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>
41 #define NF_CONNTRACK_VERSION "0.5.0"
43 DEFINE_SPINLOCK(nf_conntrack_lock
);
44 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
46 /* nf_conntrack_standalone needs this */
47 atomic_t nf_conntrack_count
= ATOMIC_INIT(0);
48 EXPORT_SYMBOL_GPL(nf_conntrack_count
);
50 unsigned int nf_conntrack_htable_size __read_mostly
;
51 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
53 int nf_conntrack_max __read_mostly
;
54 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
56 struct hlist_head
*nf_conntrack_hash __read_mostly
;
57 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
59 struct nf_conn nf_conntrack_untracked __read_mostly
;
60 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
62 unsigned int nf_ct_log_invalid __read_mostly
;
63 HLIST_HEAD(unconfirmed
);
64 static int nf_conntrack_vmalloc __read_mostly
;
65 static struct kmem_cache
*nf_conntrack_cachep __read_mostly
;
67 DEFINE_PER_CPU(struct ip_conntrack_stat
, nf_conntrack_stat
);
68 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat
);
70 static int nf_conntrack_hash_rnd_initted
;
71 static unsigned int nf_conntrack_hash_rnd
;
73 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
74 unsigned int size
, unsigned int rnd
)
79 /* The direction must be ignored, so we hash everything up to the
80 * destination ports (which is a multiple of 4) and treat the last
81 * three bytes manually.
83 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
84 h
= jhash2((u32
*)tuple
, n
,
85 rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
86 tuple
->dst
.protonum
));
88 return ((u64
)h
* size
) >> 32;
91 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
93 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
94 nf_conntrack_hash_rnd
);
98 nf_ct_get_tuple(const struct sk_buff
*skb
,
100 unsigned int dataoff
,
103 struct nf_conntrack_tuple
*tuple
,
104 const struct nf_conntrack_l3proto
*l3proto
,
105 const struct nf_conntrack_l4proto
*l4proto
)
107 memset(tuple
, 0, sizeof(*tuple
));
109 tuple
->src
.l3num
= l3num
;
110 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
113 tuple
->dst
.protonum
= protonum
;
114 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
116 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
118 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
120 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
121 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
123 struct nf_conntrack_l3proto
*l3proto
;
124 struct nf_conntrack_l4proto
*l4proto
;
125 unsigned int protoff
;
131 l3proto
= __nf_ct_l3proto_find(l3num
);
132 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
133 if (ret
!= NF_ACCEPT
) {
138 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
140 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
146 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
149 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
150 const struct nf_conntrack_tuple
*orig
,
151 const struct nf_conntrack_l3proto
*l3proto
,
152 const struct nf_conntrack_l4proto
*l4proto
)
154 memset(inverse
, 0, sizeof(*inverse
));
156 inverse
->src
.l3num
= orig
->src
.l3num
;
157 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
160 inverse
->dst
.dir
= !orig
->dst
.dir
;
162 inverse
->dst
.protonum
= orig
->dst
.protonum
;
163 return l4proto
->invert_tuple(inverse
, orig
);
165 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
168 clean_from_lists(struct nf_conn
*ct
)
170 pr_debug("clean_from_lists(%p)\n", ct
);
171 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
172 hlist_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
);
174 /* Destroy all pending expectations */
175 nf_ct_remove_expectations(ct
);
179 destroy_conntrack(struct nf_conntrack
*nfct
)
181 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
182 struct nf_conntrack_l4proto
*l4proto
;
184 pr_debug("destroy_conntrack(%p)\n", ct
);
185 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
186 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
188 nf_conntrack_event(IPCT_DESTROY
, ct
);
189 set_bit(IPS_DYING_BIT
, &ct
->status
);
191 /* To make sure we don't get any weird locking issues here:
192 * destroy_conntrack() MUST NOT be called with a write lock
193 * to nf_conntrack_lock!!! -HW */
195 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
196 if (l4proto
&& l4proto
->destroy
)
197 l4proto
->destroy(ct
);
201 spin_lock_bh(&nf_conntrack_lock
);
202 /* Expectations will have been removed in clean_from_lists,
203 * except TFTP can create an expectation on the first packet,
204 * before connection is in the list, so we need to clean here,
206 nf_ct_remove_expectations(ct
);
208 /* We overload first tuple to link into unconfirmed list. */
209 if (!nf_ct_is_confirmed(ct
)) {
210 BUG_ON(hlist_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
));
211 hlist_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
);
214 NF_CT_STAT_INC(delete);
215 spin_unlock_bh(&nf_conntrack_lock
);
218 nf_ct_put(ct
->master
);
220 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
221 nf_conntrack_free(ct
);
224 static void death_by_timeout(unsigned long ul_conntrack
)
226 struct nf_conn
*ct
= (void *)ul_conntrack
;
227 struct nf_conn_help
*help
= nfct_help(ct
);
228 struct nf_conntrack_helper
*helper
;
232 helper
= rcu_dereference(help
->helper
);
233 if (helper
&& helper
->destroy
)
238 spin_lock_bh(&nf_conntrack_lock
);
239 /* Inside lock so preempt is disabled on module removal path.
