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_conntrack_ecache.h>
43 #include <net/netfilter/nf_nat.h>
44 #include <net/netfilter/nf_nat_core.h>
46 #define NF_CONNTRACK_VERSION "0.5.0"
48 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
49 enum nf_nat_manip_type manip
,
50 struct nlattr
*attr
) __read_mostly
;
51 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
53 DEFINE_SPINLOCK(nf_conntrack_lock
);
54 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
56 unsigned int nf_conntrack_htable_size __read_mostly
;
57 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
59 unsigned int nf_conntrack_max __read_mostly
;
60 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
62 struct nf_conn nf_conntrack_untracked __read_mostly
;
63 EXPORT_SYMBOL_GPL(nf_conntrack_untracked
);
65 static struct kmem_cache
*nf_conntrack_cachep __read_mostly
;
67 static int nf_conntrack_hash_rnd_initted
;
68 static unsigned int nf_conntrack_hash_rnd
;
70 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
71 unsigned int size
, unsigned int rnd
)
76 /* The direction must be ignored, so we hash everything up to the
77 * destination ports (which is a multiple of 4) and treat the last
78 * three bytes manually.
80 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
81 h
= jhash2((u32
*)tuple
, n
,
82 rnd
^ (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
83 tuple
->dst
.protonum
));
85 return ((u64
)h
* size
) >> 32;
88 static inline u_int32_t
hash_conntrack(const struct nf_conntrack_tuple
*tuple
)
90 return __hash_conntrack(tuple
, nf_conntrack_htable_size
,
91 nf_conntrack_hash_rnd
);
95 nf_ct_get_tuple(const struct sk_buff
*skb
,
100 struct nf_conntrack_tuple
*tuple
,
101 const struct nf_conntrack_l3proto
*l3proto
,
102 const struct nf_conntrack_l4proto
*l4proto
)
104 memset(tuple
, 0, sizeof(*tuple
));
106 tuple
->src
.l3num
= l3num
;
107 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
110 tuple
->dst
.protonum
= protonum
;
111 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
113 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
115 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
117 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
118 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
120 struct nf_conntrack_l3proto
*l3proto
;
121 struct nf_conntrack_l4proto
*l4proto
;
122 unsigned int protoff
;
128 l3proto
= __nf_ct_l3proto_find(l3num
);
129 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
130 if (ret
!= NF_ACCEPT
) {
135 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
137 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
143 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
146 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
147 const struct nf_conntrack_tuple
*orig
,
148 const struct nf_conntrack_l3proto
*l3proto
,
149 const struct nf_conntrack_l4proto
*l4proto
)
151 memset(inverse
, 0, sizeof(*inverse
));
153 inverse
->src
.l3num
= orig
->src
.l3num
;
154 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
157 inverse
->dst
.dir
= !orig
->dst
.dir
;
159 inverse
->dst
.protonum
= orig
->dst
.protonum
;
160 return l4proto
->invert_tuple(inverse
, orig
);
162 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
165 clean_from_lists(struct nf_conn
*ct
)
167 pr_debug("clean_from_lists(%p)\n", ct
);
168 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
169 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
171 /* Destroy all pending expectations */
172 nf_ct_remove_expectations(ct
);
176 destroy_conntrack(struct nf_conntrack
*nfct
)
178 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
179 struct net
*net
= nf_ct_net(ct
);
180 struct nf_conntrack_l4proto
*l4proto
;
182 pr_debug("destroy_conntrack(%p)\n", ct
);
183 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
184 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
186 /* To make sure we don't get any weird locking issues here:
187 * destroy_conntrack() MUST NOT be called with a write lock
188 * to nf_conntrack_lock!!! -HW */
190 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
191 if (l4proto
&& l4proto
->destroy
)
192 l4proto
->destroy(ct
);
196 spin_lock_bh(&nf_conntrack_lock
);
197 /* Expectations will have been removed in clean_from_lists,
198 * except TFTP can create an expectation on the first packet,
199 * before connection is in the list, so we need to clean here,
201 nf_ct_remove_expectations(ct
);
203 /* We overload first tuple to link into unconfirmed list. */
204 if (!nf_ct_is_confirmed(ct
)) {
205 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
206 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
209 NF_CT_STAT_INC(net
, delete);
210 spin_unlock_bh(&nf_conntrack_lock
);
213 nf_ct_put(ct
->master
);
215 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
216 nf_conntrack_free(ct
);
219 void nf_ct_delete_from_lists(struct nf_conn
*ct
)
221 struct net
*net
= nf_ct_net(ct
);
223 nf_ct_helper_destroy(ct
);
224 spin_lock_bh(&nf_conntrack_lock
);
225 /* Inside lock so preempt is disabled on module removal path.
