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 /* nf_conntrack_standalone needs this */
48 atomic_t nf_conntrack_count
= ATOMIC_INIT(0);
49 EXPORT_SYMBOL_GPL(nf_conntrack_count
);
51 unsigned int nf_conntrack_htable_size __read_mostly
;
52 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
54 int nf_conntrack_max __read_mostly
;
55 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
57 struct hlist_head
*nf_conntrack_hash __read_mostly
;
58 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
60 struct nf_conn nf_conntrack_untracked __read_mostly
;
61 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
63 unsigned int nf_ct_log_invalid __read_mostly
;
64 HLIST_HEAD(unconfirmed
);
65 static int nf_conntrack_vmalloc __read_mostly
;
66 static struct kmem_cache
*nf_conntrack_cachep __read_mostly
;
68 DEFINE_PER_CPU(struct ip_conntrack_stat
, nf_conntrack_stat
);
69 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat
);
71 static int nf_conntrack_hash_rnd_initted
;
72 static unsigned int nf_conntrack_hash_rnd
;
74 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
75 unsigned int size
, unsigned int rnd
)
80 /* The direction must be ignored, so we hash everything up to the
81 * destination ports (which is a multiple of 4) and treat the last
82 * three bytes manually.
84 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
85 h
= jhash2((u32
*)tuple
, n
,
86 rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
87 tuple
->dst
.protonum
));
89 return ((u64
)h
* size
) >> 32;
92 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
94 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
95 nf_conntrack_hash_rnd
);
99 nf_ct_get_tuple(const struct sk_buff
*skb
,
101 unsigned int dataoff
,
104 struct nf_conntrack_tuple
*tuple
,
105 const struct nf_conntrack_l3proto
*l3proto
,
106 const struct nf_conntrack_l4proto
*l4proto
)
108 memset(tuple
, 0, sizeof(*tuple
));
110 tuple
->src
.l3num
= l3num
;
111 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
114 tuple
->dst
.protonum
= protonum
;
115 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
117 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
119 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
121 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
122 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
124 struct nf_conntrack_l3proto
*l3proto
;
125 struct nf_conntrack_l4proto
*l4proto
;
126 unsigned int protoff
;
132 l3proto
= __nf_ct_l3proto_find(l3num
);
133 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
134 if (ret
!= NF_ACCEPT
) {
139 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
141 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
147 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
150 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
151 const struct nf_conntrack_tuple
*orig
,
152 const struct nf_conntrack_l3proto
*l3proto
,
153 const struct nf_conntrack_l4proto
*l4proto
)
155 memset(inverse
, 0, sizeof(*inverse
));
157 inverse
->src
.l3num
= orig
->src
.l3num
;
158 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
161 inverse
->dst
.dir
= !orig
->dst
.dir
;
163 inverse
->dst
.protonum
= orig
->dst
.protonum
;
164 return l4proto
->invert_tuple(inverse
, orig
);
166 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
169 clean_from_lists(struct nf_conn
*ct
)
171 pr_debug("clean_from_lists(%p)\n", ct
);
172 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
173 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
);
175 /* Destroy all pending expectations */
176 nf_ct_remove_expectations(ct
);
180 destroy_conntrack(struct nf_conntrack
*nfct
)
182 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
183 struct nf_conntrack_l4proto
*l4proto
;
185 pr_debug("destroy_conntrack(%p)\n", ct
);
186 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
187 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
189 nf_conntrack_event(IPCT_DESTROY
, ct
);
190 set_bit(IPS_DYING_BIT
, &ct
->status
);
192 /* To make sure we don't get any weird locking issues here:
193 * destroy_conntrack() MUST NOT be called with a write lock
194 * to nf_conntrack_lock!!! -HW */
196 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
197 if (l4proto
&& l4proto
->destroy
)
198 l4proto
->destroy(ct
);
202 spin_lock_bh(&nf_conntrack_lock
);
203 /* Expectations will have been removed in clean_from_lists,
204 * except TFTP can create an expectation on the first packet,
205 * before connection is in the list, so we need to clean here,
207 nf_ct_remove_expectations(ct
);
209 /* We overload first tuple to link into unconfirmed list. */
210 if (!nf_ct_is_confirmed(ct
)) {
211 BUG_ON(hlist_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
));
212 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
215 NF_CT_STAT_INC(delete);
216 spin_unlock_bh(&nf_conntrack_lock
);
219 nf_ct_put(ct
->master
);
221 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
222 nf_conntrack_free(ct
);
225 static void death_by_timeout(unsigned long ul_conntrack
)
227 struct nf_conn
*ct
= (void *)ul_conntrack
;
228 struct nf_conn_help
*help
= nfct_help(ct
);
229 struct nf_conntrack_helper
*helper
;
233 helper
= rcu_dereference(help
->helper
);
234 if (helper
&& helper
->destroy
)
239 spin_lock_bh(&nf_conntrack_lock
);
240 /* Inside lock so preempt is disabled on module removal path.
