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>
33 #include <net/netfilter/nf_conntrack.h>
34 #include <net/netfilter/nf_conntrack_l3proto.h>
35 #include <net/netfilter/nf_conntrack_l4proto.h>
36 #include <net/netfilter/nf_conntrack_expect.h>
37 #include <net/netfilter/nf_conntrack_helper.h>
38 #include <net/netfilter/nf_conntrack_core.h>
39 #include <net/netfilter/nf_conntrack_extend.h>
40 #include <net/netfilter/nf_conntrack_acct.h>
42 #define NF_CONNTRACK_VERSION "0.5.0"
44 DEFINE_SPINLOCK(nf_conntrack_lock
);
45 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
47 unsigned int nf_conntrack_htable_size __read_mostly
;
48 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
50 int nf_conntrack_max __read_mostly
;
51 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
53 struct nf_conn nf_conntrack_untracked __read_mostly
;
54 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
56 static struct kmem_cache
*nf_conntrack_cachep __read_mostly
;
58 static int nf_conntrack_hash_rnd_initted
;
59 static unsigned int nf_conntrack_hash_rnd
;
61 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
62 unsigned int size
, unsigned int rnd
)
67 /* The direction must be ignored, so we hash everything up to the
68 * destination ports (which is a multiple of 4) and treat the last
69 * three bytes manually.
71 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
72 h
= jhash2((u32
*)tuple
, n
,
73 rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
74 tuple
->dst
.protonum
));
76 return ((u64
)h
* size
) >> 32;
79 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
81 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
82 nf_conntrack_hash_rnd
);
86 nf_ct_get_tuple(const struct sk_buff
*skb
,
91 struct nf_conntrack_tuple
*tuple
,
92 const struct nf_conntrack_l3proto
*l3proto
,
93 const struct nf_conntrack_l4proto
*l4proto
)
95 memset(tuple
, 0, sizeof(*tuple
));
97 tuple
->src
.l3num
= l3num
;
98 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
101 tuple
->dst
.protonum
= protonum
;
102 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
104 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
106 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
108 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
109 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
111 struct nf_conntrack_l3proto
*l3proto
;
112 struct nf_conntrack_l4proto
*l4proto
;
113 unsigned int protoff
;
119 l3proto
= __nf_ct_l3proto_find(l3num
);
120 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
121 if (ret
!= NF_ACCEPT
) {
126 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
128 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
134 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
137 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
138 const struct nf_conntrack_tuple
*orig
,
139 const struct nf_conntrack_l3proto
*l3proto
,
140 const struct nf_conntrack_l4proto
*l4proto
)
142 memset(inverse
, 0, sizeof(*inverse
));
144 inverse
->src
.l3num
= orig
->src
.l3num
;
145 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
148 inverse
->dst
.dir
= !orig
->dst
.dir
;
150 inverse
->dst
.protonum
= orig
->dst
.protonum
;
151 return l4proto
->invert_tuple(inverse
, orig
);
153 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
156 clean_from_lists(struct nf_conn
*ct
)
158 pr_debug("clean_from_lists(%p)\n", ct
);
159 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
160 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
);
162 /* Destroy all pending expectations */
163 nf_ct_remove_expectations(ct
);
167 destroy_conntrack(struct nf_conntrack
*nfct
)
169 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
170 struct net
*net
= nf_ct_net(ct
);
171 struct nf_conntrack_l4proto
*l4proto
;
173 pr_debug("destroy_conntrack(%p)\n", ct
);
174 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
175 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
177 nf_conntrack_event(IPCT_DESTROY
, ct
);
178 set_bit(IPS_DYING_BIT
, &ct
->status
);
180 /* To make sure we don't get any weird locking issues here:
181 * destroy_conntrack() MUST NOT be called with a write lock
182 * to nf_conntrack_lock!!! -HW */
184 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
185 if (l4proto
&& l4proto
->destroy
)
186 l4proto
->destroy(ct
);
190 spin_lock_bh(&nf_conntrack_lock
);
191 /* Expectations will have been removed in clean_from_lists,
192 * except TFTP can create an expectation on the first packet,
193 * before connection is in the list, so we need to clean here,
195 nf_ct_remove_expectations(ct
);
197 /* We overload first tuple to link into unconfirmed list. */
198 if (!nf_ct_is_confirmed(ct
)) {
199 BUG_ON(hlist_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
));
200 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
203 NF_CT_STAT_INC(net
, delete);
204 spin_unlock_bh(&nf_conntrack_lock
);
207 nf_ct_put(ct
->master
);
209 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
210 nf_conntrack_free(ct
);
213 static void death_by_timeout(unsigned long ul_conntrack
)
215 struct nf_conn
*ct
= (void *)ul_conntrack
;
216 struct net
*net
= nf_ct_net(ct
);
217 struct nf_conn_help
*help
= nfct_help(ct
);
218 struct nf_conntrack_helper
*helper
;
222 helper
= rcu_dereference(help
->helper
);
223 if (helper
&& helper
->destroy
)
228 spin_lock_bh(&nf_conntrack_lock
);
229 /* Inside lock so preempt is disabled on module removal path.