226 * Otherwise we can get spurious warnings. */
227 NF_CT_STAT_INC(net
, delete_list
);
228 clean_from_lists(ct
);
229 spin_unlock_bh(&nf_conntrack_lock
);
231 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists
);
233 static void death_by_event(unsigned long ul_conntrack
)
235 struct nf_conn
*ct
= (void *)ul_conntrack
;
236 struct net
*net
= nf_ct_net(ct
);
238 if (nf_conntrack_event(IPCT_DESTROY
, ct
) < 0) {
239 /* bad luck, let's retry again */
240 ct
->timeout
.expires
= jiffies
+
241 (random32() % net
->ct
.sysctl_events_retry_timeout
);
242 add_timer(&ct
->timeout
);
245 /* we've got the event delivered, now it's dying */
246 set_bit(IPS_DYING_BIT
, &ct
->status
);
247 spin_lock(&nf_conntrack_lock
);
248 hlist_nulls_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
249 spin_unlock(&nf_conntrack_lock
);
253 void nf_ct_insert_dying_list(struct nf_conn
*ct
)
255 struct net
*net
= nf_ct_net(ct
);
257 /* add this conntrack to the dying list */
258 spin_lock_bh(&nf_conntrack_lock
);
259 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
261 spin_unlock_bh(&nf_conntrack_lock
);
262 /* set a new timer to retry event delivery */
263 setup_timer(&ct
->timeout
, death_by_event
, (unsigned long)ct
);
264 ct
->timeout
.expires
= jiffies
+
265 (random32() % net
->ct
.sysctl_events_retry_timeout
);
266 add_timer(&ct
->timeout
);
268 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list
);
270 static void death_by_timeout(unsigned long ul_conntrack
)
272 struct nf_conn
*ct
= (void *)ul_conntrack
;
274 if (!test_bit(IPS_DYING_BIT
, &ct
->status
) &&
275 unlikely(nf_conntrack_event(IPCT_DESTROY
, ct
) < 0)) {
276 /* destroy event was not delivered */
277 nf_ct_delete_from_lists(ct
);
278 nf_ct_insert_dying_list(ct
);
281 set_bit(IPS_DYING_BIT
, &ct
->status
);
282 nf_ct_delete_from_lists(ct
);
288 * - Caller must take a reference on returned object
289 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
291 * - Caller must lock nf_conntrack_lock before calling this function
293 struct nf_conntrack_tuple_hash
*
294 __nf_conntrack_find(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
296 struct nf_conntrack_tuple_hash
*h
;
297 struct hlist_nulls_node
*n
;
298 unsigned int hash
= hash_conntrack(tuple
);
300 /* Disable BHs the entire time since we normally need to disable them
301 * at least once for the stats anyway.
305 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
306 if (nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
307 NF_CT_STAT_INC(net
, found
);
311 NF_CT_STAT_INC(net
, searched
);
314 * if the nulls value we got at the end of this lookup is
315 * not the expected one, we must restart lookup.
316 * We probably met an item that was moved to another chain.