240 * Otherwise we can get spurious warnings. */
241 NF_CT_STAT_INC(delete_list
);
242 clean_from_lists(ct
);
243 spin_unlock_bh(&nf_conntrack_lock
);
247 struct nf_conntrack_tuple_hash
*
248 __nf_conntrack_find(const struct nf_conntrack_tuple
*tuple
)
250 struct nf_conntrack_tuple_hash
*h
;
251 struct hlist_node
*n
;
252 unsigned int hash
= hash_conntrack(tuple
);
254 /* Disable BHs the entire time since we normally need to disable them
255 * at least once for the stats anyway.
258 hlist_for_each_entry_rcu(h
, n
, &nf_conntrack_hash
[hash
], hnode
) {
259 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
260 NF_CT_STAT_INC(found
);
264 NF_CT_STAT_INC(searched
);
270 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
272 /* Find a connection corresponding to a tuple. */
273 struct nf_conntrack_tuple_hash
*
274 nf_conntrack_find_get(const struct nf_conntrack_tuple
*tuple
)
276 struct nf_conntrack_tuple_hash
*h
;
280 h
= __nf_conntrack_find(tuple
);
282 ct
= nf_ct_tuplehash_to_ctrack(h
);
283 if (unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
290 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
292 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
294 unsigned int repl_hash
)
296 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
,
297 &nf_conntrack_hash
[hash
]);
298 hlist_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnode
,
299 &nf_conntrack_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 spin_lock_bh(&nf_conntrack_lock
);
310 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
311 spin_unlock_bh(&nf_conntrack_lock
);
313 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
315 /* Confirm a connection given skb; places it in hash table */
317 __nf_conntrack_confirm(struct sk_buff
*skb
)
319 unsigned int hash
, repl_hash
;
320 struct nf_conntrack_tuple_hash
*h
;
322 struct nf_conn_help
*help
;
323 struct hlist_node
*n
;
324 enum ip_conntrack_info ctinfo
;
326 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
, &nf_conntrack_hash
[hash
], hnode
)
354 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
357 hlist_for_each_entry(h
, n
, &nf_conntrack_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(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
, skb
);
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
, skb
);
383 nf_conntrack_event_cache(master_ct(ct
) ?
384 IPCT_RELATED
: IPCT_NEW
, skb
);
388 NF_CT_STAT_INC(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 nf_conntrack_tuple_hash
*h
;
401 struct hlist_node
*n
;
402 unsigned int hash
= hash_conntrack(tuple
);
404 /* Disable BHs the entire time since we need to disable them at
405 * least once for the stats anyway.
