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 unsigned int nf_ct_log_invalid __read_mostly
;
57 HLIST_HEAD(unconfirmed
);
58 static struct kmem_cache
*nf_conntrack_cachep __read_mostly
;
60 DEFINE_PER_CPU(struct ip_conntrack_stat
, nf_conntrack_stat
);
61 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat
);
63 static int nf_conntrack_hash_rnd_initted
;
64 static unsigned int nf_conntrack_hash_rnd
;
66 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
67 unsigned int size
, unsigned int rnd
)
72 /* The direction must be ignored, so we hash everything up to the
73 * destination ports (which is a multiple of 4) and treat the last
74 * three bytes manually.
76 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
77 h
= jhash2((u32
*)tuple
, n
,
78 rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
79 tuple
->dst
.protonum
));
81 return ((u64
)h
* size
) >> 32;
84 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
86 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
87 nf_conntrack_hash_rnd
);
91 nf_ct_get_tuple(const struct sk_buff
*skb
,
96 struct nf_conntrack_tuple
*tuple
,
97 const struct nf_conntrack_l3proto
*l3proto
,
98 const struct nf_conntrack_l4proto
*l4proto
)
100 memset(tuple
, 0, sizeof(*tuple
));
102 tuple
->src
.l3num
= l3num
;
103 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
106 tuple
->dst
.protonum
= protonum
;
107 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
109 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
111 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
113 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
114 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
116 struct nf_conntrack_l3proto
*l3proto
;
117 struct nf_conntrack_l4proto
*l4proto
;
118 unsigned int protoff
;
124 l3proto
= __nf_ct_l3proto_find(l3num
);
125 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
126 if (ret
!= NF_ACCEPT
) {
131 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
133 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
139 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
142 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
143 const struct nf_conntrack_tuple
*orig
,
144 const struct nf_conntrack_l3proto
*l3proto
,
145 const struct nf_conntrack_l4proto
*l4proto
)
147 memset(inverse
, 0, sizeof(*inverse
));
149 inverse
->src
.l3num
= orig
->src
.l3num
;
150 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
153 inverse
->dst
.dir
= !orig
->dst
.dir
;
155 inverse
->dst
.protonum
= orig
->dst
.protonum
;
156 return l4proto
->invert_tuple(inverse
, orig
);
158 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
161 clean_from_lists(struct nf_conn
*ct
)
163 pr_debug("clean_from_lists(%p)\n", ct
);
164 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
165 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
);
167 /* Destroy all pending expectations */
168 nf_ct_remove_expectations(ct
);
172 destroy_conntrack(struct nf_conntrack
*nfct
)
174 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
175 struct nf_conntrack_l4proto
*l4proto
;
177 pr_debug("destroy_conntrack(%p)\n", ct
);
178 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
179 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
181 nf_conntrack_event(IPCT_DESTROY
, ct
);
182 set_bit(IPS_DYING_BIT
, &ct
->status
);
184 /* To make sure we don't get any weird locking issues here:
185 * destroy_conntrack() MUST NOT be called with a write lock
186 * to nf_conntrack_lock!!! -HW */
188 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
189 if (l4proto
&& l4proto
->destroy
)
190 l4proto
->destroy(ct
);
194 spin_lock_bh(&nf_conntrack_lock
);
195 /* Expectations will have been removed in clean_from_lists,
196 * except TFTP can create an expectation on the first packet,
197 * before connection is in the list, so we need to clean here,
199 nf_ct_remove_expectations(ct
);
201 /* We overload first tuple to link into unconfirmed list. */
202 if (!nf_ct_is_confirmed(ct
)) {
203 BUG_ON(hlist_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
));
204 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
207 NF_CT_STAT_INC(delete);
208 spin_unlock_bh(&nf_conntrack_lock
);
211 nf_ct_put(ct
->master
);
213 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
214 nf_conntrack_free(ct
);
217 static void death_by_timeout(unsigned long ul_conntrack
)
219 struct nf_conn
*ct
= (void *)ul_conntrack
;
220 struct nf_conn_help
*help
= nfct_help(ct
);
221 struct nf_conntrack_helper
*helper
;
225 helper
= rcu_dereference(help
->helper
);
226 if (helper
&& helper
->destroy
)
231 spin_lock_bh(&nf_conntrack_lock
);
232 /* Inside lock so preempt is disabled on module removal path.