318 if (get_nulls_value(n
) != hash
)
324 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
326 /* Find a connection corresponding to a tuple. */
327 struct nf_conntrack_tuple_hash
*
328 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_tuple
*tuple
)
330 struct nf_conntrack_tuple_hash
*h
;
335 h
= __nf_conntrack_find(net
, tuple
);
337 ct
= nf_ct_tuplehash_to_ctrack(h
);
338 if (unlikely(nf_ct_is_dying(ct
) ||
339 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
342 if (unlikely(!nf_ct_tuple_equal(tuple
, &h
->tuple
))) {
352 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
354 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
356 unsigned int repl_hash
)
358 struct net
*net
= nf_ct_net(ct
);
360 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
361 &net
->ct
.hash
[hash
]);
362 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
363 &net
->ct
.hash
[repl_hash
]);
366 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
368 unsigned int hash
, repl_hash
;
370 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
371 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
373 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
375 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
377 /* Confirm a connection given skb; places it in hash table */
379 __nf_conntrack_confirm(struct sk_buff
*skb
)
381 unsigned int hash
, repl_hash
;
382 struct nf_conntrack_tuple_hash
*h
;
384 struct nf_conn_help
*help
;
385 struct hlist_nulls_node
*n
;
386 enum ip_conntrack_info ctinfo
;
389 ct
= nf_ct_get(skb
, &ctinfo
);
392 /* ipt_REJECT uses nf_conntrack_attach to attach related
393 ICMP/TCP RST packets in other direction. Actual packet
394 which created connection will be IP_CT_NEW or for an
395 expected connection, IP_CT_RELATED. */
396 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
399 hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
400 repl_hash
= hash_conntrack(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
402 /* We're not in hash table, and we refuse to set up related
403 connections for unconfirmed conns. But packet copies and
404 REJECT will give spurious warnings here. */
405 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
407 /* No external references means noone else could have
409 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
410 pr_debug("Confirming conntrack %p\n", ct
);
412 spin_lock_bh(&nf_conntrack_lock
);
414 /* See if there's one in the list already, including reverse:
415 NAT could have grabbed it without realizing, since we're
416 not in the hash. If there is, we lost race. */
417 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
418 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
421 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnnode
)
422 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
426 /* Remove from unconfirmed list */
427 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
429 /* Timer relative to confirmation time, not original
430 setting time, otherwise we'd get timer wrap in
431 weird delay cases. */
432 ct
->timeout
.expires
+= jiffies
;
433 add_timer(&ct
->timeout
);
434 atomic_inc(&ct
->ct_general
.use
);
435 set_bit(IPS_CONFIRMED_BIT
, &ct
->status
);
437 /* Since the lookup is lockless, hash insertion must be done after
438 * starting the timer and setting the CONFIRMED bit. The RCU barriers
439 * guarantee that no other CPU can find the conntrack before the above
440 * stores are visible.
442 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
443 NF_CT_STAT_INC(net
, insert
);
444 spin_unlock_bh(&nf_conntrack_lock
);
446 help
= nfct_help(ct
);
447 if (help
&& help
->helper
)
448 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
450 nf_conntrack_event_cache(master_ct(ct
) ?
451 IPCT_RELATED
: IPCT_NEW
, ct
);
455 NF_CT_STAT_INC(net
, insert_failed
);
456 spin_unlock_bh(&nf_conntrack_lock
);
459 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
461 /* Returns true if a connection correspondings to the tuple (required
464 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
465 const struct nf_conn
*ignored_conntrack
)
467 struct net
*net
= nf_ct_net(ignored_conntrack
);
468 struct nf_conntrack_tuple_hash
*h
;
469 struct hlist_nulls_node
*n
;
470 unsigned int hash
= hash_conntrack(tuple
);
472 /* Disable BHs the entire time since we need to disable them at
473 * least once for the stats anyway.
476 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
477 if (nf_ct_tuplehash_to_ctrack(h
) != ignored_conntrack
&&
478 nf_ct_tuple_equal(tuple
, &h
->tuple
)) {
479 NF_CT_STAT_INC(net
, found
);
