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>
41 #include <net/netfilter/nf_nat.h>
43 #define NF_CONNTRACK_VERSION "0.5.0"
46 (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
47 enum nf_nat_manip_type manip
,
48 struct nlattr
*attr
) __read_mostly
;
49 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
51 DEFINE_SPINLOCK(nf_conntrack_lock
);
52 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
54 unsigned int nf_conntrack_htable_size __read_mostly
;
55 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
57 int nf_conntrack_max __read_mostly
;
58 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
60 struct nf_conn nf_conntrack_untracked __read_mostly
;
61 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
63 static struct kmem_cache
*nf_conntrack_cachep __read_mostly
;
65 static int nf_conntrack_hash_rnd_initted
;
66 static unsigned int nf_conntrack_hash_rnd
;
68 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
69 unsigned int size
, unsigned int rnd
)
74 /* The direction must be ignored, so we hash everything up to the
75 * destination ports (which is a multiple of 4) and treat the last
76 * three bytes manually.
78 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
79 h
= jhash2((u32
*)tuple
, n
,
80 rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
81 tuple
->dst
.protonum
));
83 return ((u64
)h
* size
) >> 32;
86 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
88 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
89 nf_conntrack_hash_rnd
);
93 nf_ct_get_tuple(const struct sk_buff
*skb
,
98 struct nf_conntrack_tuple
*tuple
,
99 const struct nf_conntrack_l3proto
*l3proto
,
100 const struct nf_conntrack_l4proto
*l4proto
)
102 memset(tuple
, 0, sizeof(*tuple
));
104 tuple
->src
.l3num
= l3num
;
105 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
108 tuple
->dst
.protonum
= protonum
;
109 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
111 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
113 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
115 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
116 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
118 struct nf_conntrack_l3proto
*l3proto
;
119 struct nf_conntrack_l4proto
*l4proto
;
120 unsigned int protoff
;
126 l3proto
= __nf_ct_l3proto_find(l3num
);
127 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
128 if (ret
!= NF_ACCEPT
) {
133 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
135 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
141 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
144 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
145 const struct nf_conntrack_tuple
*orig
,
146 const struct nf_conntrack_l3proto
*l3proto
,
147 const struct nf_conntrack_l4proto
*l4proto
)
149 memset(inverse
, 0, sizeof(*inverse
));
151 inverse
->src
.l3num
= orig
->src
.l3num
;
152 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
155 inverse
->dst
.dir
= !orig
->dst
.dir
;
157 inverse
->dst
.protonum
= orig
->dst
.protonum
;
158 return l4proto
->invert_tuple(inverse
, orig
);
160 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
163 clean_from_lists(struct nf_conn
*ct
)
165 pr_debug("clean_from_lists(%p)\n", ct
);
166 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
167 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
);
169 /* Destroy all pending expectations */
170 nf_ct_remove_expectations(ct
);
174 destroy_conntrack(struct nf_conntrack
*nfct
)
176 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
177 struct net
*net
= nf_ct_net(ct
);
178 struct nf_conntrack_l4proto
*l4proto
;
180 pr_debug("destroy_conntrack(%p)\n", ct
);
181 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
182 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
184 nf_conntrack_event(IPCT_DESTROY
, ct
);
185 set_bit(IPS_DYING_BIT
, &ct
->status
);
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_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
));
207 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
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 static void death_by_timeout(unsigned long ul_conntrack
)
222 struct nf_conn
*ct
= (void *)ul_conntrack
;
223 struct net
*net
= nf_ct_net(ct
);
224 struct nf_conn_help
*help
= nfct_help(ct
);
225 struct nf_conntrack_helper
*helper
;
229 helper
= rcu_dereference(help
->helper
);
230 if (helper
&& helper
->destroy
)
235 spin_lock_bh(&nf_conntrack_lock
);
236 /* Inside lock so preempt is disabled on module removal path.
