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>
42 #include <net/netfilter/nf_nat_core.h>
44 #define NF_CONNTRACK_VERSION "0.5.0"
46 int (*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 if (!test_bit(IPS_DYING_BIT
, &ct
->status
))
185 nf_conntrack_event(IPCT_DESTROY
, ct
);
186 set_bit(IPS_DYING_BIT
, &ct
->status
);
188 /* To make sure we don't get any weird locking issues here:
189 * destroy_conntrack() MUST NOT be called with a write lock
190 * to nf_conntrack_lock!!! -HW */
192 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
193 if (l4proto
&& l4proto
->destroy
)
194 l4proto
->destroy(ct
);
198 spin_lock_bh(&nf_conntrack_lock
);
199 /* Expectations will have been removed in clean_from_lists,
200 * except TFTP can create an expectation on the first packet,
201 * before connection is in the list, so we need to clean here,
203 nf_ct_remove_expectations(ct
);
205 /* We overload first tuple to link into unconfirmed list. */
206 if (!nf_ct_is_confirmed(ct
)) {
207 BUG_ON(hlist_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
));
208 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
211 NF_CT_STAT_INC(net
, delete);
212 spin_unlock_bh(&nf_conntrack_lock
);
215 nf_ct_put(ct
->master
);
217 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
218 nf_conntrack_free(ct
);
221 static void death_by_timeout(unsigned long ul_conntrack
)
223 struct nf_conn
*ct
= (void *)ul_conntrack
;
224 struct net
*net
= nf_ct_net(ct
);
225 struct nf_conn_help
*help
= nfct_help(ct
);
226 struct nf_conntrack_helper
*helper
;
230 helper
= rcu_dereference(help
->helper
);
231 if (helper
&& helper
->destroy
)
236 spin_lock_bh(&nf_conntrack_lock
);
237 /* Inside lock so preempt is disabled on module removal path.
238 * Otherwise we can get spurious warnings. */
239 NF_CT_STAT_INC(net
, delete_list
);
240 clean_from_lists(ct
);
241 spin_unlock_bh(&nf_conntrack_lock
);
245 struct nf_conntrack_tuple_hash
*
246 __nf_conntrack_find(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
248 struct nf_conntrack_tuple_hash
*h
;
249 struct hlist_node
*n
;
250 unsigned int hash
= hash_conntrack(tuple
);
252 /* Disable BHs the entire time since we normally need to disable them
253 * at least once for the stats anyway.
256 hlist_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnode
) {
257 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
258 NF_CT_STAT_INC(net
, found
);
262 NF_CT_STAT_INC(net
, searched
);
268 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
270 /* Find a connection corresponding to a tuple. */
271 struct nf_conntrack_tuple_hash
*
272 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
274 struct nf_conntrack_tuple_hash
*h
;
278 h
= __nf_conntrack_find(net
, tuple
);
280 ct
= nf_ct_tuplehash_to_ctrack(h
);
281 if (unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
288 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
290 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
292 unsigned int repl_hash
)
294 struct net
*net
= nf_ct_net(ct
);
296 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
297 &net
->ct
.hash
[hash
]);
298 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
,
299 &net
->ct
.hash
[repl_hash
]);
302 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
304 unsigned int hash
, repl_hash
;
306 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
307 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
309 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
311 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
313 /* Confirm a connection given skb; places it in hash table */
315 __nf_conntrack_confirm(struct sk_buff
*skb
)
317 unsigned int hash
, repl_hash
;
318 struct nf_conntrack_tuple_hash
*h
;
320 struct nf_conn_help
*help
;
321 struct hlist_node
*n
;
322 enum ip_conntrack_info ctinfo
;
325 ct
= nf_ct_get(skb
, &ctinfo
);
328 /* ipt_REJECT uses nf_conntrack_attach to attach related
329 ICMP/TCP RST packets in other direction. Actual packet
330 which created connection will be IP_CT_NEW or for an
331 expected connection, IP_CT_RELATED. */
332 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