480 rcu_read_unlock_bh();
483 NF_CT_STAT_INC(net
, searched
);
485 rcu_read_unlock_bh();
489 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
491 #define NF_CT_EVICTION_RANGE 8
493 /* There's a small race here where we may free a just-assured
494 connection. Too bad: we're in trouble anyway. */
495 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
497 /* Use oldest entry, which is roughly LRU */
498 struct nf_conntrack_tuple_hash
*h
;
499 struct nf_conn
*ct
= NULL
, *tmp
;
500 struct hlist_nulls_node
*n
;
501 unsigned int i
, cnt
= 0;
505 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
506 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
508 tmp
= nf_ct_tuplehash_to_ctrack(h
);
509 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
514 if (ct
&& unlikely(nf_ct_is_dying(ct
) ||
515 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
517 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
519 hash
= (hash
+ 1) % nf_conntrack_htable_size
;
526 if (del_timer(&ct
->timeout
)) {
527 death_by_timeout((unsigned long)ct
);
529 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
535 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
536 const struct nf_conntrack_tuple
*orig
,
537 const struct nf_conntrack_tuple
*repl
,
542 if (unlikely(!nf_conntrack_hash_rnd_initted
)) {
543 get_random_bytes(&nf_conntrack_hash_rnd
,
544 sizeof(nf_conntrack_hash_rnd
));
545 nf_conntrack_hash_rnd_initted
= 1;
548 /* We don't want any race condition at early drop stage */
549 atomic_inc(&net
->ct
.count
);
551 if (nf_conntrack_max
&&
552 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
553 unsigned int hash
= hash_conntrack(orig
);
554 if (!early_drop(net
, hash
)) {
555 atomic_dec(&net
->ct
.count
);
558 "nf_conntrack: table full, dropping"
560 return ERR_PTR(-ENOMEM
);
564 ct
= kmem_cache_zalloc(nf_conntrack_cachep
, gfp
);
566 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
567 atomic_dec(&net
->ct
.count
);
568 return ERR_PTR(-ENOMEM
);
571 spin_lock_init(&ct
->lock
);
572 atomic_set(&ct
->ct_general
.use
, 1);
573 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
574 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
575 /* Don't set timer yet: wait for confirmation */
576 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
583 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
585 void nf_conntrack_free(struct nf_conn
*ct
)
587 struct net
*net
= nf_ct_net(ct
);
589 nf_ct_ext_destroy(ct
);
590 atomic_dec(&net
->ct
.count
);
592 kmem_cache_free(nf_conntrack_cachep
, ct
);
594 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
596 /* Allocate a new conntrack: we return -ENOMEM if classification
597 failed due to stress. Otherwise it really is unclassifiable. */
598 static struct nf_conntrack_tuple_hash
*
599 init_conntrack(struct net
*net
,
600 const struct nf_conntrack_tuple
*tuple
,
601 struct nf_conntrack_l3proto
*l3proto
,
602 struct nf_conntrack_l4proto
*l4proto
,
604 unsigned int dataoff
)
607 struct nf_conn_help
*help
;
608 struct nf_conntrack_tuple repl_tuple
;
609 struct nf_conntrack_expect
*exp
;
611 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
612 pr_debug("Can't invert tuple.\n");
616 ct
= nf_conntrack_alloc(net
, tuple
, &repl_tuple
, GFP_ATOMIC
);
618 pr_debug("Can't allocate conntrack.\n");
619 return (struct nf_conntrack_tuple_hash
*)ct
;
622 if (!l4proto
->new(ct
, skb
, dataoff
)) {
623 nf_conntrack_free(ct
);
624 pr_debug("init conntrack: can't track with proto module\n");
628 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
629 nf_ct_ecache_ext_add(ct
, GFP_ATOMIC
);
631 spin_lock_bh(&nf_conntrack_lock
);
632 exp
= nf_ct_find_expectation(net
, tuple
);
634 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
636 /* Welcome, Mr. Bond. We've been expecting you... */
637 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
638 ct
->master
= exp
->master
;
640 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
642 rcu_assign_pointer(help
->helper
, exp
->helper
);
645 #ifdef CONFIG_NF_CONNTRACK_MARK
646 ct
->mark
= exp
->master
->mark
;
648 #ifdef CONFIG_NF_CONNTRACK_SECMARK
649 ct
->secmark
= exp
->master
->secmark
;
651 nf_conntrack_get(&ct
->master
->ct_general
);
652 NF_CT_STAT_INC(net
, expect_new
);
654 __nf_ct_try_assign_helper(ct
, GFP_ATOMIC
);
655 NF_CT_STAT_INC(net
, new);
658 /* Overload tuple linked list to put us in unconfirmed list. */
659 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
660 &net
->ct
.unconfirmed
);
662 spin_unlock_bh(&nf_conntrack_lock
);
666 exp
->expectfn(ct
, exp
);
667 nf_ct_expect_put(exp
);