237 * Otherwise we can get spurious warnings. */
238 NF_CT_STAT_INC(net
, delete_list
);
239 clean_from_lists(ct
);
240 spin_unlock_bh(&nf_conntrack_lock
);
244 struct nf_conntrack_tuple_hash
*
245 __nf_conntrack_find(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
247 struct nf_conntrack_tuple_hash
*h
;
248 struct hlist_node
*n
;
249 unsigned int hash
= hash_conntrack(tuple
);
251 /* Disable BHs the entire time since we normally need to disable them
252 * at least once for the stats anyway.
255 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnode
) {
256 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
257 NF_CT_STAT_INC(net
, found
);
261 NF_CT_STAT_INC(net
, searched
);
267 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
269 /* Find a connection corresponding to a tuple. */
270 struct nf_conntrack_tuple_hash
*
271 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
273 struct nf_conntrack_tuple_hash
*h
;
277 h
= __nf_conntrack_find(net
, tuple
);
279 ct
= nf_ct_tuplehash_to_ctrack(h
);
280 if (unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
287 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
289 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
291 unsigned int repl_hash
)
293 struct net
*net
= nf_ct_net(ct
);
295 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
296 &net
->ct
.hash
[hash
]);
297 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
,
298 &net
->ct
.hash
[repl_hash
]);
301 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
303 unsigned int hash
, repl_hash
;
305 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
306 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
308 spin_lock_bh(&nf_conntrack_lock
);
309 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
310 spin_unlock_bh(&nf_conntrack_lock
);
312 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
314 /* Confirm a connection given skb; places it in hash table */
316 __nf_conntrack_confirm(struct sk_buff
*skb
)
318 unsigned int hash
, repl_hash
;
319 struct nf_conntrack_tuple_hash
*h
;
321 struct nf_conn_help
*help
;
322 struct hlist_node
*n
;
323 enum ip_conntrack_info ctinfo
;
326 ct
= nf_ct_get(skb
, &ctinfo
);
329 /* ipt_REJECT uses nf_conntrack_attach to attach related
330 ICMP/TCP RST packets in other direction. Actual packet
331 which created connection will be IP_CT_NEW or for an
332 expected connection, IP_CT_RELATED. */
333 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
336 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
337 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
339 /* We're not in hash table, and we refuse to set up related
340 connections for unconfirmed conns. But packet copies and
341 REJECT will give spurious warnings here. */
342 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
344 /* No external references means noone else could have
346 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
347 pr_debug("Confirming conntrack %p\n", ct
);
349 spin_lock_bh(&nf_conntrack_lock
);
351 /* See if there's one in the list already, including reverse:
352 NAT could have grabbed it without realizing, since we're
353 not in the hash. If there is, we lost race. */
354 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnode
)
355 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
358 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnode
)
359 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
363 /* Remove from unconfirmed list */
364 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
366 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
367 /* Timer relative to confirmation time, not original
368 setting time, otherwise we'd get timer wrap in
369 weird delay cases. */
370 ct
->timeout
.expires
+= jiffies
;
371 add_timer(&ct
->timeout
);
372 atomic_inc(&ct
->ct_general
.use
);
373 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
374 NF_CT_STAT_INC(net
, insert
);
375 spin_unlock_bh(&nf_conntrack_lock
);
376 help
= nfct_help(ct
);
377 if (help
&& help
->helper
)
378 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
379 #ifdef CONFIG_NF_NAT_NEEDED
380 if (test_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
) ||
381 test_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
))
382 nf_conntrack_event_cache(IPCT_NATINFO
, ct
);
384 nf_conntrack_event_cache(master_ct(ct
) ?
385 IPCT_RELATED
: IPCT_NEW
, ct
);
389 NF_CT_STAT_INC(net
, insert_failed
);
390 spin_unlock_bh(&nf_conntrack_lock
);
393 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
395 /* Returns true if a connection correspondings to the tuple (required
398 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
399 const struct nf_conn
*ignored_conntrack
)
401 struct net
*net
= nf_ct_net(ignored_conntrack
);
402 struct nf_conntrack_tuple_hash
*h
;
403 struct hlist_node
*n
;
404 unsigned int hash
= hash_conntrack(tuple
);
406 /* Disable BHs the entire time since we need to disable them at
407 * least once for the stats anyway.