233 * Otherwise we can get spurious warnings. */
234 NF_CT_STAT_INC(delete_list
);
235 clean_from_lists(ct
);
236 spin_unlock_bh(&nf_conntrack_lock
);
240 struct nf_conntrack_tuple_hash
*
241 __nf_conntrack_find(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
243 struct nf_conntrack_tuple_hash
*h
;
244 struct hlist_node
*n
;
245 unsigned int hash
= hash_conntrack(tuple
);
247 /* Disable BHs the entire time since we normally need to disable them
248 * at least once for the stats anyway.
251 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnode
) {
252 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
253 NF_CT_STAT_INC(found
);
257 NF_CT_STAT_INC(searched
);
263 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
265 /* Find a connection corresponding to a tuple. */
266 struct nf_conntrack_tuple_hash
*
267 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
269 struct nf_conntrack_tuple_hash
*h
;
273 h
= __nf_conntrack_find(net
, tuple
);
275 ct
= nf_ct_tuplehash_to_ctrack(h
);
276 if (unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
283 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
285 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
287 unsigned int repl_hash
)
289 struct net
*net
= nf_ct_net(ct
);
291 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
292 &net
->ct
.hash
[hash
]);
293 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
,
294 &net
->ct
.hash
[repl_hash
]);
297 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
299 unsigned int hash
, repl_hash
;
301 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
302 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
304 spin_lock_bh(&nf_conntrack_lock
);
305 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
306 spin_unlock_bh(&nf_conntrack_lock
);
308 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
310 /* Confirm a connection given skb; places it in hash table */
312 __nf_conntrack_confirm(struct sk_buff
*skb
)
314 unsigned int hash
, repl_hash
;
315 struct nf_conntrack_tuple_hash
*h
;
317 struct nf_conn_help
*help
;
318 struct hlist_node
*n
;
319 enum ip_conntrack_info ctinfo
;
322 ct
= nf_ct_get(skb
, &ctinfo
);
325 /* ipt_REJECT uses nf_conntrack_attach to attach related
326 ICMP/TCP RST packets in other direction. Actual packet
327 which created connection will be IP_CT_NEW or for an
328 expected connection, IP_CT_RELATED. */
329 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
332 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
333 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
335 /* We're not in hash table, and we refuse to set up related
336 connections for unconfirmed conns. But packet copies and
337 REJECT will give spurious warnings here. */
338 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
340 /* No external references means noone else could have
342 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
343 pr_debug("Confirming conntrack %p\n", ct
);
345 spin_lock_bh(&nf_conntrack_lock
);
347 /* See if there's one in the list already, including reverse:
348 NAT could have grabbed it without realizing, since we're
349 not in the hash. If there is, we lost race. */
350 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnode
)
351 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
354 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnode
)
355 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
359 /* Remove from unconfirmed list */
360 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
362 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
363 /* Timer relative to confirmation time, not original
364 setting time, otherwise we'd get timer wrap in
365 weird delay cases. */
366 ct
->timeout
.expires
+= jiffies
;
367 add_timer(&ct
->timeout
);
368 atomic_inc(&ct
->ct_general
.use
);
369 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
370 NF_CT_STAT_INC(insert
);
371 spin_unlock_bh(&nf_conntrack_lock
);
372 help
= nfct_help(ct
);
373 if (help
&& help
->helper
)
374 nf_conntrack_event_cache(IPCT_HELPER
, skb
);
375 #ifdef CONFIG_NF_NAT_NEEDED
376 if (test_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
) ||
377 test_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
))
378 nf_conntrack_event_cache(IPCT_NATINFO
, skb
);
380 nf_conntrack_event_cache(master_ct(ct
) ?