335 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
336 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
338 /* We're not in hash table, and we refuse to set up related
339 connections for unconfirmed conns. But packet copies and
340 REJECT will give spurious warnings here. */
341 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
343 /* No external references means noone else could have
345 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
346 pr_debug("Confirming conntrack %p\n", ct
);
348 spin_lock_bh(&nf_conntrack_lock
);
350 /* See if there's one in the list already, including reverse:
351 NAT could have grabbed it without realizing, since we're
352 not in the hash. If there is, we lost race. */
353 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnode
)
354 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
357 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnode
)
358 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
362 /* Remove from unconfirmed list */
363 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
365 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
366 /* Timer relative to confirmation time, not original
367 setting time, otherwise we'd get timer wrap in
368 weird delay cases. */
369 ct
->timeout
.expires
+= jiffies
;
370 add_timer(&ct
->timeout
);
371 atomic_inc(&ct
->ct_general
.use
);
372 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
373 NF_CT_STAT_INC(net
, insert
);
374 spin_unlock_bh(&nf_conntrack_lock
);
375 help
= nfct_help(ct
);
376 if (help
&& help
->helper
)
377 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
378 #ifdef CONFIG_NF_NAT_NEEDED
379 if (test_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
) ||
380 test_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
))
381 nf_conntrack_event_cache(IPCT_NATINFO
, ct
);
383 nf_conntrack_event_cache(master_ct(ct
) ?
384 IPCT_RELATED
: IPCT_NEW
, ct
);
388 NF_CT_STAT_INC(net
, insert_failed
);
389 spin_unlock_bh(&nf_conntrack_lock
);
392 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
394 /* Returns true if a connection correspondings to the tuple (required
397 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
398 const struct nf_conn
*ignored_conntrack
)
400 struct net
*net
= nf_ct_net(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
, &net
->ct
.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(net
, found
);
413 rcu_read_unlock_bh();
416 NF_CT_STAT_INC(net
, 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(struct net
*net
, 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
, &net
->ct
.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(net
, early_drop
);
467 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
468 const struct nf_conntrack_tuple
*orig
,
469 const struct nf_conntrack_tuple
*repl
,
474 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
475 get_random_bytes(&nf_conntrack_hash_rnd
, 4);
476 nf_conntrack_hash_rnd_initted
= 1;
479 /* We don't want any race condition at early drop stage */
480 atomic_inc(&net
->ct
.count
);
482 if (nf_conntrack_max
&&
483 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
484 unsigned int hash
= hash_conntrack(orig
);
485 if (!early_drop(net
, hash
)) {
486 atomic_dec(&net
->ct
.count
);
489 "nf_conntrack: table full, dropping"
491 return ERR_PTR(-ENOMEM
);
495 ct
= kmem_cache_zalloc(nf_conntrack_cachep
, gfp
);
497 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
498 atomic_dec(&net
->ct
.count
);
499 return ERR_PTR(-ENOMEM
);
502 atomic_set(&ct
->ct_general
.use
, 1);
503 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
504 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
505 /* Don't set timer yet: wait for confirmation */
506 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
510 INIT_RCU_HEAD(&ct
->rcu
);
514 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
516 static void nf_conntrack_free_rcu(struct rcu_head
*head
)
518 struct nf_conn
*ct
= container_of(head
, struct nf_conn
, rcu
);
519 struct net
*net
= nf_ct_net(ct
);
522 kmem_cache_free(nf_conntrack_cachep
, ct
);
523 atomic_dec(&net
->ct
.count
);