410 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnode
) {
411 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
412 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
413 NF_CT_STAT_INC(net
, found
);
414 rcu_read_unlock_bh();
417 NF_CT_STAT_INC(net
, searched
);
419 rcu_read_unlock_bh();
423 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
425 #define NF_CT_EVICTION_RANGE 8
427 /* There's a small race here where we may free a just-assured
428 connection. Too bad: we're in trouble anyway. */
429 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
431 /* Use oldest entry, which is roughly LRU */
432 struct nf_conntrack_tuple_hash
*h
;
433 struct nf_conn
*ct
= NULL
, *tmp
;
434 struct hlist_node
*n
;
435 unsigned int i
, cnt
= 0;
439 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
440 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
442 tmp
= nf_ct_tuplehash_to_ctrack(h
);
443 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
448 if (ct
&& unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
450 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
452 hash
= (hash
+ 1) % nf_conntrack_htable_size
;
459 if (del_timer(&ct
->timeout
)) {
460 death_by_timeout((unsigned long)ct
);
462 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
468 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
469 const struct nf_conntrack_tuple
*orig
,
470 const struct nf_conntrack_tuple
*repl
,
473 struct nf_conn
*ct
= NULL
;
475 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
476 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
477 nf_conntrack_hash_rnd_initted
= 1;
480 /* We don't want any race condition at early drop stage */
481 atomic_inc(&net
->ct
.count
);
483 if (nf_conntrack_max
&&
484 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
485 unsigned int hash
= hash_conntrack(orig
);
486 if (!early_drop(net
, hash
)) {
487 atomic_dec(&net
->ct
.count
);
490 "nf_conntrack: table full, dropping"
492 return ERR_PTR(-ENOMEM
);
496 ct
= kmem_cache_zalloc(nf_conntrack_cachep
, gfp
);
498 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
499 atomic_dec(&net
->ct
.count
);
500 return ERR_PTR(-ENOMEM
);
503 atomic_set(&ct
->ct_general
.use
, 1);
504 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
505 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
506 /* Don't set timer yet: wait for confirmation */
507 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
511 INIT_RCU_HEAD(&ct
->rcu
);
515 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
517 static void nf_conntrack_free_rcu(struct rcu_head
*head
)
519 struct nf_conn
*ct
= container_of(head
, struct nf_conn
, rcu
);
520 struct net
*net
= nf_ct_net(ct
);
523 kmem_cache_free(nf_conntrack_cachep
, ct
);
524 atomic_dec(&net
->ct
.count
);
527 void nf_conntrack_free(struct nf_conn
*ct
)
529 nf_ct_ext_destroy(ct
);
530 call_rcu(&ct
->rcu
, nf_conntrack_free_rcu
);
532 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
534 /* Allocate a new conntrack: we return -ENOMEM if classification
535 failed due to stress. Otherwise it really is unclassifiable. */
536 static struct nf_conntrack_tuple_hash
*
537 init_conntrack(struct net
*net
,
538 const struct nf_conntrack_tuple
*tuple
,
539 struct nf_conntrack_l3proto
*l3proto
,
540 struct nf_conntrack_l4proto
*l4proto
,
542 unsigned int dataoff
)
545 struct nf_conn_help
*help
;
546 struct nf_conntrack_tuple repl_tuple
;
547 struct nf_conntrack_expect
*exp
;
549 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
550 pr_debug("Can't invert tuple.\n");
554 ct
= nf_conntrack_alloc(net
, tuple
, &repl_tuple
, GFP_ATOMIC
);
555 if (ct
== NULL
|| IS_ERR(ct
)) {