230 * Otherwise we can get spurious warnings. */
231 NF_CT_STAT_INC(net
, delete_list
);
232 clean_from_lists(ct
);
233 spin_unlock_bh(&nf_conntrack_lock
);
237 struct nf_conntrack_tuple_hash
*
238 __nf_conntrack_find(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
240 struct nf_conntrack_tuple_hash
*h
;
241 struct hlist_node
*n
;
242 unsigned int hash
= hash_conntrack(tuple
);
244 /* Disable BHs the entire time since we normally need to disable them
245 * at least once for the stats anyway.
248 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnode
) {
249 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
250 NF_CT_STAT_INC(net
, found
);
254 NF_CT_STAT_INC(net
, searched
);
260 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
262 /* Find a connection corresponding to a tuple. */
263 struct nf_conntrack_tuple_hash
*
264 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
266 struct nf_conntrack_tuple_hash
*h
;
270 h
= __nf_conntrack_find(net
, tuple
);
272 ct
= nf_ct_tuplehash_to_ctrack(h
);
273 if (unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
280 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
282 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
284 unsigned int repl_hash
)
286 struct net
*net
= nf_ct_net(ct
);
288 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
289 &net
->ct
.hash
[hash
]);
290 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
,
291 &net
->ct
.hash
[repl_hash
]);
294 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
296 unsigned int hash
, repl_hash
;
298 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
299 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
301 spin_lock_bh(&nf_conntrack_lock
);
302 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
303 spin_unlock_bh(&nf_conntrack_lock
);
305 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
307 /* Confirm a connection given skb; places it in hash table */
309 __nf_conntrack_confirm(struct sk_buff
*skb
)
311 unsigned int hash
, repl_hash
;
312 struct nf_conntrack_tuple_hash
*h
;
314 struct nf_conn_help
*help
;
315 struct hlist_node
*n
;
316 enum ip_conntrack_info ctinfo
;
319 ct
= nf_ct_get(skb
, &ctinfo
);
322 /* ipt_REJECT uses nf_conntrack_attach to attach related
323 ICMP/TCP RST packets in other direction. Actual packet
324 which created connection will be IP_CT_NEW or for an
325 expected connection, IP_CT_RELATED. */
326 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
329 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
330 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
332 /* We're not in hash table, and we refuse to set up related
333 connections for unconfirmed conns. But packet copies and
334 REJECT will give spurious warnings here. */
335 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
337 /* No external references means noone else could have
339 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
340 pr_debug("Confirming conntrack %p\n", ct
);
342 spin_lock_bh(&nf_conntrack_lock
);
344 /* See if there's one in the list already, including reverse:
345 NAT could have grabbed it without realizing, since we're
346 not in the hash. If there is, we lost race. */
347 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnode
)
348 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
351 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnode
)
352 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
356 /* Remove from unconfirmed list */
357 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
359 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
360 /* Timer relative to confirmation time, not original
361 setting time, otherwise we'd get timer wrap in
362 weird delay cases. */
363 ct
->timeout
.expires
+= jiffies
;
364 add_timer(&ct
->timeout
);
365 atomic_inc(&ct
->ct_general
.use
);
366 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
367 NF_CT_STAT_INC(net
, insert
);
368 spin_unlock_bh(&nf_conntrack_lock
);
369 help
= nfct_help(ct
);
370 if (help
&& help
->helper
)
371 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
372 #ifdef CONFIG_NF_NAT_NEEDED
373 if (test_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
) ||
374 test_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
))
375 nf_conntrack_event_cache(IPCT_NATINFO
, ct
);
377 nf_conntrack_event_cache(master_ct(ct
) ?