526 void nf_conntrack_free(struct nf_conn
*ct
)
528 nf_ct_ext_destroy(ct
);
529 call_rcu(&ct
->rcu
, nf_conntrack_free_rcu
);
531 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
533 /* Allocate a new conntrack: we return -ENOMEM if classification
534 failed due to stress. Otherwise it really is unclassifiable. */
535 static struct nf_conntrack_tuple_hash
*
536 init_conntrack(struct net
*net
,
537 const struct nf_conntrack_tuple
*tuple
,
538 struct nf_conntrack_l3proto
*l3proto
,
539 struct nf_conntrack_l4proto
*l4proto
,
541 unsigned int dataoff
)
544 struct nf_conn_help
*help
;
545 struct nf_conntrack_tuple repl_tuple
;
546 struct nf_conntrack_expect
*exp
;
548 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
549 pr_debug("Can't invert tuple.\n");
553 ct
= nf_conntrack_alloc(net
, tuple
, &repl_tuple
, GFP_ATOMIC
);
555 pr_debug("Can't allocate conntrack.\n");
556 return (struct nf_conntrack_tuple_hash
*)ct
;
559 if (!l4proto
->new(ct
, skb
, dataoff
)) {
560 nf_conntrack_free(ct
);
561 pr_debug("init conntrack: can't track with proto module\n");
565 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
567 spin_lock_bh(&nf_conntrack_lock
);
568 exp
= nf_ct_find_expectation(net
, tuple
);
570 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
572 /* Welcome, Mr. Bond. We've been expecting you... */
573 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
574 ct
->master
= exp
->master
;
576 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
578 rcu_assign_pointer(help
->helper
, exp
->helper
);
581 #ifdef CONFIG_NF_CONNTRACK_MARK
582 ct
->mark
= exp
->master
->mark
;
584 #ifdef CONFIG_NF_CONNTRACK_SECMARK
585 ct
->secmark
= exp
->master
->secmark
;
587 nf_conntrack_get(&ct
->master
->ct_general
);
588 NF_CT_STAT_INC(net
, expect_new
);
590 __nf_ct_try_assign_helper(ct
, GFP_ATOMIC
);
591 NF_CT_STAT_INC(net
, new);
594 /* Overload tuple linked list to put us in unconfirmed list. */
595 hlist_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
596 &net
->ct
.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 net
*net
,
613 unsigned int dataoff
,
616 struct nf_conntrack_l3proto
*l3proto
,
617 struct nf_conntrack_l4proto
*l4proto
,
619 enum ip_conntrack_info
*ctinfo
)
621 struct nf_conntrack_tuple tuple
;
622 struct nf_conntrack_tuple_hash
*h
;
625 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
626 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
628 pr_debug("resolve_normal_ct: Can't get tuple\n");
632 /* look for tuple match */
633 h
= nf_conntrack_find_get(net
, &tuple
);
635 h
= init_conntrack(net
, &tuple
, l3proto
, l4proto
, skb
, dataoff
);
641 ct
= nf_ct_tuplehash_to_ctrack(h
);
643 /* It exists; we have (non-exclusive) reference. */
644 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
645 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
646 /* Please set reply bit if this packet OK */
649 /* Once we've had two way comms, always ESTABLISHED. */
650 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
651 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
652 *ctinfo
= IP_CT_ESTABLISHED
;
653 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
654 pr_debug("nf_conntrack_in: related packet for %p\n",
656 *ctinfo
= IP_CT_RELATED
;
658 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
663 skb
->nfct
= &ct
->ct_general
;
664 skb
->nfctinfo
= *ctinfo
;
669 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
673 enum ip_conntrack_info ctinfo
;
674 struct nf_conntrack_l3proto
*l3proto
;
675 struct nf_conntrack_l4proto
*l4proto
;
676 unsigned int dataoff
;
681 /* Previously seen (loopback or untracked)? Ignore. */
683 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
687 /* rcu_read_lock()ed by nf_hook_slow */
688 l3proto
= __nf_ct_l3proto_find(pf
);
689 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
690 &dataoff
, &protonum
);
692 pr_debug("not prepared to track yet or error occured\n");
693 NF_CT_STAT_INC_ATOMIC(net
, error
);
694 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
698 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
700 /* It may be an special packet, error, unclean...