556 pr_debug("Can't allocate conntrack.\n");
557 return (struct nf_conntrack_tuple_hash
*)ct
;
560 if (!l4proto
->new(ct
, skb
, dataoff
)) {
561 nf_conntrack_free(ct
);
562 pr_debug("init conntrack: can't track with proto module\n");
566 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
568 spin_lock_bh(&nf_conntrack_lock
);
569 exp
= nf_ct_find_expectation(net
, tuple
);
571 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
573 /* Welcome, Mr. Bond. We've been expecting you... */
574 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
575 ct
->master
= exp
->master
;
577 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
579 rcu_assign_pointer(help
->helper
, exp
->helper
);
582 #ifdef CONFIG_NF_CONNTRACK_MARK
583 ct
->mark
= exp
->master
->mark
;
585 #ifdef CONFIG_NF_CONNTRACK_SECMARK
586 ct
->secmark
= exp
->master
->secmark
;
588 nf_conntrack_get(&ct
->master
->ct_general
);
589 NF_CT_STAT_INC(net
, expect_new
);
591 struct nf_conntrack_helper
*helper
;
593 helper
= __nf_ct_helper_find(&repl_tuple
);
595 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
597 rcu_assign_pointer(help
->helper
, helper
);
599 NF_CT_STAT_INC(net
, new);
602 /* Overload tuple linked list to put us in unconfirmed list. */
603 hlist_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
604 &net
->ct
.unconfirmed
);
606 spin_unlock_bh(&nf_conntrack_lock
);
610 exp
->expectfn(ct
, exp
);
611 nf_ct_expect_put(exp
);
614 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
617 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
618 static inline struct nf_conn
*
619 resolve_normal_ct(struct net
*net
,
621 unsigned int dataoff
,
624 struct nf_conntrack_l3proto
*l3proto
,
625 struct nf_conntrack_l4proto
*l4proto
,
627 enum ip_conntrack_info
*ctinfo
)
629 struct nf_conntrack_tuple tuple
;
630 struct nf_conntrack_tuple_hash
*h
;
633 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
634 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
636 pr_debug("resolve_normal_ct: Can't get tuple\n");
640 /* look for tuple match */
641 h
= nf_conntrack_find_get(net
, &tuple
);
643 h
= init_conntrack(net
, &tuple
, l3proto
, l4proto
, skb
, dataoff
);
649 ct
= nf_ct_tuplehash_to_ctrack(h
);
651 /* It exists; we have (non-exclusive) reference. */
652 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
653 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
654 /* Please set reply bit if this packet OK */
657 /* Once we've had two way comms, always ESTABLISHED. */
658 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
659 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
660 *ctinfo
= IP_CT_ESTABLISHED
;
661 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
662 pr_debug("nf_conntrack_in: related packet for %p\n",
664 *ctinfo
= IP_CT_RELATED
;
666 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
671 skb
->nfct
= &ct
->ct_general
;
672 skb
->nfctinfo
= *ctinfo
;
677 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
681 enum ip_conntrack_info ctinfo
;
682 struct nf_conntrack_l3proto
*l3proto
;
683 struct nf_conntrack_l4proto
*l4proto
;
684 unsigned int dataoff
;
689 /* Previously seen (loopback or untracked)? Ignore. */
691 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
695 /* rcu_read_lock()ed by nf_hook_slow */
696 l3proto
= __nf_ct_l3proto_find(pf
);
697 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
698 &dataoff
, &protonum
);
700 pr_debug("not prepared to track yet or error occured\n");
701 NF_CT_STAT_INC_ATOMIC(net
, error
);
702 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
706 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
708 /* It may be an special packet, error, unclean...