381 IPCT_RELATED
: IPCT_NEW
, skb
);
385 NF_CT_STAT_INC(insert_failed
);
386 spin_unlock_bh(&nf_conntrack_lock
);
389 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
391 /* Returns true if a connection correspondings to the tuple (required
394 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
395 const struct nf_conn
*ignored_conntrack
)
397 struct net
*net
= nf_ct_net(ignored_conntrack
);
398 struct nf_conntrack_tuple_hash
*h
;
399 struct hlist_node
*n
;
400 unsigned int hash
= hash_conntrack(tuple
);
402 /* Disable BHs the entire time since we need to disable them at
403 * least once for the stats anyway.
406 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnode
) {
407 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
408 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
409 NF_CT_STAT_INC(found
);
410 rcu_read_unlock_bh();
413 NF_CT_STAT_INC(searched
);
415 rcu_read_unlock_bh();
419 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
421 #define NF_CT_EVICTION_RANGE 8
423 /* There's a small race here where we may free a just-assured
424 connection. Too bad: we're in trouble anyway. */
425 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
427 /* Use oldest entry, which is roughly LRU */
428 struct nf_conntrack_tuple_hash
*h
;
429 struct nf_conn
*ct
= NULL
, *tmp
;
430 struct hlist_node
*n
;
431 unsigned int i
, cnt
= 0;
435 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
436 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
438 tmp
= nf_ct_tuplehash_to_ctrack(h
);
439 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
444 if (ct
&& unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
446 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
448 hash
= (hash
+ 1) % nf_conntrack_htable_size
;
455 if (del_timer(&ct
->timeout
)) {
456 death_by_timeout((unsigned long)ct
);
458 NF_CT_STAT_INC_ATOMIC(early_drop
);
464 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
465 const struct nf_conntrack_tuple
*orig
,
466 const struct nf_conntrack_tuple
*repl
,
469 struct nf_conn
*ct
= NULL
;
471 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
472 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
473 nf_conntrack_hash_rnd_initted
= 1;
476 /* We don't want any race condition at early drop stage */
477 atomic_inc(&net
->ct
.count
);
479 if (nf_conntrack_max
&&
480 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
481 unsigned int hash
= hash_conntrack(orig
);
482 if (!early_drop(net
, hash
)) {
483 atomic_dec(&net
->ct
.count
);
486 "nf_conntrack: table full, dropping"
488 return ERR_PTR(-ENOMEM
);
492 ct
= kmem_cache_zalloc(nf_conntrack_cachep
, gfp
);
494 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
495 atomic_dec(&net
->ct
.count
);
496 return ERR_PTR(-ENOMEM
);
499 atomic_set(&ct
->ct_general
.use
, 1);
500 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
501 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
502 /* Don't set timer yet: wait for confirmation */
503 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
507 INIT_RCU_HEAD(&ct
->rcu
);
511 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
513 static void nf_conntrack_free_rcu(struct rcu_head
*head
)
515 struct nf_conn
*ct
= container_of(head
, struct nf_conn
, rcu
);
516 struct net
*net
= nf_ct_net(ct
);
519 kmem_cache_free(nf_conntrack_cachep
, ct
);
520 atomic_dec(&net
->ct
.count
);
523 void nf_conntrack_free(struct nf_conn
*ct
)
525 nf_ct_ext_destroy(ct
);
526 call_rcu(&ct
->rcu
, nf_conntrack_free_rcu
);
528 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
530 /* Allocate a new conntrack: we return -ENOMEM if classification
531 failed due to stress. Otherwise it really is unclassifiable. */
532 static struct nf_conntrack_tuple_hash
*
533 init_conntrack(struct net
*net
,
534 const struct nf_conntrack_tuple
*tuple
,
535 struct nf_conntrack_l3proto
*l3proto
,
536 struct nf_conntrack_l4proto
*l4proto
,
538 unsigned int dataoff