701 * inverse of the return code tells to the netfilter
702 * core what to do with the packet. */
703 if (l4proto
->error
!= NULL
) {
704 ret
= l4proto
->error(net
, skb
, dataoff
, &ctinfo
, pf
, hooknum
);
706 NF_CT_STAT_INC_ATOMIC(net
, error
);
707 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
712 ct
= resolve_normal_ct(net
, skb
, dataoff
, pf
, protonum
,
713 l3proto
, l4proto
, &set_reply
, &ctinfo
);
715 /* Not valid part of a connection */
716 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
721 /* Too stressed to deal. */
722 NF_CT_STAT_INC_ATOMIC(net
, 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(net
, invalid
);
739 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
740 nf_conntrack_event_cache(IPCT_STATUS
, ct
);
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
);
768 /* Should be unconfirmed, so not in hash table yet */
769 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
771 pr_debug("Altering reply tuple of %p to ", ct
);
772 nf_ct_dump_tuple(newreply
);
774 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
775 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
779 __nf_ct_try_assign_helper(ct
, GFP_ATOMIC
);
782 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
784 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
785 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
786 enum ip_conntrack_info ctinfo
,
787 const struct sk_buff
*skb
,
788 unsigned long extra_jiffies
,
793 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
796 spin_lock_bh(&nf_conntrack_lock
);
798 /* Only update if this is not a fixed timeout */
799 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
802 /* If not in hash table, timer will not be active yet */
803 if (!nf_ct_is_confirmed(ct
)) {
804 ct
->timeout
.expires
= extra_jiffies
;
805 event
= IPCT_REFRESH
;
807 unsigned long newtime
= jiffies
+ extra_jiffies
;
809 /* Only update the timeout if the new timeout is at least
810 HZ jiffies from the old timeout. Need del_timer for race
811 avoidance (may already be dying). */
812 if (newtime
- ct
->timeout
.expires
>= HZ
813 && del_timer(&ct
->timeout
)) {
814 ct
->timeout
.expires
= newtime
;
815 add_timer(&ct
->timeout
);
816 event
= IPCT_REFRESH
;
822 struct nf_conn_counter
*acct
;
824 acct
= nf_conn_acct_find(ct
);
826 acct
[CTINFO2DIR(ctinfo
)].packets
++;
827 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
828 skb
->len
- skb_network_offset(skb
);
832 spin_unlock_bh(&nf_conntrack_lock
);
834 /* must be unlocked when calling event cache */
836 nf_conntrack_event_cache(event
, ct
);
838 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
840 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
841 enum ip_conntrack_info ctinfo
,
842 const struct sk_buff
*skb
,
846 struct nf_conn_counter
*acct
;
848 spin_lock_bh(&nf_conntrack_lock
);
849 acct
= nf_conn_acct_find(ct
);
851 acct
[CTINFO2DIR(ctinfo
)].packets
++;
852 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
853 skb
->len
- skb_network_offset(skb
);
855 spin_unlock_bh(&nf_conntrack_lock
);
858 if (del_timer(&ct
->timeout
)) {
859 ct
->timeout
.function((unsigned long)ct
);
864 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
866 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
868 #include <linux/netfilter/nfnetlink.h>
869 #include <linux/netfilter/nfnetlink_conntrack.h>
870 #include <linux/mutex.h>
872 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
873 * in ip_conntrack_core, since we don't want the protocols to autoload
874 * or depend on ctnetlink */
875 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
876 const struct nf_conntrack_tuple
*tuple
)
878 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
879 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
885 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
887 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
888 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
889 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
891 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
893 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
894 struct nf_conntrack_tuple
*t
)
896 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
899 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
900 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
904 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
907 /* Used by ipt_REJECT and ip6t_REJECT. */
908 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
911 enum ip_conntrack_info ctinfo
;
913 /* This ICMP is in reverse direction to the packet which caused it */
914 ct
= nf_ct_get(skb
, &ctinfo
);