709 * inverse of the return code tells to the netfilter
710 * core what to do with the packet. */
711 if (l4proto
->error
!= NULL
) {
712 ret
= l4proto
->error(net
, skb
, dataoff
, &ctinfo
, pf
, hooknum
);
714 NF_CT_STAT_INC_ATOMIC(net
, error
);
715 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
720 ct
= resolve_normal_ct(net
, skb
, dataoff
, pf
, protonum
,
721 l3proto
, l4proto
, &set_reply
, &ctinfo
);
723 /* Not valid part of a connection */
724 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
729 /* Too stressed to deal. */
730 NF_CT_STAT_INC_ATOMIC(net
, drop
);
734 NF_CT_ASSERT(skb
->nfct
);
736 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
738 /* Invalid: inverse of the return code tells
739 * the netfilter core what to do */
740 pr_debug("nf_conntrack_in: Can't track with proto module\n");
741 nf_conntrack_put(skb
->nfct
);
743 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
747 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
748 nf_conntrack_event_cache(IPCT_STATUS
, ct
);
752 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
754 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
755 const struct nf_conntrack_tuple
*orig
)
760 ret
= nf_ct_invert_tuple(inverse
, orig
,
761 __nf_ct_l3proto_find(orig
->src
.l3num
),
762 __nf_ct_l4proto_find(orig
->src
.l3num
,
763 orig
->dst
.protonum
));
767 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
769 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
770 implicitly racy: see __nf_conntrack_confirm */
771 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
772 const struct nf_conntrack_tuple
*newreply
)
774 struct nf_conn_help
*help
= nfct_help(ct
);
775 struct nf_conntrack_helper
*helper
;
777 /* Should be unconfirmed, so not in hash table yet */
778 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
780 pr_debug("Altering reply tuple of %p to ", ct
);
781 nf_ct_dump_tuple(newreply
);
783 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
784 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
788 helper
= __nf_ct_helper_find(newreply
);
789 if (helper
== NULL
) {
791 rcu_assign_pointer(help
->helper
, NULL
);
796 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
800 memset(&help
->help
, 0, sizeof(help
->help
));
803 rcu_assign_pointer(help
->helper
, helper
);
807 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
809 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
810 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
811 enum ip_conntrack_info ctinfo
,
812 const struct sk_buff
*skb
,
813 unsigned long extra_jiffies
,
818 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
821 spin_lock_bh(&nf_conntrack_lock
);
823 /* Only update if this is not a fixed timeout */
824 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
827 /* If not in hash table, timer will not be active yet */
828 if (!nf_ct_is_confirmed(ct
)) {
829 ct
->timeout
.expires
= extra_jiffies
;
830 event
= IPCT_REFRESH
;
832 unsigned long newtime
= jiffies
+ extra_jiffies
;
834 /* Only update the timeout if the new timeout is at least
835 HZ jiffies from the old timeout. Need del_timer for race
836 avoidance (may already be dying). */
837 if (newtime
- ct
->timeout
.expires
>= HZ
838 && del_timer(&ct
->timeout
)) {
839 ct
->timeout
.expires
= newtime
;
840 add_timer(&ct
->timeout
);
841 event
= IPCT_REFRESH
;
847 struct nf_conn_counter
*acct
;
849 acct
= nf_conn_acct_find(ct
);
851 acct
[CTINFO2DIR(ctinfo
)].packets
++;
852 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
853 skb
->len
- skb_network_offset(skb
);
857 spin_unlock_bh(&nf_conntrack_lock
);
859 /* must be unlocked when calling event cache */
861 nf_conntrack_event_cache(event
, ct
);
863 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
865 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
866 enum ip_conntrack_info ctinfo
,
867 const struct sk_buff
*skb
,
871 struct nf_conn_counter
*acct
;
873 spin_lock_bh(&nf_conntrack_lock
);
874 acct
= nf_conn_acct_find(ct
);
876 acct
[CTINFO2DIR(ctinfo
)].packets
++;
877 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
878 skb
->len
- skb_network_offset(skb
);
880 spin_unlock_bh(&nf_conntrack_lock
);
883 if (del_timer(&ct
->timeout
)) {
884 ct
->timeout
.function((unsigned long)ct
);
889 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
891 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
893 #include <linux/netfilter/nfnetlink.h>
894 #include <linux/netfilter/nfnetlink_conntrack.h>
895 #include <linux/mutex.h>
897 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
898 * in ip_conntrack_core, since we don't want the protocols to autoload
899 * or depend on ctnetlink */
900 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
901 const struct nf_conntrack_tuple
*tuple
)
903 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
904 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
910 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
912 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
913 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
914 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
916 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