670 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
673 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
674 static inline struct nf_conn
*
675 resolve_normal_ct(struct net
*net
,
677 unsigned int dataoff
,
680 struct nf_conntrack_l3proto
*l3proto
,
681 struct nf_conntrack_l4proto
*l4proto
,
683 enum ip_conntrack_info
*ctinfo
)
685 struct nf_conntrack_tuple tuple
;
686 struct nf_conntrack_tuple_hash
*h
;
689 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
690 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
692 pr_debug("resolve_normal_ct: Can't get tuple\n");
696 /* look for tuple match */
697 h
= nf_conntrack_find_get(net
, &tuple
);
699 h
= init_conntrack(net
, &tuple
, l3proto
, l4proto
, skb
, dataoff
);
705 ct
= nf_ct_tuplehash_to_ctrack(h
);
707 /* It exists; we have (non-exclusive) reference. */
708 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
709 *ctinfo
= IP_CT_ESTABLISHED
+ IP_CT_IS_REPLY
;
710 /* Please set reply bit if this packet OK */
713 /* Once we've had two way comms, always ESTABLISHED. */
714 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
715 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
716 *ctinfo
= IP_CT_ESTABLISHED
;
717 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
718 pr_debug("nf_conntrack_in: related packet for %p\n",
720 *ctinfo
= IP_CT_RELATED
;
722 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
727 skb
->nfct
= &ct
->ct_general
;
728 skb
->nfctinfo
= *ctinfo
;
733 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
737 enum ip_conntrack_info ctinfo
;
738 struct nf_conntrack_l3proto
*l3proto
;
739 struct nf_conntrack_l4proto
*l4proto
;
740 unsigned int dataoff
;
745 /* Previously seen (loopback or untracked)? Ignore. */
747 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
751 /* rcu_read_lock()ed by nf_hook_slow */
752 l3proto
= __nf_ct_l3proto_find(pf
);
753 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
754 &dataoff
, &protonum
);
756 pr_debug("not prepared to track yet or error occured\n");
757 NF_CT_STAT_INC_ATOMIC(net
, error
);
758 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
762 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
764 /* It may be an special packet, error, unclean...
765 * inverse of the return code tells to the netfilter
766 * core what to do with the packet. */
767 if (l4proto
->error
!= NULL
) {
768 ret
= l4proto
->error(net
, skb
, dataoff
, &ctinfo
, pf
, hooknum
);
770 NF_CT_STAT_INC_ATOMIC(net
, error
);
771 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
776 ct
= resolve_normal_ct(net
, skb
, dataoff
, pf
, protonum
,
777 l3proto
, l4proto
, &set_reply
, &ctinfo
);
779 /* Not valid part of a connection */
780 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
785 /* Too stressed to deal. */
786 NF_CT_STAT_INC_ATOMIC(net
, drop
);
790 NF_CT_ASSERT(skb
->nfct
);
792 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
794 /* Invalid: inverse of the return code tells
795 * the netfilter core what to do */
796 pr_debug("nf_conntrack_in: Can't track with proto module\n");
797 nf_conntrack_put(skb
->nfct
);
799 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
801 NF_CT_STAT_INC_ATOMIC(net
, drop
);
805 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
806 nf_conntrack_event_cache(IPCT_STATUS
, ct
);
810 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
812 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
813 const struct nf_conntrack_tuple
*orig
)
818 ret
= nf_ct_invert_tuple(inverse
, orig
,
819 __nf_ct_l3proto_find(orig
->src
.l3num
),
820 __nf_ct_l4proto_find(orig
->src
.l3num
,
821 orig
->dst
.protonum
));
825 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
827 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
828 implicitly racy: see __nf_conntrack_confirm */
829 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
830 const struct nf_conntrack_tuple
*newreply
)
832 struct nf_conn_help
*help
= nfct_help(ct
);
834 /* Should be unconfirmed, so not in hash table yet */
835 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
837 pr_debug("Altering reply tuple of %p to ", ct
);
838 nf_ct_dump_tuple(newreply
);
840 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
841 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
845 __nf_ct_try_assign_helper(ct
, GFP_ATOMIC
);
848 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
850 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
851 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
852 enum ip_conntrack_info ctinfo
,
853 const struct sk_buff
*skb
,
854 unsigned long extra_jiffies
,