241 * Otherwise we can get spurious warnings. */
242 NF_CT_STAT_INC(delete_list
);
243 clean_from_lists(ct
);
244 spin_unlock_bh(&nf_conntrack_lock
);
248 struct nf_conntrack_tuple_hash
*
249 __nf_conntrack_find(const struct nf_conntrack_tuple
*tuple
)
251 struct nf_conntrack_tuple_hash
*h
;
252 struct hlist_node
*n
;
253 unsigned int hash
= hash_conntrack(tuple
);
255 /* Disable BHs the entire time since we normally need to disable them
256 * at least once for the stats anyway.
259 hlist_for_each_entry_rcu(h
, n
, &nf_conntrack_hash
[hash
], hnode
) {
260 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
261 NF_CT_STAT_INC(found
);
265 NF_CT_STAT_INC(searched
);
271 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
273 /* Find a connection corresponding to a tuple. */
274 struct nf_conntrack_tuple_hash
*
275 nf_conntrack_find_get(const struct nf_conntrack_tuple
*tuple
)
277 struct nf_conntrack_tuple_hash
*h
;
281 h
= __nf_conntrack_find(tuple
);
283 ct
= nf_ct_tuplehash_to_ctrack(h
);
284 if (unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
291 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
293 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
295 unsigned int repl_hash
)
297 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
298 &nf_conntrack_hash
[hash
]);
299 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
,
300 &nf_conntrack_hash
[repl_hash
]);
303 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
305 unsigned int hash
, repl_hash
;
307 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
308 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
310 spin_lock_bh(&nf_conntrack_lock
);
311 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
312 spin_unlock_bh(&nf_conntrack_lock
);
314 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
316 /* Confirm a connection given skb; places it in hash table */
318 __nf_conntrack_confirm(struct sk_buff
*skb
)
320 unsigned int hash
, repl_hash
;
321 struct nf_conntrack_tuple_hash
*h
;
323 struct nf_conn_help
*help
;
324 struct hlist_node
*n
;
325 enum ip_conntrack_info ctinfo
;
327 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
, &nf_conntrack_hash
[hash
], hnode
)
355 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
358 hlist_for_each_entry(h
, n
, &nf_conntrack_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(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
, skb
);
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
, skb
);
384 nf_conntrack_event_cache(master_ct(ct
) ?
385 IPCT_RELATED
: IPCT_NEW
, skb
);
389 NF_CT_STAT_INC(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 nf_conntrack_tuple_hash
*h
;
402 struct hlist_node
*n
;
403 unsigned int hash
= hash_conntrack(tuple
);
405 /* Disable BHs the entire time since we need to disable them at
406 * least once for the stats anyway.