378 IPCT_RELATED
: IPCT_NEW
, ct
);
382 NF_CT_STAT_INC(net
, insert_failed
);
383 spin_unlock_bh(&nf_conntrack_lock
);
386 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
388 /* Returns true if a connection correspondings to the tuple (required
391 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
392 const struct nf_conn
*ignored_conntrack
)
394 struct net
*net
= nf_ct_net(ignored_conntrack
);
395 struct nf_conntrack_tuple_hash
*h
;
396 struct hlist_node
*n
;
397 unsigned int hash
= hash_conntrack(tuple
);
399 /* Disable BHs the entire time since we need to disable them at
400 * least once for the stats anyway.
403 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnode
) {
404 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
405 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
406 NF_CT_STAT_INC(net
, found
);
407 rcu_read_unlock_bh();
410 NF_CT_STAT_INC(net
, searched
);
412 rcu_read_unlock_bh();
416 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
418 #define NF_CT_EVICTION_RANGE 8
420 /* There's a small race here where we may free a just-assured
421 connection. Too bad: we're in trouble anyway. */
422 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
424 /* Use oldest entry, which is roughly LRU */
425 struct nf_conntrack_tuple_hash
*h
;
426 struct nf_conn
*ct
= NULL
, *tmp
;
427 struct hlist_node
*n
;
428 unsigned int i
, cnt
= 0;
432 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
433 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
435 tmp
= nf_ct_tuplehash_to_ctrack(h
);
436 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
441 if (ct
&& unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
443 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
445 hash
= (hash
+ 1) % nf_conntrack_htable_size
;
452 if (del_timer(&ct
->timeout
)) {
453 death_by_timeout((unsigned long)ct
);
455 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
461 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
462 const struct nf_conntrack_tuple
*orig
,
463 const struct nf_conntrack_tuple
*repl
,
466 struct nf_conn
*ct
= NULL
;
468 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
469 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
470 nf_conntrack_hash_rnd_initted
= 1;
473 /* We don't want any race condition at early drop stage */
474 atomic_inc(&net
->ct
.count
);
476 if (nf_conntrack_max
&&
477 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
478 unsigned int hash
= hash_conntrack(orig
);
479 if (!early_drop(net
, hash
)) {
480 atomic_dec(&net
->ct
.count
);
483 "nf_conntrack: table full, dropping"
485 return ERR_PTR(-ENOMEM
);
489 ct
= kmem_cache_zalloc(nf_conntrack_cachep
, gfp
);
491 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
492 atomic_dec(&net
->ct
.count
);
493 return ERR_PTR(-ENOMEM
);
496 atomic_set(&ct
->ct_general
.use
, 1);
497 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
498 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
499 /* Don't set timer yet: wait for confirmation */
500 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
504 INIT_RCU_HEAD(&ct
->rcu
);
508 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
510 static void nf_conntrack_free_rcu(struct rcu_head
*head
)
512 struct nf_conn
*ct
= container_of(head
, struct nf_conn
, rcu
);
513 struct net
*net
= nf_ct_net(ct
);
516 kmem_cache_free(nf_conntrack_cachep
, ct
);
517 atomic_dec(&net
->ct
.count
);
520 void nf_conntrack_free(struct nf_conn
*ct
)
522 nf_ct_ext_destroy(ct
);
523 call_rcu(&ct
->rcu
, nf_conntrack_free_rcu
);
525 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
527 /* Allocate a new conntrack: we return -ENOMEM if classification
528 failed due to stress. Otherwise it really is unclassifiable. */
529 static struct nf_conntrack_tuple_hash
*
530 init_conntrack(struct net
*net
,
531 const struct nf_conntrack_tuple
*tuple
,
532 struct nf_conntrack_l3proto
*l3proto
,
533 struct nf_conntrack_l4proto
*l4proto
,
535 unsigned int dataoff
)
