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>
32 #include <linux/rculist_nulls.h>
34 #include <net/netfilter/nf_conntrack.h>
35 #include <net/netfilter/nf_conntrack_l3proto.h>
36 #include <net/netfilter/nf_conntrack_l4proto.h>
37 #include <net/netfilter/nf_conntrack_expect.h>
38 #include <net/netfilter/nf_conntrack_helper.h>
39 #include <net/netfilter/nf_conntrack_core.h>
40 #include <net/netfilter/nf_conntrack_extend.h>
41 #include <net/netfilter/nf_conntrack_acct.h>
42 #include <net/netfilter/nf_nat.h>
43 #include <net/netfilter/nf_nat_core.h>
45 #define NF_CONNTRACK_VERSION "0.5.0"
47 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
48 enum nf_nat_manip_type manip
,
49 struct nlattr
*attr
) __read_mostly
;
50 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
52 DEFINE_SPINLOCK(nf_conntrack_lock
);
53 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
55 unsigned int nf_conntrack_htable_size __read_mostly
;
56 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
58 unsigned int nf_conntrack_max __read_mostly
;
59 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
61 struct nf_conn nf_conntrack_untracked __read_mostly
;
62 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
64 static struct kmem_cache
*nf_conntrack_cachep __read_mostly
;
66 static int nf_conntrack_hash_rnd_initted
;
67 static unsigned int nf_conntrack_hash_rnd
;
69 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
70 unsigned int size
, unsigned int rnd
)
75 /* The direction must be ignored, so we hash everything up to the
76 * destination ports (which is a multiple of 4) and treat the last
77 * three bytes manually.
79 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
80 h
= jhash2((u32
*)tuple
, n
,
81 rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
82 tuple
->dst
.protonum
));
84 return ((u64
)h
* size
) >> 32;
87 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
89 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
90 nf_conntrack_hash_rnd
);
94 nf_ct_get_tuple(const struct sk_buff
*skb
,
99 struct nf_conntrack_tuple
*tuple
,
100 const struct nf_conntrack_l3proto
*l3proto
,
101 const struct nf_conntrack_l4proto
*l4proto
)
103 memset(tuple
, 0, sizeof(*tuple
));
105 tuple
->src
.l3num
= l3num
;
106 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
109 tuple
->dst
.protonum
= protonum
;
110 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
112 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
114 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
116 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
117 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
119 struct nf_conntrack_l3proto
*l3proto
;
120 struct nf_conntrack_l4proto
*l4proto
;
121 unsigned int protoff
;
127 l3proto
= __nf_ct_l3proto_find(l3num
);
128 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
129 if (ret
!= NF_ACCEPT
) {
134 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
136 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
142 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
145 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
146 const struct nf_conntrack_tuple
*orig
,
147 const struct nf_conntrack_l3proto
*l3proto
,
148 const struct nf_conntrack_l4proto
*l4proto
)
150 memset(inverse
, 0, sizeof(*inverse
));
152 inverse
->src
.l3num
= orig
->src
.l3num
;
153 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
156 inverse
->dst
.dir
= !orig
->dst
.dir
;
158 inverse
->dst
.protonum
= orig
->dst
.protonum
;
159 return l4proto
->invert_tuple(inverse
, orig
);
161 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
164 clean_from_lists(struct nf_conn
*ct
)
166 pr_debug("clean_from_lists(%p)\n", ct
);
167 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
168 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
170 /* Destroy all pending expectations */
171 nf_ct_remove_expectations(ct
);
175 destroy_conntrack(struct nf_conntrack
*nfct
)
177 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
178 struct net
*net
= nf_ct_net(ct
);
179 struct nf_conntrack_l4proto
*l4proto
;
181 pr_debug("destroy_conntrack(%p)\n", ct
);
182 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
183 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
185 if (!test_bit(IPS_DYING_BIT
, &ct
->status
))
186 nf_conntrack_event(IPCT_DESTROY
, ct
);
187 set_bit(IPS_DYING_BIT
, &ct
->status
);
189 /* To make sure we don't get any weird locking issues here:
190 * destroy_conntrack() MUST NOT be called with a write lock
191 * to nf_conntrack_lock!!! -HW */
193 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
194 if (l4proto
&& l4proto
->destroy
)
195 l4proto
->destroy(ct
);
199 spin_lock_bh(&nf_conntrack_lock
);
200 /* Expectations will have been removed in clean_from_lists,
201 * except TFTP can create an expectation on the first packet,
202 * before connection is in the list, so we need to clean here,
204 nf_ct_remove_expectations(ct
);
206 /* We overload first tuple to link into unconfirmed list. */
207 if (!nf_ct_is_confirmed(ct
)) {
208 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
209 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
212 NF_CT_STAT_INC(net
, delete);
213 spin_unlock_bh(&nf_conntrack_lock
);
216 nf_ct_put(ct
->master
);
218 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
219 nf_conntrack_free(ct
);
222 static void death_by_timeout(unsigned long ul_conntrack
)
224 struct nf_conn
*ct
= (void *)ul_conntrack
;
225 struct net
*net
= nf_ct_net(ct
);
226 struct nf_conn_help
*help
= nfct_help(ct
);
227 struct nf_conntrack_helper
*helper
;
231 helper
= rcu_dereference(help
->helper
);
232 if (helper
&& helper
->destroy
)
237 spin_lock_bh(&nf_conntrack_lock
);
238 /* Inside lock so preempt is disabled on module removal path.