408 hlist_for_each_entry_rcu(h
, n
, &nf_conntrack_hash
[hash
], hnode
) {
409 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
410 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
411 NF_CT_STAT_INC(found
);
412 rcu_read_unlock_bh();
415 NF_CT_STAT_INC(searched
);
417 rcu_read_unlock_bh();
421 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
423 #define NF_CT_EVICTION_RANGE 8
425 /* There's a small race here where we may free a just-assured
426 connection. Too bad: we're in trouble anyway. */
427 static noinline
int early_drop(unsigned int hash
)
429 /* Use oldest entry, which is roughly LRU */
430 struct nf_conntrack_tuple_hash
*h
;
431 struct nf_conn
*ct
= NULL
, *tmp
;
432 struct hlist_node
*n
;
433 unsigned int i
, cnt
= 0;
437 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
438 hlist_for_each_entry_rcu(h
, n
, &nf_conntrack_hash
[hash
],
440 tmp
= nf_ct_tuplehash_to_ctrack(h
);
441 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
446 if (ct
&& unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
448 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
450 hash
= (hash
+ 1) % nf_conntrack_htable_size
;
457 if (del_timer(&ct
->timeout
)) {
458 death_by_timeout((unsigned long)ct
);
460 NF_CT_STAT_INC_ATOMIC(early_drop
);
466 struct nf_conn
*nf_conntrack_alloc(const struct nf_conntrack_tuple
*orig
,
467 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(&nf_conntrack_count
);
479 if (nf_conntrack_max
&&
480 unlikely(atomic_read(&nf_conntrack_count
) > nf_conntrack_max
)) {
481 unsigned int hash
= hash_conntrack(orig
);
482 if (!early_drop(hash
)) {
483 atomic_dec(&nf_conntrack_count
);
486 "nf_conntrack: table full, dropping"
488 return ERR_PTR(-ENOMEM
);
492 ct
= kmem_cache_zalloc(nf_conntrack_cachep
, GFP_ATOMIC
);
494 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
495 atomic_dec(&nf_conntrack_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
);
504 INIT_RCU_HEAD(&ct
->rcu
);
508 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
510 static void nf_conntrack_free_rcu(struct rcu_head
*head
)
512 struct nf_conn
*ct
= container_of(head
, struct nf_conn
, rcu
);
515 kmem_cache_free(nf_conntrack_cachep
, ct
);
516 atomic_dec(&nf_conntrack_count
);
519 void nf_conntrack_free(struct nf_conn
*ct
)
521 nf_ct_ext_destroy(ct
);
522 call_rcu(&ct
->rcu
, nf_conntrack_free_rcu
);
524 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
526 /* Allocate a new conntrack: we return -ENOMEM if classification
527 failed due to stress. Otherwise it really is unclassifiable. */
528 static struct nf_conntrack_tuple_hash
*
529 init_conntrack(const struct nf_conntrack_tuple
*tuple
,
530 struct nf_conntrack_l3proto
*l3proto
,
531 struct nf_conntrack_l4proto
*l4proto
,
533 unsigned int dataoff
)
536 struct nf_conn_help
*help
;
537 struct nf_conntrack_tuple repl_tuple
;
538 struct nf_conntrack_expect
*exp
;
540 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
541 pr_debug("Can't invert tuple.\n");
545 ct
= nf_conntrack_alloc(tuple
, &repl_tuple
);
546 if (ct
== NULL
|| IS_ERR(ct
)) {
547 pr_debug("Can't allocate conntrack.\n");
548 return (struct nf_conntrack_tuple_hash
*)ct
;
551 if (!l4proto
->new(ct
, skb
, dataoff
)) {
552 nf_conntrack_free(ct
);
553 pr_debug("init conntrack: can't track with proto module\n");
557 spin_lock_bh(&nf_conntrack_lock
);
558 exp
= nf_ct_find_expectation(tuple
);
560 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
562 /* Welcome, Mr. Bond. We've been expecting you... */
563 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
564 ct
->master
= exp
->master
;
566 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
568 rcu_assign_pointer(help
->helper
, exp
->helper
);
571 #ifdef CONFIG_NF_CONNTRACK_MARK
572 ct
->mark
= exp
->master
->mark
;
574 #ifdef CONFIG_NF_CONNTRACK_SECMARK
575 ct
->secmark
= exp
->master
->secmark
;
577 nf_conntrack_get(&ct
->master
->ct_general
);
578 NF_CT_STAT_INC(expect_new
);
580 struct nf_conntrack_helper
*helper
;
582 helper
= __nf_ct_helper_find(&repl_tuple
);
584 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
586 rcu_assign_pointer(help
->helper
, helper
);
591 /* Overload tuple linked list to put us in unconfirmed list. */
592 hlist_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnode
, &unconfirmed
);
594 spin_unlock_bh(&nf_conntrack_lock
);
598 exp
->expectfn(ct
, exp
);
599 nf_ct_expect_put(exp
);
602 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
605 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
606 static inline struct nf_conn
*
607 resolve_normal_ct(struct sk_buff
*skb
,
608 unsigned int dataoff
,
611 struct nf_conntrack_l3proto
*l3proto
,
612 struct nf_conntrack_l4proto
*l4proto
,
614 enum ip_conntrack_info
*ctinfo
)