918 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
919 struct nf_conntrack_tuple
*t
)
921 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
924 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
925 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
929 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
932 /* Used by ipt_REJECT and ip6t_REJECT. */
933 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
936 enum ip_conntrack_info ctinfo
;
938 /* This ICMP is in reverse direction to the packet which caused it */
939 ct
= nf_ct_get(skb
, &ctinfo
);
940 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
941 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
943 ctinfo
= IP_CT_RELATED
;
945 /* Attach to new skbuff, and increment count */
946 nskb
->nfct
= &ct
->ct_general
;
947 nskb
->nfctinfo
= ctinfo
;
948 nf_conntrack_get(nskb
->nfct
);
951 /* Bring out ya dead! */
952 static struct nf_conn
*
953 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
954 void *data
, unsigned int *bucket
)
956 struct nf_conntrack_tuple_hash
*h
;
958 struct hlist_node
*n
;
960 spin_lock_bh(&nf_conntrack_lock
);
961 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
962 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnode
) {
963 ct
= nf_ct_tuplehash_to_ctrack(h
);
968 hlist_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnode
) {
969 ct
= nf_ct_tuplehash_to_ctrack(h
);
971 set_bit(IPS_DYING_BIT
, &ct
->status
);
973 spin_unlock_bh(&nf_conntrack_lock
);
976 atomic_inc(&ct
->ct_general
.use
);
977 spin_unlock_bh(&nf_conntrack_lock
);
981 void nf_ct_iterate_cleanup(struct net
*net
,
982 int (*iter
)(struct nf_conn
*i
, void *data
),
986 unsigned int bucket
= 0;
988 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
989 /* Time to push up daises... */
990 if (del_timer(&ct
->timeout
))
991 death_by_timeout((unsigned long)ct
);
992 /* ... else the timer will get him soon. */
997 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
999 static int kill_all(struct nf_conn
*i
, void *data
)
1004 void nf_ct_free_hashtable(struct hlist_head
*hash
, int vmalloced
, unsigned int size
)
1009 free_pages((unsigned long)hash
,
1010 get_order(sizeof(struct hlist_head
) * size
));
1012 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1014 void nf_conntrack_flush(struct net
*net
)
1016 nf_ct_iterate_cleanup(net
, kill_all
, NULL
);
1018 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
1020 static void nf_conntrack_cleanup_init_net(void)
1022 nf_conntrack_helper_fini();
1023 nf_conntrack_proto_fini();
1024 kmem_cache_destroy(nf_conntrack_cachep
);
1027 static void nf_conntrack_cleanup_net(struct net
*net
)
1029 nf_ct_event_cache_flush(net
);
1030 nf_conntrack_ecache_fini(net
);
1032 nf_conntrack_flush(net
);
1033 if (atomic_read(&net
->ct
.count
) != 0) {
1035 goto i_see_dead_people
;
1037 /* wait until all references to nf_conntrack_untracked are dropped */
1038 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1041 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1042 nf_conntrack_htable_size
);
1043 nf_conntrack_acct_fini(net
);
1044 nf_conntrack_expect_fini(net
);
1045 free_percpu(net
->ct
.stat
);
1048 /* Mishearing the voices in his head, our hero wonders how he's
1049 supposed to kill the mall. */
1050 void nf_conntrack_cleanup(struct net
*net
)
1052 if (net_eq(net
, &init_net
))
1053 rcu_assign_pointer(ip_ct_attach
, NULL
);
1055 /* This makes sure all current packets have passed through
1056 netfilter framework. Roll on, two-stage module
1060 nf_conntrack_cleanup_net(net
);
1062 if (net_eq(net
, &init_net
)) {
1063 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1064 nf_conntrack_cleanup_init_net();
1068 struct hlist_head
*nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
)
1070 struct hlist_head
*hash
;
1071 unsigned int size
, i
;
1075 size
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_head
));
1076 hash
= (void*)__get_free_pages(GFP_KERNEL
|__GFP_NOWARN
,
1077 get_order(sizeof(struct hlist_head
)
1081 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1082 hash
= vmalloc(sizeof(struct hlist_head
) * size
);
1086 for (i
= 0; i
< size
; i
++)
1087 INIT_HLIST_HEAD(&hash
[i
]);
1091 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1093 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1095 int i
, bucket
, vmalloced
, old_vmalloced
;
1096 unsigned int hashsize
, old_size
;
1098 struct hlist_head
*hash
, *old_hash
;
1099 struct nf_conntrack_tuple_hash
*h
;
1101 /* On boot, we can set this without any fancy locking. */
1102 if (!nf_conntrack_htable_size
)
1103 return param_set_uint(val
, kp
);
1105 hashsize
= simple_strtoul(val
, NULL
, 0);
1109 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
);
1113 /* We have to rehahs for the new table anyway, so we also can
1114 * use a newrandom seed */
1115 get_random_bytes(&rnd
, 4);
1117 /* Lookups in the old hash might happen in parallel, which means we
1118 * might get false negatives during connection lookup. New connections
1119 * created because of a false negative won't make it into the hash
1120 * though since that required taking the lock.