239 * Otherwise we can get spurious warnings. */
240 NF_CT_STAT_INC(net
, delete_list
);
241 clean_from_lists(ct
);
242 spin_unlock_bh(&nf_conntrack_lock
);
248 * - Caller must take a reference on returned object
249 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
251 * - Caller must lock nf_conntrack_lock before calling this function
253 struct nf_conntrack_tuple_hash
*
254 __nf_conntrack_find(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
256 struct nf_conntrack_tuple_hash
*h
;
257 struct hlist_nulls_node
*n
;
258 unsigned int hash
= hash_conntrack(tuple
);
260 /* Disable BHs the entire time since we normally need to disable them
261 * at least once for the stats anyway.
265 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
266 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
267 NF_CT_STAT_INC(net
, found
);
271 NF_CT_STAT_INC(net
, searched
);
274 * if the nulls value we got at the end of this lookup is
275 * not the expected one, we must restart lookup.
276 * We probably met an item that was moved to another chain.
278 if (get_nulls_value(n
) != hash
)
284 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
286 /* Find a connection corresponding to a tuple. */
287 struct nf_conntrack_tuple_hash
*
288 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
290 struct nf_conntrack_tuple_hash
*h
;
295 h
= __nf_conntrack_find(net
, tuple
);
297 ct
= nf_ct_tuplehash_to_ctrack(h
);
298 if (unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
301 if (unlikely(!nf_ct_tuple_equal(tuple
, &h
->tuple
))) {
311 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
313 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
315 unsigned int repl_hash
)
317 struct net
*net
= nf_ct_net(ct
);
319 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
320 &net
->ct
.hash
[hash
]);
321 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
322 &net
->ct
.hash
[repl_hash
]);
325 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
327 unsigned int hash
, repl_hash
;
329 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
330 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
332 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
334 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
336 /* Confirm a connection given skb; places it in hash table */
338 __nf_conntrack_confirm(struct sk_buff
*skb
)
340 unsigned int hash
, repl_hash
;
341 struct nf_conntrack_tuple_hash
*h
;
343 struct nf_conn_help
*help
;
344 struct hlist_nulls_node
*n
;
345 enum ip_conntrack_info ctinfo
;
348 ct
= nf_ct_get(skb
, &ctinfo
);
351 /* ipt_REJECT uses nf_conntrack_attach to attach related
352 ICMP/TCP RST packets in other direction. Actual packet
353 which created connection will be IP_CT_NEW or for an
354 expected connection, IP_CT_RELATED. */
355 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
358 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
359 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
361 /* We're not in hash table, and we refuse to set up related
362 connections for unconfirmed conns. But packet copies and
363 REJECT will give spurious warnings here. */
364 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
366 /* No external references means noone else could have
368 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
369 pr_debug("Confirming conntrack %p\n", ct
);
371 spin_lock_bh(&nf_conntrack_lock
);
373 /* See if there's one in the list already, including reverse:
374 NAT could have grabbed it without realizing, since we're
375 not in the hash. If there is, we lost race. */
376 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
377 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
380 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnnode
)
381 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
385 /* Remove from unconfirmed list */
386 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
388 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
389 /* Timer relative to confirmation time, not original
390 setting time, otherwise we'd get timer wrap in
391 weird delay cases. */
392 ct
->timeout
.expires
+= jiffies
;
393 add_timer(&ct
->timeout
);
394 atomic_inc(&ct
->ct_general
.use
);
395 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
396 NF_CT_STAT_INC(net
, insert
);
397 spin_unlock_bh(&nf_conntrack_lock
);
398 help
= nfct_help(ct
);
399 if (help
&& help
->helper
)
400 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
402 nf_conntrack_event_cache(master_ct(ct
) ?