616 struct nf_conntrack_tuple tuple
;
617 struct nf_conntrack_tuple_hash
*h
;
620 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
621 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
623 pr_debug("resolve_normal_ct: Can't get tuple\n");
627 /* look for tuple match */
628 h
= nf_conntrack_find_get(&tuple
);
630 h
= init_conntrack(&tuple
, l3proto
, l4proto
, skb
, dataoff
);
636 ct
= nf_ct_tuplehash_to_ctrack(h
);
638 /* It exists; we have (non-exclusive) reference. */
639 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
640 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
641 /* Please set reply bit if this packet OK */
644 /* Once we've had two way comms, always ESTABLISHED. */
645 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
646 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
647 *ctinfo
= IP_CT_ESTABLISHED
;
648 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
649 pr_debug("nf_conntrack_in: related packet for %p\n",
651 *ctinfo
= IP_CT_RELATED
;
653 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
658 skb
->nfct
= &ct
->ct_general
;
659 skb
->nfctinfo
= *ctinfo
;
664 nf_conntrack_in(int pf
, unsigned int hooknum
, struct sk_buff
*skb
)
667 enum ip_conntrack_info ctinfo
;
668 struct nf_conntrack_l3proto
*l3proto
;
669 struct nf_conntrack_l4proto
*l4proto
;
670 unsigned int dataoff
;
675 /* Previously seen (loopback or untracked)? Ignore. */
677 NF_CT_STAT_INC_ATOMIC(ignore
);
681 /* rcu_read_lock()ed by nf_hook_slow */
682 l3proto
= __nf_ct_l3proto_find((u_int16_t
)pf
);
683 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
684 &dataoff
, &protonum
);
686 pr_debug("not prepared to track yet or error occured\n");
687 NF_CT_STAT_INC_ATOMIC(error
);
688 NF_CT_STAT_INC_ATOMIC(invalid
);
692 l4proto
= __nf_ct_l4proto_find((u_int16_t
)pf
, protonum
);
694 /* It may be an special packet, error, unclean...
695 * inverse of the return code tells to the netfilter
696 * core what to do with the packet. */
697 if (l4proto
->error
!= NULL
&&
698 (ret
= l4proto
->error(skb
, dataoff
, &ctinfo
, pf
, hooknum
)) <= 0) {
699 NF_CT_STAT_INC_ATOMIC(error
);
700 NF_CT_STAT_INC_ATOMIC(invalid
);
704 ct
= resolve_normal_ct(skb
, dataoff
, pf
, protonum
, l3proto
, l4proto
,
705 &set_reply
, &ctinfo
);
707 /* Not valid part of a connection */
708 NF_CT_STAT_INC_ATOMIC(invalid
);
713 /* Too stressed to deal. */
714 NF_CT_STAT_INC_ATOMIC(drop
);
718 NF_CT_ASSERT(skb
->nfct
);
720 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
722 /* Invalid: inverse of the return code tells
723 * the netfilter core what to do */
724 pr_debug("nf_conntrack_in: Can't track with proto module\n");
725 nf_conntrack_put(skb
->nfct
);
727 NF_CT_STAT_INC_ATOMIC(invalid
);
731 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
732 nf_conntrack_event_cache(IPCT_STATUS
, skb
);
736 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
738 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
739 const struct nf_conntrack_tuple
*orig
)
744 ret
= nf_ct_invert_tuple(inverse
, orig
,
745 __nf_ct_l3proto_find(orig
->src
.l3num
),
746 __nf_ct_l4proto_find(orig
->src
.l3num
,
747 orig
->dst
.protonum
));
751 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
753 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
754 implicitly racy: see __nf_conntrack_confirm */
755 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
756 const struct nf_conntrack_tuple
*newreply
)
758 struct nf_conn_help
*help
= nfct_help(ct
);
759 struct nf_conntrack_helper
*helper
;
761 /* Should be unconfirmed, so not in hash table yet */
762 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
764 pr_debug("Altering reply tuple of %p to ", ct
);
765 nf_ct_dump_tuple(newreply
);
767 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
768 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
772 helper
= __nf_ct_helper_find(newreply
);
773 if (helper
== NULL
) {
775 rcu_assign_pointer(help
->helper
, NULL
);
780 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
784 memset(&help
->help
, 0, sizeof(help
->help
));
787 rcu_assign_pointer(help
->helper
, helper
);
791 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
793 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
794 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
795 enum ip_conntrack_info ctinfo
,
796 const struct sk_buff
*skb
,
797 unsigned long extra_jiffies
,
802 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
805 spin_lock_bh(&nf_conntrack_lock
);
807 /* Only update if this is not a fixed timeout */
808 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
811 /* If not in hash table, timer will not be active yet */
812 if (!nf_ct_is_confirmed(ct
)) {
813 ct
->timeout
.expires
= extra_jiffies