915 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
916 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
918 ctinfo
= IP_CT_RELATED
;
920 /* Attach to new skbuff, and increment count */
921 nskb
->nfct
= &ct
->ct_general
;
922 nskb
->nfctinfo
= ctinfo
;
923 nf_conntrack_get(nskb
->nfct
);
926 /* Bring out ya dead! */
927 static struct nf_conn
*
928 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
929 void *data
, unsigned int *bucket
)
931 struct nf_conntrack_tuple_hash
*h
;
933 struct hlist_node
*n
;
935 spin_lock_bh(&nf_conntrack_lock
);
936 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
937 hlist_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnode
) {
938 ct
= nf_ct_tuplehash_to_ctrack(h
);
943 hlist_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnode
) {
944 ct
= nf_ct_tuplehash_to_ctrack(h
);
946 set_bit(IPS_DYING_BIT
, &ct
->status
);
948 spin_unlock_bh(&nf_conntrack_lock
);
951 atomic_inc(&ct
->ct_general
.use
);
952 spin_unlock_bh(&nf_conntrack_lock
);
956 void nf_ct_iterate_cleanup(struct net
*net
,
957 int (*iter
)(struct nf_conn
*i
, void *data
),
961 unsigned int bucket
= 0;
963 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
964 /* Time to push up daises... */
965 if (del_timer(&ct
->timeout
))
966 death_by_timeout((unsigned long)ct
);
967 /* ... else the timer will get him soon. */
972 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
974 struct __nf_ct_flush_report
{
979 static int kill_all(struct nf_conn
*i
, void *data
)
981 struct __nf_ct_flush_report
*fr
= (struct __nf_ct_flush_report
*)data
;
983 /* get_next_corpse sets the dying bit for us */
984 nf_conntrack_event_report(IPCT_DESTROY
,
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(struct net
*net
, u32 pid
, int report
)
1003 struct __nf_ct_flush_report fr
= {
1007 nf_ct_iterate_cleanup(net
, kill_all
, &fr
);
1009 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
1011 static void nf_conntrack_cleanup_init_net(void)
1013 nf_conntrack_helper_fini();
1014 nf_conntrack_proto_fini();
1015 kmem_cache_destroy(nf_conntrack_cachep
);
1018 static void nf_conntrack_cleanup_net(struct net
*net
)
1020 nf_ct_event_cache_flush(net
);
1021 nf_conntrack_ecache_fini(net
);
1023 nf_conntrack_flush(net
, 0, 0);
1024 if (atomic_read(&net
->ct
.count
) != 0) {
1026 goto i_see_dead_people
;
1028 /* wait until all references to nf_conntrack_untracked are dropped */
1029 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1032 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1033 nf_conntrack_htable_size
);
1034 nf_conntrack_acct_fini(net
);
1035 nf_conntrack_expect_fini(net
);
1036 free_percpu(net
->ct
.stat
);
1039 /* Mishearing the voices in his head, our hero wonders how he's
1040 supposed to kill the mall. */
1041 void nf_conntrack_cleanup(struct net
*net
)
1043 if (net_eq(net
, &init_net
))
1044 rcu_assign_pointer(ip_ct_attach
, NULL
);
1046 /* This makes sure all current packets have passed through
1047 netfilter framework. Roll on, two-stage module
1051 nf_conntrack_cleanup_net(net
);
1053 if (net_eq(net
, &init_net
)) {
1054 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1055 nf_conntrack_cleanup_init_net();
1059 struct hlist_head
*nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
)
1061 struct hlist_head
*hash
;
1062 unsigned int size
, i
;
1066 size
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_head
));
1067 hash
= (void*)__get_free_pages(GFP_KERNEL
|__GFP_NOWARN
,
1068 get_order(sizeof(struct hlist_head
)
1072 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1073 hash
= vmalloc(sizeof(struct hlist_head
) * size
);
1077 for (i
= 0; i
< size
; i
++)
1078 INIT_HLIST_HEAD(&hash
[i
]);
1082 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1084 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1086 int i
, bucket
, vmalloced
, old_vmalloced
;
1087 unsigned int hashsize
, old_size
;
1089 struct hlist_head
*hash
, *old_hash
;
1090 struct nf_conntrack_tuple_hash
*h
;
1092 /* On boot, we can set this without any fancy locking. */
1093 if (!nf_conntrack_htable_size
)
1094 return param_set_uint(val
, kp
);
1096 hashsize
= simple_strtoul(val
, NULL
, 0);
1100 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
);
1104 /* We have to rehahs for the new table anyway, so we also can
1105 * use a newrandom seed */
1106 get_random_bytes(&rnd
, 4);
1108 /* Lookups in the old hash might happen in parallel, which means we
1109 * might get false negatives during connection lookup. New connections
1110 * created because of a false negative won't make it into the hash
1111 * though since that required taking the lock.