403 IPCT_RELATED
: IPCT_NEW
, ct
);
407 NF_CT_STAT_INC(net
, insert_failed
);
408 spin_unlock_bh(&nf_conntrack_lock
);
411 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
413 /* Returns true if a connection correspondings to the tuple (required
416 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
417 const struct nf_conn
*ignored_conntrack
)
419 struct net
*net
= nf_ct_net(ignored_conntrack
);
420 struct nf_conntrack_tuple_hash
*h
;
421 struct hlist_nulls_node
*n
;
422 unsigned int hash
= hash_conntrack(tuple
);
424 /* Disable BHs the entire time since we need to disable them at
425 * least once for the stats anyway.
428 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
429 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
430 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
431 NF_CT_STAT_INC(net
, found
);
432 rcu_read_unlock_bh();
435 NF_CT_STAT_INC(net
, searched
);
437 rcu_read_unlock_bh();
441 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
443 #define NF_CT_EVICTION_RANGE 8
445 /* There's a small race here where we may free a just-assured
446 connection. Too bad: we're in trouble anyway. */
447 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
449 /* Use oldest entry, which is roughly LRU */
450 struct nf_conntrack_tuple_hash
*h
;
451 struct nf_conn
*ct
= NULL
, *tmp
;
452 struct hlist_nulls_node
*n
;
453 unsigned int i
, cnt
= 0;
457 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
458 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
460 tmp
= nf_ct_tuplehash_to_ctrack(h
);
461 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
466 if (ct
&& unlikely(!atomic_inc_not_zero(&ct
->ct_general
.use
)))
468 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
470 hash
= (hash
+ 1) % nf_conntrack_htable_size
;
477 if (del_timer(&ct
->timeout
)) {
478 death_by_timeout((unsigned long)ct
);
480 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
486 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
487 const struct nf_conntrack_tuple
*orig
,
488 const struct nf_conntrack_tuple
*repl
,
493 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
494 get_random_bytes(&nf_conntrack_hash_rnd
,
495 sizeof(nf_conntrack_hash_rnd
));
496 nf_conntrack_hash_rnd_initted
= 1;
499 /* We don't want any race condition at early drop stage */
500 atomic_inc(&net
->ct
.count
);
502 if (nf_conntrack_max
&&
503 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
504 unsigned int hash
= hash_conntrack(orig
);
505 if (!early_drop(net
, hash
)) {
506 atomic_dec(&net
->ct
.count
);
509 "nf_conntrack: table full, dropping"
511 return ERR_PTR(-ENOMEM
);
515 ct
= kmem_cache_zalloc(nf_conntrack_cachep
, gfp
);
517 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
518 atomic_dec(&net
->ct
.count
);
519 return ERR_PTR(-ENOMEM
);
522 spin_lock_init(&ct
->lock
);
523 atomic_set(&ct
->ct_general
.use
, 1);
524 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
525 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
526 /* Don't set timer yet: wait for confirmation */
527 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
534 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
536 void nf_conntrack_free(struct nf_conn
*ct
)
538 struct net
*net
= nf_ct_net(ct
);
540 nf_ct_ext_destroy(ct
);
541 atomic_dec(&net
->ct
.count
);
543 kmem_cache_free(nf_conntrack_cachep
, ct
);
545 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
547 /* Allocate a new conntrack: we return -ENOMEM if classification
548 failed due to stress. Otherwise it really is unclassifiable. */
549 static struct nf_conntrack_tuple_hash
*
550 init_conntrack(struct net
*net
,
551 const struct nf_conntrack_tuple
*tuple
,
552 struct nf_conntrack_l3proto
*l3proto
,
553 struct nf_conntrack_l4proto
*l4proto
,
555 unsigned int dataoff
)