409 hlist_for_each_entry_rcu(h
, n
, &nf_conntrack_hash
[hash
], hnode
) {
410 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
411 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
412 NF_CT_STAT_INC(found
);
413 rcu_read_unlock_bh();
416 NF_CT_STAT_INC(searched
);
418 rcu_read_unlock_bh();
422 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
424 #define NF_CT_EVICTION_RANGE 8
426 /* There's a small race here where we may free a just-assured
427 connection. Too bad: we're in trouble anyway. */
428 static noinline
int early_drop(unsigned int hash
)
430 /* Use oldest entry, which is roughly LRU */
431 struct nf_conntrack_tuple_hash
*h
;
432 struct nf_conn
*ct
= NULL
, *tmp
;
433 struct hlist_node
*n
;
434 unsigned int i
, cnt
= 0;
438 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
439 hlist_for_each_entry_rcu(h
, n
, &nf_conntrack_hash
[hash
],
441 tmp
= nf_ct_tuplehash_to_ctrack(h
);
442 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
447 if (ct
&& unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
449 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
451 hash
= (hash
+ 1) % nf_conntrack_htable_size
;
458 if (del_timer(&ct
->timeout
)) {
459 death_by_timeout((unsigned long)ct
);
461 NF_CT_STAT_INC_ATOMIC(early_drop
);
467 struct nf_conn
*nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
468 const struct nf_conntrack_tuple
*repl
,
471 struct nf_conn
*ct
= NULL
;
473 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
474 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
475 nf_conntrack_hash_rnd_initted
= 1;
478 /* We don't want any race condition at early drop stage */
479 atomic_inc(&nf_conntrack_count
);
481 if (nf_conntrack_max
&&
482 unlikely(atomic_read(&nf_conntrack_count
) > nf_conntrack_max
)) {
483 unsigned int hash
= hash_conntrack(orig
);
484 if (!early_drop(hash
)) {
485 atomic_dec(&nf_conntrack_count
);
488 "nf_conntrack: table full, dropping"
490 return ERR_PTR(-ENOMEM
);
494 ct
= kmem_cache_zalloc(nf_conntrack_cachep
, gfp
);
496 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
497 atomic_dec(&nf_conntrack_count
);
498 return ERR_PTR(-ENOMEM
);
501 atomic_set(&ct
->ct_general
.use
, 1);
502 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
503 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
504 /* Don't set timer yet: wait for confirmation */
505 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
506 INIT_RCU_HEAD(&ct
->rcu
);
510 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
512 static void nf_conntrack_free_rcu(struct rcu_head
*head
)
514 struct nf_conn
*ct
= container_of(head
, struct nf_conn
, rcu
);
517 kmem_cache_free(nf_conntrack_cachep
, ct
);
518 atomic_dec(&nf_conntrack_count
);
521 void nf_conntrack_free(struct nf_conn
*ct
)
523 nf_ct_ext_destroy(ct
);
524 call_rcu(&ct
->rcu
, nf_conntrack_free_rcu
);
526 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
528 /* Allocate a new conntrack: we return -ENOMEM if classification
529 failed due to stress. Otherwise it really is unclassifiable. */
530 static struct nf_conntrack_tuple_hash
*
531 init_conntrack(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(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(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(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
);
595 /* Overload tuple linked list to put us in unconfirmed list. */
596 hlist_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
, &unconfirmed
);
598 spin_unlock_bh(&nf_conntrack_lock
);
602 exp
->expectfn(ct
, exp
);
603 nf_ct_expect_put(exp
);
606 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
609 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
610 static inline struct nf_conn
*
611 resolve_normal_ct(struct sk_buff
*skb
,
612 unsigned int dataoff
,
615 struct nf_conntrack_l3proto
*l3proto
,
616 struct nf_conntrack_l4proto
*l4proto
,
618 enum ip_conntrack_info
*ctinfo
)
620 struct nf_conntrack_tuple tuple
;
621 struct nf_conntrack_tuple_hash
*h
;
624 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
625 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
627 pr_debug("resolve_normal_ct: Can't get tuple\n");
631 /* look for tuple match */
632 h
= nf_conntrack_find_get(&tuple
);
634 h
= init_conntrack(&tuple
, l3proto
, l4proto
, skb
, dataoff
);
640 ct
= nf_ct_tuplehash_to_ctrack(h
);
642 /* It exists; we have (non-exclusive) reference. */
643 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
644 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
645 /* Please set reply bit if this packet OK */
648 /* Once we've had two way comms, always ESTABLISHED. */
649 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
650 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
651 *ctinfo
= IP_CT_ESTABLISHED
;
652 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
653 pr_debug("nf_conntrack_in: related packet for %p\n",
655 *ctinfo
= IP_CT_RELATED
;
657 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
662 skb
->nfct
= &ct
->ct_general
;
663 skb
->nfctinfo
= *ctinfo
;
668 nf_conntrack_in(int pf
, unsigned int hooknum
, struct sk_buff
*skb
)
671 enum ip_conntrack_info ctinfo
;
672 struct nf_conntrack_l3proto
*l3proto
;
673 struct nf_conntrack_l4proto
*l4proto
;
674 unsigned int dataoff
;
679 /* Previously seen (loopback or untracked)? Ignore. */
681 NF_CT_STAT_INC_ATOMIC(ignore
);
685 /* rcu_read_lock()ed by nf_hook_slow */
686 l3proto
= __nf_ct_l3proto_find((u_int16_t
)pf
);
687 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
688 &dataoff
, &protonum
);
690 pr_debug("not prepared to track yet or error occured\n");
691 NF_CT_STAT_INC_ATOMIC(error
);
692 NF_CT_STAT_INC_ATOMIC(invalid
);
696 l4proto
= __nf_ct_l4proto_find((u_int16_t
)pf
, protonum
);
698 /* It may be an special packet, error, unclean...