)
541 struct nf_conn_help
*help
;
542 struct nf_conntrack_tuple repl_tuple
;
543 struct nf_conntrack_expect
*exp
;
545 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
546 pr_debug("Can't invert tuple.\n");
550 ct
= nf_conntrack_alloc(net
, tuple
, &repl_tuple
, GFP_ATOMIC
);
551 if (ct
== NULL
|| IS_ERR(ct
)) {
552 pr_debug("Can't allocate conntrack.\n");
553 return (struct nf_conntrack_tuple_hash
*)ct
;
556 if (!l4proto
->new(ct
, skb
, dataoff
)) {
557 nf_conntrack_free(ct
);
558 pr_debug("init conntrack: can't track with proto module\n");
562 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
564 spin_lock_bh(&nf_conntrack_lock
);
565 exp
= nf_ct_find_expectation(tuple
);
567 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
569 /* Welcome, Mr. Bond. We've been expecting you... */
570 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
571 ct
->master
= exp
->master
;
573 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
575 rcu_assign_pointer(help
->helper
, exp
->helper
);
578 #ifdef CONFIG_NF_CONNTRACK_MARK
579 ct
->mark
= exp
->master
->mark
;
581 #ifdef CONFIG_NF_CONNTRACK_SECMARK
582 ct
->secmark
= exp
->master
->secmark
;
584 nf_conntrack_get(&ct
->master
->ct_general
);
585 NF_CT_STAT_INC(expect_new
);
587 struct nf_conntrack_helper
*helper
;
589 helper
= __nf_ct_helper_find(&repl_tuple
);
591 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
593 rcu_assign_pointer(help
->helper
, helper
);
598 /* Overload tuple linked list to put us in unconfirmed list. */
599 hlist_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
, &unconfirmed
);
601 spin_unlock_bh(&nf_conntrack_lock
);
605 exp
->expectfn(ct
, exp
);
606 nf_ct_expect_put(exp
);
609 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
612 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
613 static inline struct nf_conn
*
614 resolve_normal_ct(struct sk_buff
*skb
,
615 unsigned int dataoff
,
618 struct nf_conntrack_l3proto
*l3proto
,
619 struct nf_conntrack_l4proto
*l4proto
,
621 enum ip_conntrack_info
*ctinfo
)
623 struct nf_conntrack_tuple tuple
;
624 struct nf_conntrack_tuple_hash
*h
;
627 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
628 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
630 pr_debug("resolve_normal_ct: Can't get tuple\n");
634 /* look for tuple match */
635 h
= nf_conntrack_find_get(&init_net
, &tuple
);
637 h
= init_conntrack(&init_net
, &tuple
, l3proto
, l4proto
, skb
,
644 ct
= nf_ct_tuplehash_to_ctrack(h
);
646 /* It exists; we have (non-exclusive) reference. */
647 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
648 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
649 /* Please set reply bit if this packet OK */
652 /* Once we've had two way comms, always ESTABLISHED. */
653 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
654 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
655 *ctinfo
= IP_CT_ESTABLISHED
;
656 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
657 pr_debug("nf_conntrack_in: related packet for %p\n",
659 *ctinfo
= IP_CT_RELATED
;
661 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
666 skb
->nfct
= &ct
->ct_general
;
667 skb
->nfctinfo
= *ctinfo
;
672 nf_conntrack_in(u_int8_t pf
, unsigned int hooknum
, struct sk_buff
*skb
)
675 enum ip_conntrack_info ctinfo
;
676 struct nf_conntrack_l3proto
*l3proto
;
677 struct nf_conntrack_l4proto
*l4proto
;
678 unsigned int dataoff
;
683 /* Previously seen (loopback or untracked)? Ignore. */
685 NF_CT_STAT_INC_ATOMIC(ignore
);
689 /* rcu_read_lock()ed by nf_hook_slow */
690 l3proto
= __nf_ct_l3proto_find(pf
);
691 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
692 &dataoff
, &protonum
);
694 pr_debug("not prepared to track yet or error occured\n");
695 NF_CT_STAT_INC_ATOMIC(error
);
696 NF_CT_STAT_INC_ATOMIC(invalid
);
700 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
702 /* It may be an special packet, error, unclean...