558 struct nf_conn_help
*help
;
559 struct nf_conntrack_tuple repl_tuple
;
560 struct nf_conntrack_expect
*exp
;
562 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
563 pr_debug("Can't invert tuple.\n");
567 ct
= nf_conntrack_alloc(net
, tuple
, &repl_tuple
, GFP_ATOMIC
);
569 pr_debug("Can't allocate conntrack.\n");
570 return (struct nf_conntrack_tuple_hash
*)ct
;
573 if (!l4proto
->new(ct
, skb
, dataoff
)) {
574 nf_conntrack_free(ct
);
575 pr_debug("init conntrack: can't track with proto module\n");
579 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
581 spin_lock_bh(&nf_conntrack_lock
);
582 exp
= nf_ct_find_expectation(net
, tuple
);
584 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
586 /* Welcome, Mr. Bond. We've been expecting you... */
587 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
588 ct
->master
= exp
->master
;
590 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
592 rcu_assign_pointer(help
->helper
, exp
->helper
);
595 #ifdef CONFIG_NF_CONNTRACK_MARK
596 ct
->mark
= exp
->master
->mark
;
598 #ifdef CONFIG_NF_CONNTRACK_SECMARK
599 ct
->secmark
= exp
->master
->secmark
;
601 nf_conntrack_get(&ct
->master
->ct_general
);
602 NF_CT_STAT_INC(net
, expect_new
);
604 __nf_ct_try_assign_helper(ct
, GFP_ATOMIC
);
605 NF_CT_STAT_INC(net
, new);
608 /* Overload tuple linked list to put us in unconfirmed list. */
609 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
610 &net
->ct
.unconfirmed
);
612 spin_unlock_bh(&nf_conntrack_lock
);
616 exp
->expectfn(ct
, exp
);
617 nf_ct_expect_put(exp
);
620 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
623 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
624 static inline struct nf_conn
*
625 resolve_normal_ct(struct net
*net
,
627 unsigned int dataoff
,
630 struct nf_conntrack_l3proto
*l3proto
,
631 struct nf_conntrack_l4proto
*l4proto
,
633 enum ip_conntrack_info
*ctinfo
)
635 struct nf_conntrack_tuple tuple
;
636 struct nf_conntrack_tuple_hash
*h
;
639 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
640 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
642 pr_debug("resolve_normal_ct: Can't get tuple\n");
646 /* look for tuple match */
647 h
= nf_conntrack_find_get(net
, &tuple
);
649 h
= init_conntrack(net
, &tuple
, l3proto
, l4proto
, skb
, dataoff
);
655 ct
= nf_ct_tuplehash_to_ctrack(h
);
657 /* It exists; we have (non-exclusive) reference. */
658 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
659 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
660 /* Please set reply bit if this packet OK */
663 /* Once we've had two way comms, always ESTABLISHED. */
664 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
665 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
666 *ctinfo
= IP_CT_ESTABLISHED
;
667 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
668 pr_debug("nf_conntrack_in: related packet for %p\n",
670 *ctinfo
= IP_CT_RELATED
;
672 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
677 skb
->nfct
= &ct
->ct_general
;
678 skb
->nfctinfo
= *ctinfo
;
683 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
687 enum ip_conntrack_info ctinfo
;
688 struct nf_conntrack_l3proto
*l3proto
;
689 struct nf_conntrack_l4proto
*l4proto
;
690 unsigned int dataoff
;
695 /* Previously seen (loopback or untracked)? Ignore. */
697 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
701 /* rcu_read_lock()ed by nf_hook_slow */
702 l3proto
= __nf_ct_l3proto_find(pf
);
703 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
704 &dataoff
, &protonum
);
706 pr_debug("not prepared to track yet or error occured\n");
707 NF_CT_STAT_INC_ATOMIC(net
, error
);
708 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
712 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
714 /* It may be an special packet, error, unclean...