699 * inverse of the return code tells to the netfilter
700 * core what to do with the packet. */
701 if (l4proto
->error
!= NULL
&&
702 (ret
= l4proto
->error(skb
, dataoff
, &ctinfo
, pf
, hooknum
)) <= 0) {
703 NF_CT_STAT_INC_ATOMIC(error
);
704 NF_CT_STAT_INC_ATOMIC(invalid
);
708 ct
= resolve_normal_ct(skb
, dataoff
, pf
, protonum
, l3proto
, l4proto
,
709 &set_reply
, &ctinfo
);
711 /* Not valid part of a connection */
712 NF_CT_STAT_INC_ATOMIC(invalid
);
717 /* Too stressed to deal. */
718 NF_CT_STAT_INC_ATOMIC(drop
);
722 NF_CT_ASSERT(skb
->nfct
);
724 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
726 /* Invalid: inverse of the return code tells
727 * the netfilter core what to do */
728 pr_debug("nf_conntrack_in: Can't track with proto module\n");
729 nf_conntrack_put(skb
->nfct
);
731 NF_CT_STAT_INC_ATOMIC(invalid
);
735 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
736 nf_conntrack_event_cache(IPCT_STATUS
, skb
);
740 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
742 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
743 const struct nf_conntrack_tuple
*orig
)
748 ret
= nf_ct_invert_tuple(inverse
, orig
,
749 __nf_ct_l3proto_find(orig
->src
.l3num
),
750 __nf_ct_l4proto_find(orig
->src
.l3num
,
751 orig
->dst
.protonum
));
755 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
757 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
758 implicitly racy: see __nf_conntrack_confirm */
759 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
760 const struct nf_conntrack_tuple
*newreply
)
762 struct nf_conn_help
*help
= nfct_help(ct
);
763 struct nf_conntrack_helper
*helper
;
765 /* Should be unconfirmed, so not in hash table yet */
766 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
768 pr_debug("Altering reply tuple of %p to ", ct
);
769 nf_ct_dump_tuple(newreply
);
771 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
772 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
776 helper
= __nf_ct_helper_find(newreply
);
777 if (helper
== NULL
) {
779 rcu_assign_pointer(help
->helper
, NULL
);
784 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
788 memset(&help
->help
, 0, sizeof(help
->help
));
791 rcu_assign_pointer(help
->helper
, helper
);
795 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
797 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
798 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
799 enum ip_conntrack_info ctinfo
,
800 const struct sk_buff
*skb
,
801 unsigned long extra_jiffies
,
806 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
809 spin_lock_bh(&nf_conntrack_lock
);
811 /* Only update if this is not a fixed timeout */
812 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
815 /* If not in hash table, timer will not be active yet */
816 if (!nf_ct_is_confirmed(ct
)) {
817 ct
->timeout
.expires
= extra_jiffies
;
818 event
= IPCT_REFRESH
;
820 unsigned long newtime
= jiffies
+ extra_jiffies
;
822 /* Only update the timeout if the new timeout is at least
823 HZ jiffies from the old timeout. Need del_timer for race
824 avoidance (may already be dying). */
825 if (newtime
- ct
->timeout
.expires
>= HZ
826 && del_timer(&ct
->timeout
)) {
827 ct
->timeout
.expires
= newtime
;
828 add_timer(&ct
->timeout
);
829 event
= IPCT_REFRESH
;
835 struct nf_conn_counter
*acct
;
837 acct
= nf_conn_acct_find(ct
);
839 acct
[CTINFO2DIR(ctinfo
)].packets
++;
840 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
841 skb
->len
- skb_network_offset(skb
);
845 spin_unlock_bh(&nf_conntrack_lock
);
847 /* must be unlocked when calling event cache */
849 nf_conntrack_event_cache(event
, skb
);
851 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
853 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
854 enum ip_conntrack_info ctinfo
,
855 const struct sk_buff
*skb
,
859 struct nf_conn_counter
*acct
;
861 spin_lock_bh(&nf_conntrack_lock
);
862 acct
= nf_conn_acct_find(ct
);
864 acct
[CTINFO2DIR(ctinfo
)].packets
++;
865 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
866 skb
->len
- skb_network_offset(skb
);
868 spin_unlock_bh(&nf_conntrack_lock
);
871 if (del_timer(&ct
->timeout
)) {
872 ct
->timeout
.function((unsigned long)ct
);
877 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
879 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