538 struct nf_conn_help
*help
;
539 struct nf_conntrack_tuple repl_tuple
;
540 struct nf_conntrack_expect
*exp
;
542 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
543 pr_debug("Can't invert tuple.\n");
547 ct
= nf_conntrack_alloc(net
, tuple
, &repl_tuple
, GFP_ATOMIC
);
548 if (ct
== NULL
|| IS_ERR(ct
)) {
549 pr_debug("Can't allocate conntrack.\n");
550 return (struct nf_conntrack_tuple_hash
*)ct
;
553 if (!l4proto
->new(ct
, skb
, dataoff
)) {
554 nf_conntrack_free(ct
);
555 pr_debug("init conntrack: can't track with proto module\n");
559 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
561 spin_lock_bh(&nf_conntrack_lock
);
562 exp
= nf_ct_find_expectation(net
, tuple
);
564 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
566 /* Welcome, Mr. Bond. We've been expecting you... */
567 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
568 ct
->master
= exp
->master
;
570 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
572 rcu_assign_pointer(help
->helper
, exp
->helper
);
575 #ifdef CONFIG_NF_CONNTRACK_MARK
576 ct
->mark
= exp
->master
->mark
;
578 #ifdef CONFIG_NF_CONNTRACK_SECMARK
579 ct
->secmark
= exp
->master
->secmark
;
581 nf_conntrack_get(&ct
->master
->ct_general
);
582 NF_CT_STAT_INC(net
, expect_new
);
584 struct nf_conntrack_helper
*helper
;
586 helper
= __nf_ct_helper_find(&repl_tuple
);
588 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
590 rcu_assign_pointer(help
->helper
, helper
);
592 NF_CT_STAT_INC(net
, new);
595 /* Overload tuple linked list to put us in unconfirmed list. */
596 hlist_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
597 &net
->ct
.unconfirmed
);
599 spin_unlock_bh(&nf_conntrack_lock
);
603 exp
->expectfn(ct
, exp
);
604 nf_ct_expect_put(exp
);
607 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
610 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
611 static inline struct nf_conn
*
612 resolve_normal_ct(struct net
*net
,
614 unsigned int dataoff
,
617 struct nf_conntrack_l3proto
*l3proto
,
618 struct nf_conntrack_l4proto
*l4proto
,
620 enum ip_conntrack_info
*ctinfo
)
622 struct nf_conntrack_tuple tuple
;
623 struct nf_conntrack_tuple_hash
*h
;
626 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
627 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
629 pr_debug("resolve_normal_ct: Can't get tuple\n");
633 /* look for tuple match */
634 h
= nf_conntrack_find_get(net
, &tuple
);
636 h
= init_conntrack(net
, &tuple
, l3proto
, l4proto
, skb
, dataoff
);
642 ct
= nf_ct_tuplehash_to_ctrack(h
);
644 /* It exists; we have (non-exclusive) reference. */
645 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
646 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
647 /* Please set reply bit if this packet OK */
650 /* Once we've had two way comms, always ESTABLISHED. */
651 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
652 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
653 *ctinfo
= IP_CT_ESTABLISHED
;
654 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
655 pr_debug("nf_conntrack_in: related packet for %p\n",
657 *ctinfo
= IP_CT_RELATED
;
659 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
664 skb
->nfct
= &ct
->ct_general
;
665 skb
->nfctinfo
= *ctinfo
;
670 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
674 enum ip_conntrack_info ctinfo
;
675 struct nf_conntrack_l3proto
*l3proto
;
676 struct nf_conntrack_l4proto
*l4proto
;
677 unsigned int dataoff
;
682 /* Previously seen (loopback or untracked)? Ignore. */
684 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
688 /* rcu_read_lock()ed by nf_hook_slow */
689 l3proto
= __nf_ct_l3proto_find(pf
);
690 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
691 &dataoff
, &protonum
);
693 pr_debug("not prepared to track yet or error occured\n");
694 NF_CT_STAT_INC_ATOMIC(net
, error
);
695 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
699 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
701 /* It may be an special packet, error, unclean...