703 * inverse of the return code tells to the netfilter
704 * core what to do with the packet. */
705 if (l4proto
->error
!= NULL
&&
706 (ret
= l4proto
->error(skb
, dataoff
, &ctinfo
, pf
, hooknum
)) <= 0) {
707 NF_CT_STAT_INC_ATOMIC(error
);
708 NF_CT_STAT_INC_ATOMIC(invalid
);
712 ct
= resolve_normal_ct(skb
, dataoff
, pf
, protonum
, l3proto
, l4proto
,
713 &set_reply
, &ctinfo
);
715 /* Not valid part of a connection */
716 NF_CT_STAT_INC_ATOMIC(invalid
);
721 /* Too stressed to deal. */
722 NF_CT_STAT_INC_ATOMIC(drop
);
726 NF_CT_ASSERT(skb
->nfct
);
728 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
730 /* Invalid: inverse of the return code tells
731 * the netfilter core what to do */
732 pr_debug("nf_conntrack_in: Can't track with proto module\n");
733 nf_conntrack_put(skb
->nfct
);
735 NF_CT_STAT_INC_ATOMIC(invalid
);
739 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
740 nf_conntrack_event_cache(IPCT_STATUS
, skb
);
744 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
746 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
747 const struct nf_conntrack_tuple
*orig
)
752 ret
= nf_ct_invert_tuple(inverse
, orig
,
753 __nf_ct_l3proto_find(orig
->src
.l3num
),
754 __nf_ct_l4proto_find(orig
->src
.l3num
,
755 orig
->dst
.protonum
));
759 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
761 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
762 implicitly racy: see __nf_conntrack_confirm */
763 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
764 const struct nf_conntrack_tuple
*newreply
)
766 struct nf_conn_help
*help
= nfct_help(ct
);
767 struct nf_conntrack_helper
*helper
;
769 /* Should be unconfirmed, so not in hash table yet */
770 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
772 pr_debug("Altering reply tuple of %p to ", ct
);
773 nf_ct_dump_tuple(newreply
);
775 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
776 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
780 helper
= __nf_ct_helper_find(newreply
);
781 if (helper
== NULL
) {
783 rcu_assign_pointer(help
->helper
, NULL
);
788 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
792 memset(&help
->help
, 0, sizeof(help
->help
));
795 rcu_assign_pointer(help
->helper
, helper
);
799 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
801 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
802 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
803 enum ip_conntrack_info ctinfo
,
804 const struct sk_buff
*skb
,
805 unsigned long extra_jiffies
,
810 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
813 spin_lock_bh(&nf_conntrack_lock
);
815 /* Only update if this is not a fixed timeout */
816 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
819 /* If not in hash table, timer will not be active yet */
820 if (!nf_ct_is_confirmed(ct
)) {
821 ct
->timeout
.expires
= extra_jiffies
;
822 event
= IPCT_REFRESH
;
824 unsigned long newtime
= jiffies
+ extra_jiffies
;
826 /* Only update the timeout if the new timeout is at least
827 HZ jiffies from the old timeout. Need del_timer for race
828 avoidance (may already be dying). */
829 if (newtime
- ct
->timeout
.expires
>= HZ
830 && del_timer(&ct
->timeout
)) {
831 ct
->timeout
.expires
= newtime
;
832 add_timer(&ct
->timeout
);
833 event
= IPCT_REFRESH
;
839 struct nf_conn_counter
*acct
;
841 acct
= nf_conn_acct_find(ct
);
843 acct
[CTINFO2DIR(ctinfo
)].packets
++;
844 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
845 skb
->len
- skb_network_offset(skb
);
849 spin_unlock_bh(&nf_conntrack_lock
);
851 /* must be unlocked when calling event cache */
853 nf_conntrack_event_cache(event
, skb
);
855 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
857 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
858 enum ip_conntrack_info ctinfo
,
859 const struct sk_buff
*skb
,
863 struct nf_conn_counter
*acct
;
865 spin_lock_bh(&nf_conntrack_lock
);
866 acct
= nf_conn_acct_find(ct
);
868 acct
[CTINFO2DIR(ctinfo
)].packets
++;
869 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
870 skb
->len
- skb_network_offset(skb
);
872 spin_unlock_bh(&nf_conntrack_lock
);
875 if (del_timer(&ct
->timeout
)) {
876 ct
->timeout
.function((unsigned long)ct
);
881 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
883 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
885 #include <linux/netfilter/nfnetlink.h>
886 #include <linux/netfilter/nfnetlink_conntrack.h>
887 #include <linux/mutex.h>
889 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
890 * in ip_conntrack_core, since we don't want the protocols to autoload
891 * or depend on ctnetlink */
892 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
893 const struct nf_conntrack_tuple
*tuple
)
895 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
896 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