881 #include <linux/netfilter/nfnetlink.h>
882 #include <linux/netfilter/nfnetlink_conntrack.h>
883 #include <linux/mutex.h>
885 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
886 * in ip_conntrack_core, since we don't want the protocols to autoload
887 * or depend on ctnetlink */
888 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
889 const struct nf_conntrack_tuple
*tuple
)
891 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
892 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
898 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
900 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
901 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
902 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
904 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
906 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
907 struct nf_conntrack_tuple
*t
)
909 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
912 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
913 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
917 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
920 /* Used by ipt_REJECT and ip6t_REJECT. */
921 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
924 enum ip_conntrack_info ctinfo
;
926 /* This ICMP is in reverse direction to the packet which caused it */
927 ct
= nf_ct_get(skb
, &ctinfo
);
928 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
929 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
931 ctinfo
= IP_CT_RELATED
;
933 /* Attach to new skbuff, and increment count */
934 nskb
->nfct
= &ct
->ct_general
;
935 nskb
->nfctinfo
= ctinfo
;
936 nf_conntrack_get(nskb
->nfct
);
939 /* Bring out ya dead! */
940 static struct nf_conn
*
941 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
942 void *data
, unsigned int *bucket
)
944 struct nf_conntrack_tuple_hash
*h
;
946 struct hlist_node
*n
;
948 spin_lock_bh(&nf_conntrack_lock
);
949 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
950 hlist_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnode
) {
951 ct
= nf_ct_tuplehash_to_ctrack(h
);
956 hlist_for_each_entry(h
, n
, &unconfirmed
, hnode
) {
957 ct
= nf_ct_tuplehash_to_ctrack(h
);
959 set_bit(IPS_DYING_BIT
, &ct
->status
);
961 spin_unlock_bh(&nf_conntrack_lock
);
964 atomic_inc(&ct
->ct_general
.use
);
965 spin_unlock_bh(&nf_conntrack_lock
);
970 nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
973 unsigned int bucket
= 0;
975 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
976 /* Time to push up daises... */
977 if (del_timer(&ct
->timeout
))
978 death_by_timeout((unsigned long)ct
);
979 /* ... else the timer will get him soon. */
984 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
986 static int kill_all(struct nf_conn
*i
, void *data
)
991 void nf_ct_free_hashtable(struct hlist_head
*hash
, int vmalloced
, unsigned int size
)
996 free_pages((unsigned long)hash
,
997 get_order(sizeof(struct hlist_head
) * size
));
999 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1001 void nf_conntrack_flush(void)
1003 nf_ct_iterate_cleanup(kill_all
, NULL
);
1005 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
1007 /* Mishearing the voices in his head, our hero wonders how he's
1008 supposed to kill the mall. */
1009 void nf_conntrack_cleanup(void)
1011 rcu_assign_pointer(ip_ct_attach
, NULL
);
1013 /* This makes sure all current packets have passed through
1014 netfilter framework. Roll on, two-stage module
1018 nf_ct_event_cache_flush();
1020 nf_conntrack_flush();
1021 if (atomic_read(&nf_conntrack_count
) != 0) {
1023 goto i_see_dead_people
;
1025 /* wait until all references to nf_conntrack_untracked are dropped */
1026 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1029 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1031 kmem_cache_destroy(nf_conntrack_cachep
);
1032 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1033 nf_conntrack_htable_size
);
1035 nf_conntrack_proto_fini();
1036 nf_conntrack_helper_fini();
1037 nf_conntrack_expect_fini();
1038 nf_conntrack_acct_fini();
1041 struct hlist_head
*nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
)
1043 struct hlist_head
*hash
;
1044 unsigned int size
, i
;
1048 size
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_head
));
1049 hash