702 * inverse of the return code tells to the netfilter
703 * core what to do with the packet. */
704 if (l4proto
->error
!= NULL
) {
705 ret
= l4proto
->error(net
, skb
, dataoff
, &ctinfo
, pf
, hooknum
);
707 NF_CT_STAT_INC_ATOMIC(net
, error
);
708 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
713 ct
= resolve_normal_ct(net
, skb
, dataoff
, pf
, protonum
,
714 l3proto
, l4proto
, &set_reply
, &ctinfo
);
716 /* Not valid part of a connection */
717 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
722 /* Too stressed to deal. */
723 NF_CT_STAT_INC_ATOMIC(net
, drop
);
727 NF_CT_ASSERT(skb
->nfct
);
729 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
731 /* Invalid: inverse of the return code tells
732 * the netfilter core what to do */
733 pr_debug("nf_conntrack_in: Can't track with proto module\n");
734 nf_conntrack_put(skb
->nfct
);
736 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
740 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
741 nf_conntrack_event_cache(IPCT_STATUS
, ct
);
745 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
747 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
748 const struct nf_conntrack_tuple
*orig
)
753 ret
= nf_ct_invert_tuple(inverse
, orig
,
754 __nf_ct_l3proto_find(orig
->src
.l3num
),
755 __nf_ct_l4proto_find(orig
->src
.l3num
,
756 orig
->dst
.protonum
));
760 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
762 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
763 implicitly racy: see __nf_conntrack_confirm */
764 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
765 const struct nf_conntrack_tuple
*newreply
)
767 struct nf_conn_help
*help
= nfct_help(ct
);
768 struct nf_conntrack_helper
*helper
;
770 /* Should be unconfirmed, so not in hash table yet */
771 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
773 pr_debug("Altering reply tuple of %p to ", ct
);
774 nf_ct_dump_tuple(newreply
);
776 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
777 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
781 helper
= __nf_ct_helper_find(newreply
);
782 if (helper
== NULL
) {
784 rcu_assign_pointer(help
->helper
, NULL
);
789 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
793 memset(&help
->help
, 0, sizeof(help
->help
));
796 rcu_assign_pointer(help
->helper
, helper
);
800 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
802 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
803 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
804 enum ip_conntrack_info ctinfo
,
805 const struct sk_buff
*skb
,
806 unsigned long extra_jiffies
,
811 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
814 spin_lock_bh(&nf_conntrack_lock
);
816 /* Only update if this is not a fixed timeout */
817 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
820 /* If not in hash table, timer will not be active yet */
821 if (!nf_ct_is_confirmed(ct
)) {
822 ct
->timeout
.expires
= extra_jiffies
;
823 event
= IPCT_REFRESH
;
825 unsigned long newtime
= jiffies
+ extra_jiffies
;
827 /* Only update the timeout if the new timeout is at least
828 HZ jiffies from the old timeout. Need del_timer for race
829 avoidance (may already be dying). */
830 if (newtime
- ct
->timeout
.expires
>= HZ
831 && del_timer(&ct
->timeout
)) {
832 ct
->timeout
.expires
= newtime
;
833 add_timer(&ct
->timeout
);
834 event
= IPCT_REFRESH
;
840 struct nf_conn_counter
*acct
;
842 acct
= nf_conn_acct_find(ct
);
844 acct
[CTINFO2DIR(ctinfo
)].packets
++;
845 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
846 skb
->len
- skb_network_offset(skb
);
850 spin_unlock_bh(&nf_conntrack_lock
);
852 /* must be unlocked when calling event cache */
854 nf_conntrack_event_cache(event
, ct
);
856 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
858 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
859 enum ip_conntrack_info ctinfo
,
860 const struct sk_buff
*skb
,
864 struct nf_conn_counter
*acct
;
866 spin_lock_bh(&nf_conntrack_lock
);
867 acct
= nf_conn_acct_find(ct
);
869 acct
[CTINFO2DIR(ctinfo
)].packets
++;
870 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
871 skb
->len
- skb_network_offset(skb
);
873 spin_unlock_bh(&nf_conntrack_lock
);
876 if (del_timer(&ct
->timeout
)) {
877 ct
->timeout
.function((unsigned long)ct
);
882 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
884 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
886 #include <linux/netfilter/nfnetlink.h>
887 #include <linux/netfilter/nfnetlink_conntrack.h>
888 #include <linux/mutex.h>
890 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
891 * in ip_conntrack_core, since we don't want the protocols to autoload
892 * or depend on ctnetlink */
893 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
894 const struct nf_conntrack_tuple
*tuple
)
896 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
897 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
903 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
905 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
906 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
907 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
909 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
