
, void *data
)
1151 void nf_ct_free_hashtable(void *hash
, int vmalloced
, unsigned int size
)
1156 free_pages((unsigned long)hash
,
1157 get_order(sizeof(struct hlist_head
) * size
));
1159 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1161 void nf_conntrack_flush_report(struct net
*net
, u32 pid
, int report
)
1163 struct __nf_ct_flush_report fr
= {
1167 nf_ct_iterate_cleanup(net
, kill_report
, &fr
);
1169 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report
);
1171 static void nf_ct_release_dying_list(struct net
*net
)
1173 struct nf_conntrack_tuple_hash
*h
;
1175 struct hlist_nulls_node
*n
;
1177 spin_lock_bh(&nf_conntrack_lock
);
1178 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.dying
, hnnode
) {
1179 ct
= nf_ct_tuplehash_to_ctrack(h
);
1180 /* never fails to remove them, no listeners at this point */
1183 spin_unlock_bh(&nf_conntrack_lock
);
1186 static void nf_conntrack_cleanup_init_net(void)
1188 /* wait until all references to nf_conntrack_untracked are dropped */
1189 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1192 nf_conntrack_helper_fini();
1193 nf_conntrack_proto_fini();
1194 #ifdef CONFIG_NF_CONNTRACK_ZONES
1195 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1199 static void nf_conntrack_cleanup_net(struct net
*net
)
1202 nf_ct_iterate_cleanup(net
, kill_all
, NULL
);
1203 nf_ct_release_dying_list(net
);
1204 if (atomic_read(&net
->ct
.count
) != 0) {
1206 goto i_see_dead_people
;
1209 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1210 net
->ct
.htable_size
);
1211 nf_conntrack_ecache_fini(net
);
1212 nf_conntrack_acct_fini(net
);
1213 nf_conntrack_expect_fini(net
);
1214 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1215 kfree(net
->ct
.slabname
);
1216 free_percpu(net
->ct
.stat
);
1219 /* Mishearing the voices in his head, our hero wonders how he's
1220 supposed to kill the mall. */
1221 void nf_conntrack_cleanup(struct net
*net
)
1223 if (net_eq(net
, &init_net
))
1224 rcu_assign_pointer(ip_ct_attach
, NULL
);
1226 /* This makes sure all current packets have passed through
1227 netfilter framework. Roll on, two-stage module
1231 nf_conntrack_cleanup_net(net
);
1233 if (net_eq(net
, &init_net
)) {
1234 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1235 nf_conntrack_cleanup_init_net();
1239 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
, int nulls
)
1241 struct hlist_nulls_head
*hash
;
1242 unsigned int nr_slots
, i
;
1247 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1248 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1249 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1250 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1254 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1255 hash
= __vmalloc(sz
, GFP_KERNEL
| __GFP_ZERO
, PAGE_KERNEL
);
1259 for (i
= 0; i
< nr_slots
; i
++)
1260 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1264 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1266 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1268 int i
, bucket
, vmalloced
, old_vmalloced
;
1269 unsigned int hashsize
, old_size
;
1270 struct hlist_nulls_head
*hash
, *old_hash
;
1271 struct nf_conntrack_tuple_hash
*h
;
1274 if (current
->nsproxy
->net_ns
!= &init_net
)
1277 /* On boot, we can set this without any fancy locking. */
1278 if (!nf_conntrack_htable_size
)
1279 return param_set_uint(val
, kp
);
1281 hashsize
= simple_strtoul(val
, NULL
, 0);
1285 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
, 1);
1289 /* Lookups in the old hash might happen in parallel, which means we
1290 * might get false negatives during connection lookup. New connections
1291 * created because of a false negative won't make it into the hash
1292 * though since that required taking the lock.