;
814 event
= IPCT_REFRESH
;
816 unsigned long newtime
= jiffies
+ extra_jiffies
;
818 /* Only update the timeout if the new timeout is at least
819 HZ jiffies from the old timeout. Need del_timer for race
820 avoidance (may already be dying). */
821 if (newtime
- ct
->timeout
.expires
>= HZ
822 && del_timer(&ct
->timeout
)) {
823 ct
->timeout
.expires
= newtime
;
824 add_timer(&ct
->timeout
);
825 event
= IPCT_REFRESH
;
830 #ifdef CONFIG_NF_CT_ACCT
832 ct
->counters
[CTINFO2DIR(ctinfo
)].packets
++;
833 ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
+=
834 skb
->len
- skb_network_offset(skb
);
836 if ((ct
->counters
[CTINFO2DIR(ctinfo
)].packets
& 0x80000000)
837 || (ct
->counters
[CTINFO2DIR(ctinfo
)].bytes
& 0x80000000))
838 event
|= IPCT_COUNTER_FILLING
;
842 spin_unlock_bh(&nf_conntrack_lock
);
844 /* must be unlocked when calling event cache */
846 nf_conntrack_event_cache(event
, skb
);
848 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
850 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
852 #include <linux/netfilter/nfnetlink.h>
853 #include <linux/netfilter/nfnetlink_conntrack.h>
854 #include <linux/mutex.h>
856 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
857 * in ip_conntrack_core, since we don't want the protocols to autoload
858 * or depend on ctnetlink */
859 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
860 const struct nf_conntrack_tuple
*tuple
)
862 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
863 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
869 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
871 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
872 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
873 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
875 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
877 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
878 struct nf_conntrack_tuple
*t
)
880 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
883 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
884 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
888 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
891 /* Used by ipt_REJECT and ip6t_REJECT. */
892 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
895 enum ip_conntrack_info ctinfo
;
897 /* This ICMP is in reverse direction to the packet which caused it */
898 ct
= nf_ct_get(skb
, &ctinfo
);
899 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
900 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
902 ctinfo
= IP_CT_RELATED
;
904 /* Attach to new skbuff, and increment count */
905 nskb
->nfct
= &ct
->ct_general
;
906 nskb
->nfctinfo
= ctinfo
;
907 nf_conntrack_get(nskb
->nfct
);
910 /* Bring out ya dead! */
911 static struct nf_conn
*
912 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
913 void *data
, unsigned int *bucket
)
915 struct nf_conntrack_tuple_hash
*h
;
917 struct hlist_node
*n
;
919 spin_lock_bh(&nf_conntrack_lock
);
920 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
921 hlist_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnode
) {
922 ct
= nf_ct_tuplehash_to_ctrack(h
);
927 hlist_for_each_entry(h
, n
, &unconfirmed
, hnode
) {
928 ct
= nf_ct_tuplehash_to_ctrack(h
);
930 set_bit(IPS_DYING_BIT
, &ct
->status
);
932 spin_unlock_bh(&nf_conntrack_lock
);
935 atomic_inc(&ct
->ct_general
.use
);
936 spin_unlock_bh(&nf_conntrack_lock
);
941 nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
944 unsigned int bucket
= 0;
946 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
947 /* Time to push up daises... */
948 if (del_timer(&ct
->timeout
))
949 death_by_timeout((unsigned long)ct
);
950 /* ... else the timer will get him soon. */
955 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
957 static int kill_all(struct nf_conn
*i
, void *data
)
962 void nf_ct_free_hashtable(struct hlist_head
*hash
, int vmalloced
, unsigned int size
)
967 free_pages((unsigned long)hash
,
968 get_order(sizeof(struct hlist_head
) * size
));
970 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
972 void nf_conntrack_flush(void)
974 nf_ct_iterate_cleanup(kill_all
, NULL
);
976 EXPORT_SYMBOL_GPL(nf_conntrack_flush
);
978 /* Mishearing the voices in his head, our hero wonders how he's
979 supposed to kill the mall. */
980 void nf_conntrack_cleanup(void)
982 rcu_assign_pointer(ip_ct_attach
, NULL
);
984 /* This makes sure all current packets have passed through
985 netfilter framework. Roll on, two-stage module
989 nf_ct_event_cache_flush();
991 nf_conntrack_flush();
992 if (atomic_read(&nf_conntrack_count
) != 0) {
994 goto i_see_dead_people
;
996 /* wait until all references to nf_conntrack_untracked are dropped */