911 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
912 struct nf_conntrack_tuple
*t
)
914 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
917 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
918 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
922 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
925 /* Used by ipt_REJECT and ip6t_REJECT. */
926 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
929 enum ip_conntrack_info ctinfo
;
931 /* This ICMP is in reverse direction to the packet which caused it */
932 ct
= nf_ct_get(skb
, &ctinfo
);
933 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
934 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
936 ctinfo
= IP_CT_RELATED
;
938 /* Attach to new skbuff, and increment count */
939 nskb
->nfct
= &ct
->ct_general
;
940 nskb
->nfctinfo
= ctinfo
;
941 nf_conntrack_get(nskb
->nfct
);
944 /* Bring out ya dead! */
945 static struct nf_conn
*
946 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
947 void *data
, unsigned int *bucket
)
949 struct nf_conntrack_tuple_hash
*h
;
951 struct hlist_node
*n
;
953 spin_lock_bh(&nf_conntrack_lock
);
954 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
955 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnode
) {
956 ct
= nf_ct_tuplehash_to_ctrack(h
);
961 hlist_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnode
) {
962 ct
= nf_ct_tuplehash_to_ctrack(h
);
964 set_bit(IPS_DYING_BIT
, &ct
->status
);
966 spin_unlock_bh(&nf_conntrack_lock
);
969 atomic_inc(&ct
->ct_general
.use
);
970 spin_unlock_bh(&nf_conntrack_lock
);
974 void nf_ct_iterate_cleanup(struct net
*net
,
975 int (*iter
)(struct nf_conn
*i
, void *data
),
979 unsigned int bucket
= 0;
981 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
982 /* Time to push up daises... */
983 if (del_timer(&ct
->timeout
))
984 death_by_timeout((unsigned long)ct
);
985 /* ... else the timer will get him soon. */
990 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
992 static int kill_all(struct nf_conn
*i
, void *data
)
997 void nf_ct_free_hashtable(struct hlist_head
*hash
, int vmalloced
, unsigned int size
)
1002 free_pages((unsigned long)hash
,
1003 get_order(sizeof(struct hlist_head
) * size
));
1005 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1007 void nf_conntrack_flush(struct net
*net
)
1009 nf_ct_iterate_cleanup(net
, kill_all
, NULL
);
1011 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
1013 static void nf_conntrack_cleanup_init_net(void)
1015 nf_conntrack_helper_fini();
1016 nf_conntrack_proto_fini();
1017 kmem_cache_destroy(nf_conntrack_cachep
);
1020 static void nf_conntrack_cleanup_net(struct net
*net
)
1022 nf_ct_event_cache_flush(net
);
1023 nf_conntrack_ecache_fini(net
);
1025 nf_conntrack_flush(net
);
1026 if (atomic_read(&net
->ct
.count
) != 0) {
1028 goto i_see_dead_people
;
1030 /* wait until all references to nf_conntrack_untracked are dropped */
1031 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1034 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1035 nf_conntrack_htable_size
);
1036 nf_conntrack_acct_fini(net
);
1037 nf_conntrack_expect_fini(net
);
1038 free_percpu(net
->ct
.stat
);
1041 /* Mishearing the voices in his head, our hero wonders how he's
1042 supposed to kill the mall. */
1043 void nf_conntrack_cleanup(struct net
*net
)
1045 if (net_eq(net
, &init_net
))
1046 rcu_assign_pointer(ip_ct_attach
, NULL
);
1048 /* This makes sure all current packets have passed through
1049 netfilter framework. Roll on, two-stage module
1053 nf_conntrack_cleanup_net(net
);
1055 if (net_eq(net
, &init_net
)) {
1056 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1057 nf_conntrack_cleanup_init_net();
1061 struct hlist_head
*nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
)
1063 struct hlist_head
*hash
;
1064 unsigned int size
, i
;
1068 size
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_head
));
1069 hash
= (void*)__get_free_pages(GFP_KERNEL
|__GFP_NOWARN
,
1070 get_order(sizeof(struct hlist_head
)
1074 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1075 hash
= vmalloc(sizeof(struct hlist_head
) * size
);
1079 for (i
= 0; i
< size
; i
++)
1080 INIT_HLIST_HEAD(&hash
[i
]);
1084 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1086 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1088 int i
, bucket
, vmalloced
, old_vmalloced
;
1089 unsigned int hashsize
, old_size
;
1091 struct hlist_head
*hash
, *old_hash
;
1092 struct nf_conntrack_tuple_hash
*h
;
1094 /* On boot, we can set this without any fancy locking. */
1095 if (!nf_conntrack_htable_size
)
1096 return param_set_uint(val
, kp
);
1098 hashsize
= simple_strtoul(val
, NULL
, 0);
1102 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
);
1106 /* We have to rehahs for the new table anyway, so we also can
1107 * use a newrandom seed */
1108 get_random_bytes(&rnd
, 4);
1110 /* Lookups in the old hash might happen in parallel, which means we
1111 * might get false negatives during connection lookup. New connections
1112 * created because of a false negative won't make it into the hash
1113 * though since that required taking the lock.