1294 spin_lock_bh(&nf_conntrack_lock
);
1295 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1296 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1297 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1298 struct nf_conntrack_tuple_hash
, hnnode
);
1299 ct
= nf_ct_tuplehash_to_ctrack(h
);
1300 hlist_nulls_del_rcu(&h
->hnnode
);
1301 bucket
= __hash_conntrack(&h
->tuple
, nf_ct_zone(ct
),
1303 nf_conntrack_hash_rnd
);
1304 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1307 old_size
= init_net
.ct
.htable_size
;
1308 old_vmalloced
= init_net
.ct
.hash_vmalloc
;
1309 old_hash
= init_net
.ct
.hash
;
1311 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1312 init_net
.ct
.hash_vmalloc
= vmalloced
;
1313 init_net
.ct
.hash
= hash
;
1314 spin_unlock_bh(&nf_conntrack_lock
);
1316 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1319 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1321 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1322 &nf_conntrack_htable_size
, 0600);
1324 static int nf_conntrack_init_init_net(void)
1329 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1330 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1331 if (!nf_conntrack_htable_size
) {
1332 nf_conntrack_htable_size
1333 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1334 / sizeof(struct hlist_head
));
1335 if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1336 nf_conntrack_htable_size
= 16384;
1337 if (nf_conntrack_htable_size
< 32)
1338 nf_conntrack_htable_size
= 32;
1340 /* Use a max. factor of four by default to get the same max as
1341 * with the old struct list_heads. When a table size is given
1342 * we use the old value of 8 to avoid reducing the max.
1346 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1348 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1349 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1352 ret
= nf_conntrack_proto_init();
1356 ret
= nf_conntrack_helper_init();
1360 #ifdef CONFIG_NF_CONNTRACK_ZONES
1361 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1365 /* Set up fake conntrack: to never be deleted, not in any hashes */
1366 #ifdef CONFIG_NET_NS
1367 nf_conntrack_untracked
.ct_net
= &init_net
;
1369 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1370 /* - and look it like as a confirmed connection */
1371 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1375 #ifdef CONFIG_NF_CONNTRACK_ZONES
1377 nf_conntrack_helper_fini();
1380 nf_conntrack_proto_fini();
1386 * We need to use special "null" values, not used in hash table
1388 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1389 #define DYING_NULLS_VAL ((1<<30)+1)
1391 static int nf_conntrack_init_net(struct net
*net
)
1395 atomic_set(&net
->ct
.count
, 0);
1396 INIT_HLIST_NULLS_HEAD(&net
->ct
.unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1397 INIT_HLIST_NULLS_HEAD(&net
->ct
.dying
, DYING_NULLS_VAL
);
1398 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1399 if (!net
->ct
.stat
) {
1404 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1405 if (!net
->ct
.slabname
) {
1410 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1411 sizeof(struct nf_conn
), 0,
1412 SLAB_DESTROY_BY_RCU
, NULL
);
1413 if (!net
->ct
.nf_conntrack_cachep
) {
1414 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1419 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1420 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
,
1421 &net
->ct
.hash_vmalloc
, 1);
1422 if (!net
->ct
.hash
) {
1424 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1427 ret
= nf_conntrack_expect_init(net
);
1430 ret
= nf_conntrack_acct_init(net
);
1433 ret
= nf_conntrack_ecache_init(net
);
1440 nf_conntrack_acct_fini(net
);
1442 nf_conntrack_expect_fini(net
);
1444 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1445 net
->ct
.htable_size
);
1447 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1449 kfree(net
->ct
.slabname
);
1451 free_percpu(net
->ct
.stat
);
1456 s16 (*nf_ct_nat_offset
)(const struct nf_conn
*ct
,
1457 enum ip_conntrack_dir dir
,
1459 EXPORT_SYMBOL_GPL(nf_ct_nat_offset
);
1461 int nf_conntrack_init(struct net
*net
)
1465 if (net_eq(net
, &init_net
)) {
1466 ret
= nf_conntrack_init_init_net();
1470 ret
= nf_conntrack_init_net(net
);
1474 if (net_eq(net
, &init_net
)) {
1475 /* For use by REJECT target */
1476 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1477 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1479 /* Howto get NAT offsets */
1480 rcu_assign_pointer(nf_ct_nat_offset
, NULL
);
1485 if (net_eq(net
, &init_net
))
1486 nf_conntrack_cleanup_init_net();