857 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
860 /* Only update if this is not a fixed timeout */
861 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
864 /* If not in hash table, timer will not be active yet */
865 if (!nf_ct_is_confirmed(ct
)) {
866 ct
->timeout
.expires
= extra_jiffies
;
868 unsigned long newtime
= jiffies
+ extra_jiffies
;
870 /* Only update the timeout if the new timeout is at least
871 HZ jiffies from the old timeout. Need del_timer for race
872 avoidance (may already be dying). */
873 if (newtime
- ct
->timeout
.expires
>= HZ
)
874 mod_timer_pending(&ct
->timeout
, newtime
);
879 struct nf_conn_counter
*acct
;
881 acct
= nf_conn_acct_find(ct
);
883 spin_lock_bh(&ct
->lock
);
884 acct
[CTINFO2DIR(ctinfo
)].packets
++;
885 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
886 skb
->len
- skb_network_offset(skb
);
887 spin_unlock_bh(&ct
->lock
);
891 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
893 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
894 enum ip_conntrack_info ctinfo
,
895 const struct sk_buff
*skb
,
899 struct nf_conn_counter
*acct
;
901 acct
= nf_conn_acct_find(ct
);
903 spin_lock_bh(&ct
->lock
);
904 acct
[CTINFO2DIR(ctinfo
)].packets
++;
905 acct
[CTINFO2DIR(ctinfo
)].bytes
+=
906 skb
->len
- skb_network_offset(skb
);
907 spin_unlock_bh(&ct
->lock
);
911 if (del_timer(&ct
->timeout
)) {
912 ct
->timeout
.function((unsigned long)ct
);
917 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
919 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
921 #include <linux/netfilter/nfnetlink.h>
922 #include <linux/netfilter/nfnetlink_conntrack.h>
923 #include <linux/mutex.h>
925 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
926 * in ip_conntrack_core, since we don't want the protocols to autoload
927 * or depend on ctnetlink */
928 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
929 const struct nf_conntrack_tuple
*tuple
)
931 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
932 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
938 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
940 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
941 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
942 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
944 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
946 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
947 struct nf_conntrack_tuple
*t
)
949 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
952 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
953 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
957 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
959 int nf_ct_port_nlattr_tuple_size(void)
961 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
963 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
966 /* Used by ipt_REJECT and ip6t_REJECT. */
967 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
970 enum ip_conntrack_info ctinfo
;
972 /* This ICMP is in reverse direction to the packet which caused it */
973 ct
= nf_ct_get(skb
, &ctinfo
);
974 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
975 ctinfo
= IP_CT_RELATED
+ IP_CT_IS_REPLY
;
977 ctinfo
= IP_CT_RELATED
;
979 /* Attach to new skbuff, and increment count */
980 nskb
->nfct
= &ct
->ct_general
;
981 nskb
->nfctinfo
= ctinfo
;
982 nf_conntrack_get(nskb
->nfct
);
985 /* Bring out ya dead! */
986 static struct nf_conn
*
987 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
988 void *data
, unsigned int *bucket
)
990 struct nf_conntrack_tuple_hash
*h
;
992 struct hlist_nulls_node
*n
;
994 spin_lock_bh(&nf_conntrack_lock
);
995 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
996 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
997 ct
= nf_ct_tuplehash_to_ctrack(h
);
1002 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnnode
) {
1003 ct
= nf_ct_tuplehash_to_ctrack(h
);
1005 set_bit(IPS_DYING_BIT
, &ct
->status
);
1007 spin_unlock_bh(&nf_conntrack_lock
);
1010 atomic_inc(&ct
->ct_general
.use
);
1011 spin_unlock_bh(&nf_conntrack_lock
);
1015 void nf_ct_iterate_cleanup(struct net
*net
,
1016 int (*iter
)(struct nf_conn
*i
, void *data
),
1020 unsigned int bucket
= 0;
1022 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1023 /* Time to push up daises... */
1024 if (del_timer(&ct
->timeout
))
1025 death_by_timeout((unsigned long)ct
);
1026 /* ... else the timer will get him soon. */
1031 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1033 struct __nf_ct_flush_report
{
1038 static int kill_report(struct nf_conn
*i
, void *data
)
1040 struct __nf_ct_flush_report
*fr
= (struct __nf_ct_flush_report