902 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
904 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
905 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
906 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
908 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
910 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
911 struct nf_conntrack_tuple
*t
)
913 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
916 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
917 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
921 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
924 /* Used by ipt_REJECT and ip6t_REJECT. */
925 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
928 enum ip_conntrack_info ctinfo
;
930 /* This ICMP is in reverse direction to the packet which caused it */
931 ct
= nf_ct_get(skb
, &ctinfo
);
932 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
933 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
935 ctinfo
= IP_CT_RELATED
;
937 /* Attach to new skbuff, and increment count */
938 nskb
->nfct
= &ct
->ct_general
;
939 nskb
->nfctinfo
= ctinfo
;
940 nf_conntrack_get(nskb
->nfct
);
943 /* Bring out ya dead! */
944 static struct nf_conn
*
945 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
946 void *data
, unsigned int *bucket
)
948 struct nf_conntrack_tuple_hash
*h
;
950 struct hlist_node
*n
;
952 spin_lock_bh(&nf_conntrack_lock
);
953 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
954 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnode
) {
955 ct
= nf_ct_tuplehash_to_ctrack(h
);
960 hlist_for_each_entry(h
, n
, &unconfirmed
, hnode
) {
961 ct
= nf_ct_tuplehash_to_ctrack(h
);
963 set_bit(IPS_DYING_BIT
, &ct
->status
);
965 spin_unlock_bh(&nf_conntrack_lock
);
968 atomic_inc(&ct
->ct_general
.use
);
969 spin_unlock_bh(&nf_conntrack_lock
);
973 void nf_ct_iterate_cleanup(struct net
*net
,
974 int (*iter
)(struct nf_conn
*i
, void *data
),
978 unsigned int bucket
= 0;
980 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
981 /* Time to push up daises... */
982 if (del_timer(&ct
->timeout
))
983 death_by_timeout((unsigned long)ct
);
984 /* ... else the timer will get him soon. */
989 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
991 static int kill_all(struct nf_conn
*i
, void *data
)
996 void nf_ct_free_hashtable(struct hlist_head
*hash
, int vmalloced
, unsigned int size
)
1001 free_pages((unsigned long)hash
,
1002 get_order(sizeof(struct hlist_head
) * size
));
1004 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1006 void nf_conntrack_flush(struct net
*net
)
1008 nf_ct_iterate_cleanup(net
, kill_all
, NULL
);
1010 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
1012 /* Mishearing the voices in his head, our hero wonders how he's
1013 supposed to kill the mall. */
1014 void nf_conntrack_cleanup(struct net
*net
)
1016 rcu_assign_pointer(ip_ct_attach
, NULL
);
1018 /* This makes sure all current packets have passed through
1019 netfilter framework. Roll on, two-stage module
1023 nf_ct_event_cache_flush();
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 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1036 kmem_cache_destroy(nf_conntrack_cachep
);
1037 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1038 nf_conntrack_htable_size
);
1040 nf_conntrack_acct_fini();
1041 nf_conntrack_expect_fini();
1042 nf_conntrack_helper_fini();
1043 nf_conntrack_proto_fini();
1046 struct hlist_head
*nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
)
1048 struct hlist_head
*hash
;
1049 unsigned int size
, i
;
1053 size
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_head
));
1054 hash
= (void*)__get_free_pages(GFP_KERNEL
|__GFP_NOWARN
,
1055 get_order(sizeof(struct hlist_head
)
1059 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1060 hash
= vmalloc(sizeof(struct hlist_head
) * size
);
1064 for (i
= 0; i
< size
; i
++)
1065 INIT_HLIST_HEAD(&hash
[i
]);
1069 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1071 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1073 int i
, bucket
, vmalloced
, old_vmalloced
;
1074 unsigned int hashsize
, old_size
;
1076 struct hlist_head
*hash
, *old_hash
;
1077 struct nf_conntrack_tuple_hash
*h
;
1079 /* On boot, we can set this without any fancy locking. */
1080 if (!nf_conntrack_htable_size
)
1081 return param_set_uint(val
, kp
);
1083 hashsize
= simple_strtoul(val
, NULL
, 0);
1087 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
);
1091 /* We have to rehahs for the new table anyway, so we also can
1092 * use a newrandom seed */
1093 get_random_bytes(&rnd
, 4);
1095 /* Lookups in the old hash might happen in parallel, which means we
1096 * might get false negatives during connection lookup. New connections
1097 * created because of a false negative won't make it into the hash
1098 * though since that required taking the lock.