1115 spin_lock_bh(&nf_conntrack_lock
);
1116 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1117 while (!hlist_empty(&init_net
.ct
.hash
[i
])) {
1118 h
= hlist_entry(init_net
.ct
.hash
[i
].first
,
1119 struct nf_conntrack_tuple_hash
, hnode
);
1120 hlist_del_rcu(&h
->hnode
);
1121 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1122 hlist_add_head(&h
->hnode
, &hash
[bucket
]);
1125 old_size
= nf_conntrack_htable_size
;
1126 old_vmalloced
= init_net
.ct
.hash_vmalloc
;
1127 old_hash
= init_net
.ct
.hash
;
1129 nf_conntrack_htable_size
= hashsize
;
1130 init_net
.ct
.hash_vmalloc
= vmalloced
;
1131 init_net
.ct
.hash
= hash
;
1132 nf_conntrack_hash_rnd
= rnd
;
1133 spin_unlock_bh(&nf_conntrack_lock
);
1135 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1138 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1140 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1141 &nf_conntrack_htable_size
, 0600);
1143 static int nf_conntrack_init_init_net(void)
1148 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1149 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1150 if (!nf_conntrack_htable_size
) {
1151 nf_conntrack_htable_size
1152 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1153 / sizeof(struct hlist_head
));
1154 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1155 nf_conntrack_htable_size
= 16384;
1156 if (nf_conntrack_htable_size
< 32)
1157 nf_conntrack_htable_size
= 32;
1159 /* Use a max. factor of four by default to get the same max as
1160 * with the old struct list_heads. When a table size is given
1161 * we use the old value of 8 to avoid reducing the max.
1165 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1167 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1168 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1171 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1172 sizeof(struct nf_conn
),
1174 if (!nf_conntrack_cachep
) {
1175 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1180 ret
= nf_conntrack_proto_init();
1184 ret
= nf_conntrack_helper_init();
1191 nf_conntrack_proto_fini();
1193 kmem_cache_destroy(nf_conntrack_cachep
);
1198 static int nf_conntrack_init_net(struct net
*net
)
1202 atomic_set(&net
->ct
.count
, 0);
1203 INIT_HLIST_HEAD(&net
->ct
.unconfirmed
);
1204 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1205 if (!net
->ct
.stat
) {
1209 ret
= nf_conntrack_ecache_init(net
);
1212 net
->ct
.hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
,
1213 &net
->ct
.hash_vmalloc
);
1214 if (!net
->ct
.hash
) {
1216 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1219 ret
= nf_conntrack_expect_init(net
);
1222 ret
= nf_conntrack_acct_init(net
);
1226 /* Set up fake conntrack:
1227 - to never be deleted, not in any hashes */
1228 #ifdef CONFIG_NET_NS
1229 nf_conntrack_untracked
.ct_net
= &init_net
;
1231 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1232 /* - and look it like as a confirmed connection */
1233 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1238 nf_conntrack_expect_fini(net
);
1240 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1241 nf_conntrack_htable_size
);
1243 nf_conntrack_ecache_fini(net
);
1245 free_percpu(net
->ct
.stat
);
1250 int nf_conntrack_init(struct net
*net
)
1254 if (net_eq(net
, &init_net
)) {
1255 ret
= nf_conntrack_init_init_net();
1259 ret
= nf_conntrack_init_net(net
);
1263 if (net_eq(net
, &init_net
)) {
1264 /* For use by REJECT target */
1265 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1266 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1271 if (net_eq(net
, &init_net
))
1272 nf_conntrack_cleanup_init_net();