715 * inverse of the return code tells to the netfilter
716 * core what to do with the packet. */
717 if (l4proto
->error
!= NULL
) {
718 ret
= l4proto
->error(net
, skb
, dataoff
, &ctinfo
, pf
, hooknum
);
720 NF_CT_STAT_INC_ATOMIC(net
, error
);
721 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
726 ct
= resolve_normal_ct(net
, skb
, dataoff
, pf
, protonum
,
727 l3proto
, l4proto
, &set_reply
, &ctinfo
);
729 /* Not valid part of a connection */
730 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
735 /* Too stressed to deal. */
736 NF_CT_STAT_INC_ATOMIC(net
, drop
);
740 NF_CT_ASSERT(skb
->nfct
);
742 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
744 /* Invalid: inverse of the return code tells
745 * the netfilter core what to do */
746 pr_debug("nf_conntrack_in: Can't track with proto module\n");
747 nf_conntrack_put(skb
->nfct
);
749 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
751 NF_CT_STAT_INC_ATOMIC(net
, drop
);
755 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
756 nf_conntrack_event_cache(IPCT_STATUS
, ct
);
760 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
762 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
763 const struct nf_conntrack_tuple
*orig
)
768 ret
= nf_ct_invert_tuple(inverse
, orig
,
769 __nf_ct_l3proto_find(orig
->src
.l3num
),
770 __nf_ct_l4proto_find(orig
->src
.l3num
,
771 orig
->dst
.protonum
));
775 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
777 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
778 implicitly racy: see __nf_conntrack_confirm */
779 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
780 const struct nf_conntrack_tuple
*newreply
)
782 struct nf_conn_help
*help
= nfct_help(ct
);
784 /* Should be unconfirmed, so not in hash table yet */
785 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
787 pr_debug("Altering reply tuple of %p to ", ct
);
788 nf_ct_dump_tuple(newreply
);
790 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
791 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
795 __nf_ct_try_assign_helper(ct
, GFP_ATOMIC
);
798 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
800 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
801 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
802 enum ip_conntrack_info ctinfo
,
803 const struct sk_buff
*skb
,
804 unsigned long extra_jiffies
,
807 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
810 /* Only update if this is not a fixed timeout */
811 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
814 /* If not in hash table, timer will not be active yet */
815 if (!nf_ct_is_confirmed(ct
)) {
816 ct
->timeout
.expires
= extra_jiffies
;
818 unsigned long newtime
= jiffies
+ extra_jiffies
;
820 /* Only update the timeout if the new timeout is at least
821 HZ jiffies from the old timeout. Need del_timer for race
822 avoidance (may already be dying). */
823 if (newtime
- ct
->timeout
.expires
>= HZ
)
824 mod_timer_pending(&ct
->timeout
, newtime
);
829 struct nf_conn_counter
*acct
;
831 acct
= nf_conn_acct_find(ct
);
833 spin_lock_bh(&ct
->lock
);
834 acct
[CTINFO2DIR(ctinfo
)].packets
++;
835 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
836 skb
->len
- skb_network_offset(skb
);
837 spin_unlock_bh(&ct
->lock
);
841 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
843 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
844 enum ip_conntrack_info ctinfo
,
845 const struct sk_buff
*skb
,
849 struct nf_conn_counter
*acct
;
851 acct
= nf_conn_acct_find(ct
);
853 spin_lock_bh(&ct
->lock
);
854 acct
[CTINFO2DIR(ctinfo
)].packets
++;
855 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
856 skb
->len
- skb_network_offset(skb
);
857 spin_unlock_bh(&ct
->lock
);
861 if (del_timer(&ct
->timeout
)) {
862 ct
->timeout
.function((unsigned long)ct
);
867 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
869 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
871 #include <linux/netfilter/nfnetlink.h>
872 #include <linux/netfilter/nfnetlink_conntrack.h>
873 #include <linux/mutex.h>
875 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
876 * in ip_conntrack_core, since we don't want the protocols to autoload
877 * or depend on ctnetlink */
878 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
879 const struct nf_conntrack_tuple
*tuple
)
881 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
882 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
888 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
890 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
891 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
892 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
894 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
896 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
897 struct nf_conntrack_tuple
*t
)
899 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
902 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
903 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
907 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
909 int nf_ct_port_nlattr_tuple_size(void)
911 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
913 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
916 /* Used by ipt_REJECT and ip6t_REJECT. */