1122 spin_lock_bh(&nf_conntrack_lock
);
1123 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1124 while (!hlist_empty(&init_net
.ct
.hash
[i
])) {
1125 h
= hlist_entry(init_net
.ct
.hash
[i
].first
,
1126 struct nf_conntrack_tuple_hash
, hnode
);
1127 hlist_del_rcu(&h
->hnode
);
1128 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1129 hlist_add_head(&h
->hnode
, &hash
[bucket
]);
1132 old_size
= nf_conntrack_htable_size
;
1133 old_vmalloced
= init_net
.ct
.hash_vmalloc
;
1134 old_hash
= init_net
.ct
.hash
;
1136 nf_conntrack_htable_size
= hashsize
;
1137 init_net
.ct
.hash_vmalloc
= vmalloced
;
1138 init_net
.ct
.hash
= hash
;
1139 nf_conntrack_hash_rnd
= rnd
;
1140 spin_unlock_bh(&nf_conntrack_lock
);
1142 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1145 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1147 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1148 &nf_conntrack_htable_size
, 0600);
1150 static int nf_conntrack_init_init_net(void)
1155 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1156 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1157 if (!nf_conntrack_htable_size
) {
1158 nf_conntrack_htable_size
1159 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1160 / sizeof(struct hlist_head
));
1161 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1162 nf_conntrack_htable_size
= 16384;
1163 if (nf_conntrack_htable_size
< 32)
1164 nf_conntrack_htable_size
= 32;
1166 /* Use a max. factor of four by default to get the same max as
1167 * with the old struct list_heads. When a table size is given
1168 * we use the old value of 8 to avoid reducing the max.
1172 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1174 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1175 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1178 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1179 sizeof(struct nf_conn
),
1181 if (!nf_conntrack_cachep
) {
1182 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1187 ret
= nf_conntrack_proto_init();
1191 ret
= nf_conntrack_helper_init();
1198 nf_conntrack_proto_fini();
1200 kmem_cache_destroy(nf_conntrack_cachep
);
1205 static int nf_conntrack_init_net(struct net
*net
)
1209 atomic_set(&net
->ct
.count
, 0);
1210 INIT_HLIST_HEAD(&net
->ct
.unconfirmed
);
1211 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1212 if (!net
->ct
.stat
) {
1216 ret
= nf_conntrack_ecache_init(net
);
1219 net
->ct
.hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
,
1220 &net
->ct
.hash_vmalloc
);
1221 if (!net
->ct
.hash
) {
1223 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1226 ret
= nf_conntrack_expect_init(net
);
1229 ret
= nf_conntrack_acct_init(net
);
1233 /* Set up fake conntrack:
1234 - to never be deleted, not in any hashes */
1235 #ifdef CONFIG_NET_NS
1236 nf_conntrack_untracked
.ct_net
= &init_net
;
1238 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1239 /* - and look it like as a confirmed connection */
1240 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1245 nf_conntrack_expect_fini(net
);
1247 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1248 nf_conntrack_htable_size
);
1250 nf_conntrack_ecache_fini(net
);
1252 free_percpu(net
->ct
.stat
);
1257 int nf_conntrack_init(struct net
*net
)
1261 if (net_eq(net
, &init_net
)) {
1262 ret
= nf_conntrack_init_init_net();
1266 ret
= nf_conntrack_init_net(net
);
1270 if (net_eq(net
, &init_net
)) {
1271 /* For use by REJECT target */
1272 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1273 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1278 if (net_eq(net
, &init_net
))
1279 nf_conntrack_cleanup_init_net();