997 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1000 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1002 kmem_cache_destroy(nf_conntrack_cachep
);
1003 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1004 nf_conntrack_htable_size
);
1006 nf_conntrack_proto_fini();
1007 nf_conntrack_helper_fini();
1008 nf_conntrack_expect_fini();
1011 struct hlist_head
*nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
)
1013 struct hlist_head
*hash
;
1014 unsigned int size
, i
;
1018 size
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_head
));
1019 hash
= (void*)__get_free_pages(GFP_KERNEL
|__GFP_NOWARN
,
1020 get_order(sizeof(struct hlist_head
)
1024 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1025 hash
= vmalloc(sizeof(struct hlist_head
) * size
);
1029 for (i
= 0; i
< size
; i
++)
1030 INIT_HLIST_HEAD(&hash
[i
]);
1034 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1036 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1038 int i
, bucket
, vmalloced
, old_vmalloced
;
1039 unsigned int hashsize
, old_size
;
1041 struct hlist_head
*hash
, *old_hash
;
1042 struct nf_conntrack_tuple_hash
*h
;
1044 /* On boot, we can set this without any fancy locking. */
1045 if (!nf_conntrack_htable_size
)
1046 return param_set_uint(val
, kp
);
1048 hashsize
= simple_strtoul(val
, NULL
, 0);
1052 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
);
1056 /* We have to rehahs for the new table anyway, so we also can
1057 * use a newrandom seed */
1058 get_random_bytes(&rnd
, 4);
1060 /* Lookups in the old hash might happen in parallel, which means we
1061 * might get false negatives during connection lookup. New connections
1062 * created because of a false negative won't make it into the hash
1063 * though since that required taking the lock.
1065 spin_lock_bh(&nf_conntrack_lock
);
1066 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1067 while (!hlist_empty(&nf_conntrack_hash
[i
])) {
1068 h
= hlist_entry(nf_conntrack_hash
[i
].first
,
1069 struct nf_conntrack_tuple_hash
, hnode
);
1070 hlist_del_rcu(&h
->hnode
);
1071 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1072 hlist_add_head(&h
->hnode
, &hash
[bucket
]);
1075 old_size
= nf_conntrack_htable_size
;
1076 old_vmalloced
= nf_conntrack_vmalloc
;
1077 old_hash
= nf_conntrack_hash
;
1079 nf_conntrack_htable_size
= hashsize
;
1080 nf_conntrack_vmalloc
= vmalloced
;
1081 nf_conntrack_hash
= hash
;
1082 nf_conntrack_hash_rnd
= rnd
;
1083 spin_unlock_bh(&nf_conntrack_lock
);
1085 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1088 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1090 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1091 &nf_conntrack_htable_size
, 0600);
1093 int __init
nf_conntrack_init(void)
1098 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1099 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1100 if (!nf_conntrack_htable_size
) {
1101 nf_conntrack_htable_size
1102 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1103 / sizeof(struct hlist_head
));
1104 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1105 nf_conntrack_htable_size
= 16384;
1106 if (nf_conntrack_htable_size
< 32)
1107 nf_conntrack_htable_size
= 32;
1109 /* Use a max. factor of four by default to get the same max as
1110 * with the old struct list_heads. When a table size is given
1111 * we use the old value of 8 to avoid reducing the max.
1115 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
,
1116 &nf_conntrack_vmalloc
);
1117 if (!nf_conntrack_hash
) {
1118 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1122 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1124 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1125 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1128 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1129 sizeof(struct nf_conn
),
1131 if (!nf_conntrack_cachep
) {
1132 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1136 ret
= nf_conntrack_proto_init();
1138 goto err_free_conntrack_slab
;
1140 ret
= nf_conntrack_expect_init();
1142 goto out_fini_proto
;
1144 ret
= nf_conntrack_helper_init();
1146 goto out_fini_expect
;
1148 /* For use by REJECT target */
1149 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1150 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1152 /* Set up fake conntrack:
1153 - to never be deleted, not in any hashes */
1154 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1155 /* - and look it like as a confirmed connection */
1156 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1161 nf_conntrack_expect_fini();
1163 nf_conntrack_proto_fini();
1164 err_free_conntrack_slab
:
1165 kmem_cache_destroy(nf_conntrack_cachep
);
1167 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_vmalloc
,
1168 nf_conntrack_htable_size
);