= (void*)__get_free_pages(GFP_KERNEL
|__GFP_NOWARN
,
1050 get_order(sizeof(struct hlist_head
)
1054 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1055 hash
= vmalloc(sizeof(struct hlist_head
) * size
);
1059 for (i
= 0; i
< size
; i
++)
1060 INIT_HLIST_HEAD(&hash
[i
]);
1064 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1066 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1068 int i
, bucket
, vmalloced
, old_vmalloced
;
1069 unsigned int hashsize
, old_size
;
1071 struct hlist_head
*hash
, *old_hash
;
1072 struct nf_conntrack_tuple_hash
*h
;
1074 /* On boot, we can set this without any fancy locking. */
1075 if (!nf_conntrack_htable_size
)
1076 return param_set_uint(val
, kp
);
1078 hashsize
= simple_strtoul(val
, NULL
, 0);
1082 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
);
1086 /* We have to rehahs for the new table anyway, so we also can
1087 * use a newrandom seed */
1088 get_random_bytes(&rnd
, 4);
1090 /* Lookups in the old hash might happen in parallel, which means we
1091 * might get false negatives during connection lookup. New connections
1092 * created because of a false negative won't make it into the hash
1093 * though since that required taking the lock.
1095 spin_lock_bh(&nf_conntrack_lock
);
1096 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1097 while (!hlist_empty(&nf_conntrack_hash
[i
])) {
1098 h
= hlist_entry(nf_conntrack_hash
[i
].first
,
1099 struct nf_conntrack_tuple_hash
, hnode
);
1100 hlist_del_rcu(&h
->hnode
);
1101 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1102 hlist_add_head(&h
->hnode
, &hash
[bucket
]);
1105 old_size
= nf_conntrack_htable_size
;
1106 old_vmalloced
= nf_conntrack_vmalloc
;
1107 old_hash
= nf_conntrack_hash
;
1109 nf_conntrack_htable_size
= hashsize
;
1110 nf_conntrack_vmalloc
= vmalloced
;
1111 nf_conntrack_hash
= hash
;
1112 nf_conntrack_hash_rnd
= rnd
;
1113 spin_unlock_bh(&nf_conntrack_lock
);
1115 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1118 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1120 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1121 &nf_conntrack_htable_size
, 0600);
1123 int __init
nf_conntrack_init(void)
1128 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1129 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1130 if (!nf_conntrack_htable_size
) {
1131 nf_conntrack_htable_size
1132 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1133 / sizeof(struct hlist_head
));
1134 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1135 nf_conntrack_htable_size
= 16384;
1136 if (nf_conntrack_htable_size
< 32)
1137 nf_conntrack_htable_size
= 32;
1139 /* Use a max. factor of four by default to get the same max as
1140 * with the old struct list_heads. When a table size is given
1141 * we use the old value of 8 to avoid reducing the max.
1145 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
,
1146 &nf_conntrack_vmalloc
);
1147 if (!nf_conntrack_hash
) {
1148 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1152 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1154 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1155 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1158 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1159 sizeof(struct nf_conn
),
1161 if (!nf_conntrack_cachep
) {
1162 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1166 ret
= nf_conntrack_proto_init();
1168 goto err_free_conntrack_slab
;
1170 ret
= nf_conntrack_expect_init();
1172 goto out_fini_proto
;
1174 ret
= nf_conntrack_helper_init();
1176 goto out_fini_expect
;
1178 ret
= nf_conntrack_acct_init();
1180 goto out_fini_helper
;
1182 /* For use by REJECT target */
1183 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1184 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1186 /* Set up fake conntrack:
1187 - to never be deleted, not in any hashes */
1188 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1189 /* - and look it like as a confirmed connection */
1190 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1195 nf_conntrack_helper_fini();
1197 nf_conntrack_expect_fini();
1199 nf_conntrack_proto_fini();
1200 err_free_conntrack_slab
:
1201 kmem_cache_destroy(nf_conntrack_cachep
);
1203 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1204 nf_conntrack_htable_size
);