917 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
920 enum ip_conntrack_info ctinfo
;
922 /* This ICMP is in reverse direction to the packet which caused it */
923 ct
= nf_ct_get(skb
, &ctinfo
);
924 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
925 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
927 ctinfo
= IP_CT_RELATED
;
929 /* Attach to new skbuff, and increment count */
930 nskb
->nfct
= &ct
->ct_general
;
931 nskb
->nfctinfo
= ctinfo
;
932 nf_conntrack_get(nskb
->nfct
);
935 /* Bring out ya dead! */
936 static struct nf_conn
*
937 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
938 void *data
, unsigned int *bucket
)
940 struct nf_conntrack_tuple_hash
*h
;
942 struct hlist_nulls_node
*n
;
944 spin_lock_bh(&nf_conntrack_lock
);
945 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
946 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
947 ct
= nf_ct_tuplehash_to_ctrack(h
);
952 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnnode
) {
953 ct
= nf_ct_tuplehash_to_ctrack(h
);
955 set_bit(IPS_DYING_BIT
, &ct
->status
);
957 spin_unlock_bh(&nf_conntrack_lock
);
960 atomic_inc(&ct
->ct_general
.use
);
961 spin_unlock_bh(&nf_conntrack_lock
);
965 void nf_ct_iterate_cleanup(struct net
*net
,
966 int (*iter
)(struct nf_conn
*i
, void *data
),
970 unsigned int bucket
= 0;
972 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
973 /* Time to push up daises... */
974 if (del_timer(&ct
->timeout
))
975 death_by_timeout((unsigned long)ct
);
976 /* ... else the timer will get him soon. */
981 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
983 struct __nf_ct_flush_report
{
988 static int kill_report(struct nf_conn
*i
, void *data
)
990 struct __nf_ct_flush_report
*fr
= (struct __nf_ct_flush_report
*)data
;
992 /* get_next_corpse sets the dying bit for us */
993 nf_conntrack_event_report(IPCT_DESTROY
,
1000 static int kill_all(struct nf_conn
*i
, void *data
)
1005 void nf_ct_free_hashtable(void *hash
, int vmalloced
, unsigned int size
)
1010 free_pages((unsigned long)hash
,
1011 get_order(sizeof(struct hlist_head
) * size
));
1013 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1015 void nf_conntrack_flush_report(struct net
*net
, u32 pid
, int report
)
1017 struct __nf_ct_flush_report fr
= {
1021 nf_ct_iterate_cleanup(net
, kill_report
, &fr
);
1023 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report
);
1025 static void nf_conntrack_cleanup_init_net(void)
1027 nf_conntrack_helper_fini();
1028 nf_conntrack_proto_fini();
1029 kmem_cache_destroy(nf_conntrack_cachep
);
1032 static void nf_conntrack_cleanup_net(struct net
*net
)
1034 nf_ct_event_cache_flush(net
);
1035 nf_conntrack_ecache_fini(net
);
1037 nf_ct_iterate_cleanup(net
, kill_all
, NULL
);
1038 if (atomic_read(&net
->ct
.count
) != 0) {
1040 goto i_see_dead_people
;
1042 /* wait until all references to nf_conntrack_untracked are dropped */
1043 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1046 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1047 nf_conntrack_htable_size
);
1048 nf_conntrack_acct_fini(net
);
1049 nf_conntrack_expect_fini(net
);
1050 free_percpu(net
->ct
.stat
);
1053 /* Mishearing the voices in his head, our hero wonders how he's
1054 supposed to kill the mall. */
1055 void nf_conntrack_cleanup(struct net
*net
)
1057 if (net_eq(net
, &init_net
))
1058 rcu_assign_pointer(ip_ct_attach
, NULL
);
1060 /* This makes sure all current packets have passed through
1061 netfilter framework. Roll on, two-stage module
1065 nf_conntrack_cleanup_net(net
);
1067 if (net_eq(net
, &init_net
)) {
1068 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1069 nf_conntrack_cleanup_init_net();
1073 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
, int nulls
)
1075 struct hlist_nulls_head
*hash
;
1076 unsigned int nr_slots
, i
;
1081 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1082 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1083 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1084 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1088 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1089 hash
= __vmalloc(sz
, GFP_KERNEL
| __GFP_ZERO
, PAGE_KERNEL
);
1093 for (i
= 0; i
< nr_slots
; i
++)
1094 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1098 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1100 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1102 int i
, bucket
, vmalloced
, old_vmalloced
;
1103 unsigned int hashsize
, old_size
;
1105 struct hlist_nulls_head
*hash
, *old_hash
;
1106 struct nf_conntrack_tuple_hash
*h
;
1108 /* On boot, we can set this without any fancy locking. */
1109 if (!nf_conntrack_htable_size
)
1110 return param_set_uint(val
, kp
);
1112 hashsize
= simple_strtoul(val
, NULL
, 0);
1116 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
, 1);
1120 /* We have to rehahs for the new table anyway, so we also can
1121 * use a newrandom seed */
1122 get_random_bytes(&rnd
, sizeof(rnd
));
1124 /* Lookups in the old hash might happen in parallel, which means we
1125 * might get false negatives during connection lookup. New connections
1126 * created because of a false negative won't make it into the hash
1127 * though since that required taking the lock.