1100 spin_lock_bh(&nf_conntrack_lock
);
1101 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1102 while (!hlist_empty(&init_net
.ct
.hash
[i
])) {
1103 h
= hlist_entry(init_net
.ct
.hash
[i
].first
,
1104 struct nf_conntrack_tuple_hash
, hnode
);
1105 hlist_del_rcu(&h
->hnode
);
1106 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1107 hlist_add_head(&h
->hnode
, &hash
[bucket
]);
1110 old_size
= nf_conntrack_htable_size
;
1111 old_vmalloced
= init_net
.ct
.hash_vmalloc
;
1112 old_hash
= init_net
.ct
.hash
;
1114 nf_conntrack_htable_size
= hashsize
;
1115 init_net
.ct
.hash_vmalloc
= vmalloced
;
1116 init_net
.ct
.hash
= hash
;
1117 nf_conntrack_hash_rnd
= rnd
;
1118 spin_unlock_bh(&nf_conntrack_lock
);
1120 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1123 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1125 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1126 &nf_conntrack_htable_size
, 0600);
1128 int nf_conntrack_init(struct net
*net
)
1133 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1134 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1135 if (!nf_conntrack_htable_size
) {
1136 nf_conntrack_htable_size
1137 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1138 / sizeof(struct hlist_head
));
1139 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1140 nf_conntrack_htable_size
= 16384;
1141 if (nf_conntrack_htable_size
< 32)
1142 nf_conntrack_htable_size
= 32;
1144 /* Use a max. factor of four by default to get the same max as
1145 * with the old struct list_heads. When a table size is given
1146 * we use the old value of 8 to avoid reducing the max.
1150 atomic_set(&net
->ct
.count
, 0);
1151 net
->ct
.hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
,
1152 &net
->ct
.hash_vmalloc
);
1153 if (!net
->ct
.hash
) {
1154 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1158 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1160 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1161 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1164 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1165 sizeof(struct nf_conn
),
1167 if (!nf_conntrack_cachep
) {
1168 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1172 ret
= nf_conntrack_proto_init();
1174 goto err_free_conntrack_slab
;
1176 ret
= nf_conntrack_expect_init();
1178 goto out_fini_proto
;
1180 ret
= nf_conntrack_helper_init();
1182 goto out_fini_expect
;
1184 ret
= nf_conntrack_acct_init();
1186 goto out_fini_helper
;
1188 /* For use by REJECT target */
1189 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1190 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1192 /* Set up fake conntrack:
1193 - to never be deleted, not in any hashes */
1194 #ifdef CONFIG_NET_NS
1195 nf_conntrack_untracked
.ct_net
= &init_net
;
1197 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1198 /* - and look it like as a confirmed connection */
1199 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1204 nf_conntrack_helper_fini();
1206 nf_conntrack_expect_fini();
1208 nf_conntrack_proto_fini();
1209 err_free_conntrack_slab
:
1210 kmem_cache_destroy(nf_conntrack_cachep
);
1212 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1213 nf_conntrack_htable_size
);