1113 spin_lock_bh(&nf_conntrack_lock
);
1114 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1115 while (!hlist_empty(&init_net
.ct
.hash
[i
])) {
1116 h
= hlist_entry(init_net
.ct
.hash
[i
].first
,
1117 struct nf_conntrack_tuple_hash
, hnode
);
1118 hlist_del_rcu(&h
->hnode
);
1119 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1120 hlist_add_head(&h
->hnode
, &hash
[bucket
]);
1123 old_size
= nf_conntrack_htable_size
;
1124 old_vmalloced
= init_net
.ct
.hash_vmalloc
;
1125 old_hash
= init_net
.ct
.hash
;
1127 nf_conntrack_htable_size
= hashsize
;
1128 init_net
.ct
.hash_vmalloc
= vmalloced
;
1129 init_net
.ct
.hash
= hash
;
1130 nf_conntrack_hash_rnd
= rnd
;
1131 spin_unlock_bh(&nf_conntrack_lock
);
1133 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1136 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1138 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1139 &nf_conntrack_htable_size
, 0600);
1141 static int nf_conntrack_init_init_net(void)
1146 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1147 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1148 if (!nf_conntrack_htable_size
) {
1149 nf_conntrack_htable_size
1150 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1151 / sizeof(struct hlist_head
));
1152 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1153 nf_conntrack_htable_size
= 16384;
1154 if (nf_conntrack_htable_size
< 32)
1155 nf_conntrack_htable_size
= 32;
1157 /* Use a max. factor of four by default to get the same max as
1158 * with the old struct list_heads. When a table size is given
1159 * we use the old value of 8 to avoid reducing the max.
1163 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1165 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1166 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1169 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1170 sizeof(struct nf_conn
),
1172 if (!nf_conntrack_cachep
) {
1173 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1178 ret
= nf_conntrack_proto_init();
1182 ret
= nf_conntrack_helper_init();
1189 nf_conntrack_proto_fini();
1191 kmem_cache_destroy(nf_conntrack_cachep
);
1196 static int nf_conntrack_init_net(struct net
*net
)
1200 atomic_set(&net
->ct
.count
, 0);
1201 INIT_HLIST_HEAD(&net
->ct
.unconfirmed
);
1202 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1203 if (!net
->ct
.stat
) {
1207 ret
= nf_conntrack_ecache_init(net
);
1210 net
->ct
.hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
,
1211 &net
->ct
.hash_vmalloc
);
1212 if (!net
->ct
.hash
) {
1214 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1217 ret
= nf_conntrack_expect_init(net
);
1220 ret
= nf_conntrack_acct_init(net
);
1224 /* Set up fake conntrack:
1225 - to never be deleted, not in any hashes */
1226 #ifdef CONFIG_NET_NS
1227 nf_conntrack_untracked
.ct_net
= &init_net
;
1229 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1230 /* - and look it like as a confirmed connection */
1231 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1236 nf_conntrack_expect_fini(net
);
1238 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1239 nf_conntrack_htable_size
);
1241 nf_conntrack_ecache_fini(net
);
1243 free_percpu(net
->ct
.stat
);
1248 int nf_conntrack_init(struct net
*net
)
1252 if (net_eq(net
, &init_net
)) {
1253 ret
= nf_conntrack_init_init_net();
1257 ret
= nf_conntrack_init_net(net
);
1261 if (net_eq(net
, &init_net
)) {
1262 /* For use by REJECT target */
1263 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1264 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1269 if (net_eq(net
, &init_net
))
1270 nf_conntrack_cleanup_init_net();