1129 spin_lock_bh(&nf_conntrack_lock
);
1130 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1131 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1132 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1133 struct nf_conntrack_tuple_hash
, hnnode
);
1134 hlist_nulls_del_rcu(&h
->hnnode
);
1135 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1136 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1139 old_size
= nf_conntrack_htable_size
;
1140 old_vmalloced
= init_net
.ct
.hash_vmalloc
;
1141 old_hash
= init_net
.ct
.hash
;
1143 nf_conntrack_htable_size
= hashsize
;
1144 init_net
.ct
.hash_vmalloc
= vmalloced
;
1145 init_net
.ct
.hash
= hash
;
1146 nf_conntrack_hash_rnd
= rnd
;
1147 spin_unlock_bh(&nf_conntrack_lock
);
1149 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1152 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1154 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1155 &nf_conntrack_htable_size
, 0600);
1157 static int nf_conntrack_init_init_net(void)
1162 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1163 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1164 if (!nf_conntrack_htable_size
) {
1165 nf_conntrack_htable_size
1166 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1167 / sizeof(struct hlist_head
));
1168 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1169 nf_conntrack_htable_size
= 16384;
1170 if (nf_conntrack_htable_size
< 32)
1171 nf_conntrack_htable_size
= 32;
1173 /* Use a max. factor of four by default to get the same max as
1174 * with the old struct list_heads. When a table size is given
1175 * we use the old value of 8 to avoid reducing the max.
1179 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1181 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1182 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1185 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1186 sizeof(struct nf_conn
),
1187 0, SLAB_DESTROY_BY_RCU
, NULL
);
1188 if (!nf_conntrack_cachep
) {
1189 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1194 ret
= nf_conntrack_proto_init();
1198 ret
= nf_conntrack_helper_init();
1205 nf_conntrack_proto_fini();
1207 kmem_cache_destroy(nf_conntrack_cachep
);
1212 static int nf_conntrack_init_net(struct net
*net
)
1216 atomic_set(&net
->ct
.count
, 0);
1217 INIT_HLIST_NULLS_HEAD(&net
->ct
.unconfirmed
, 0);
1218 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1219 if (!net
->ct
.stat
) {
1223 ret
= nf_conntrack_ecache_init(net
);
1226 net
->ct
.hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
,
1227 &net
->ct
.hash_vmalloc
, 1);
1228 if (!net
->ct
.hash
) {
1230 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1233 ret
= nf_conntrack_expect_init(net
);
1236 ret
= nf_conntrack_acct_init(net
);
1240 /* Set up fake conntrack:
1241 - to never be deleted, not in any hashes */
1242 #ifdef CONFIG_NET_NS
1243 nf_conntrack_untracked
.ct_net
= &init_net
;
1245 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1246 /* - and look it like as a confirmed connection */
1247 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1252 nf_conntrack_expect_fini(net
);
1254 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1255 nf_conntrack_htable_size
);
1257 nf_conntrack_ecache_fini(net
);
1259 free_percpu(net
->ct
.stat
);
1264 int nf_conntrack_init(struct net
*net
)
1268 if (net_eq(net
, &init_net
)) {
1269 ret
= nf_conntrack_init_init_net();
1273 ret
= nf_conntrack_init_net(net
);
1277 if (net_eq(net
, &init_net
)) {
1278 /* For use by REJECT target */
1279 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1280 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1285 if (net_eq(net
, &init_net
))
1286 nf_conntrack_cleanup_init_net();