*)data
;
1042 /* If we fail to deliver the event, death_by_timeout() will retry */
1043 if (nf_conntrack_event_report(IPCT_DESTROY
, i
,
1044 fr
->pid
, fr
->report
) < 0)
1047 /* Avoid the delivery of the destroy event in death_by_timeout(). */
1048 set_bit(IPS_DYING_BIT
, &i
->status
);
1052 static int kill_all(struct nf_conn
*i
, void *data
)
1057 void nf_ct_free_hashtable(void *hash
, int vmalloced
, unsigned int size
)
1062 free_pages((unsigned long)hash
,
1063 get_order(sizeof(struct hlist_head
) * size
));
1065 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1067 void nf_conntrack_flush_report(struct net
*net
, u32 pid
, int report
)
1069 struct __nf_ct_flush_report fr
= {
1073 nf_ct_iterate_cleanup(net
, kill_report
, &fr
);
1075 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report
);
1077 static void nf_ct_release_dying_list(void)
1079 struct nf_conntrack_tuple_hash
*h
;
1081 struct hlist_nulls_node
*n
;
1083 spin_lock_bh(&nf_conntrack_lock
);
1084 hlist_nulls_for_each_entry(h
, n
, &init_net
.ct
.dying
, hnnode
) {
1085 ct
= nf_ct_tuplehash_to_ctrack(h
);
1086 /* never fails to remove them, no listeners at this point */
1089 spin_unlock_bh(&nf_conntrack_lock
);
1092 static void nf_conntrack_cleanup_init_net(void)
1094 nf_conntrack_helper_fini();
1095 nf_conntrack_proto_fini();
1096 kmem_cache_destroy(nf_conntrack_cachep
);
1099 static void nf_conntrack_cleanup_net(struct net
*net
)
1102 nf_ct_iterate_cleanup(net
, kill_all
, NULL
);
1103 nf_ct_release_dying_list();
1104 if (atomic_read(&net
->ct
.count
) != 0) {
1106 goto i_see_dead_people
;
1108 /* wait until all references to nf_conntrack_untracked are dropped */
1109 while (atomic_read(&nf_conntrack_untracked
.ct_general
.use
) > 1)
1112 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1113 nf_conntrack_htable_size
);
1114 nf_conntrack_ecache_fini(net
);
1115 nf_conntrack_acct_fini(net
);
1116 nf_conntrack_expect_fini(net
);
1117 free_percpu(net
->ct
.stat
);
1120 /* Mishearing the voices in his head, our hero wonders how he's
1121 supposed to kill the mall. */
1122 void nf_conntrack_cleanup(struct net
*net
)
1124 if (net_eq(net
, &init_net
))
1125 rcu_assign_pointer(ip_ct_attach
, NULL
);
1127 /* This makes sure all current packets have passed through
1128 netfilter framework. Roll on, two-stage module
1132 nf_conntrack_cleanup_net(net
);
1134 if (net_eq(net
, &init_net
)) {
1135 rcu_assign_pointer(nf_ct_destroy
, NULL
);
1136 nf_conntrack_cleanup_init_net();
1140 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int *vmalloced
, int nulls
)
1142 struct hlist_nulls_head
*hash
;
1143 unsigned int nr_slots
, i
;
1148 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1149 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1150 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1151 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1155 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1156 hash
= __vmalloc(sz
, GFP_KERNEL
| __GFP_ZERO
, PAGE_KERNEL
);
1160 for (i
= 0; i
< nr_slots
; i
++)
1161 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1165 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1167 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1169 int i
, bucket
, vmalloced
, old_vmalloced
;
1170 unsigned int hashsize
, old_size
;
1172 struct hlist_nulls_head
*hash
, *old_hash
;
1173 struct nf_conntrack_tuple_hash
*h
;
1175 /* On boot, we can set this without any fancy locking. */
1176 if (!nf_conntrack_htable_size
)
1177 return param_set_uint(val
, kp
);
1179 hashsize
= simple_strtoul(val
, NULL
, 0);
1183 hash
= nf_ct_alloc_hashtable(&hashsize
, &vmalloced
, 1);
1187 /* We have to rehahs for the new table anyway, so we also can
1188 * use a newrandom seed */
1189 get_random_bytes(&rnd
, sizeof(rnd
));
1191 /* Lookups in the old hash might happen in parallel, which means we
1192 * might get false negatives during connection lookup. New connections
1193 * created because of a false negative won't make it into the hash
1194 * though since that required taking the lock.
1196 spin_lock_bh(&nf_conntrack_lock
);
1197 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1198 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1199 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1200 struct nf_conntrack_tuple_hash
, hnnode
);
1201 hlist_nulls_del_rcu(&h
->hnnode
);
1202 bucket
= __hash_conntrack(&h
->tuple
, hashsize
, rnd
);
1203 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1206 old_size
= nf_conntrack_htable_size
;
1207 old_vmalloced
= init_net
.ct
.hash_vmalloc
;
1208 old_hash
= init_net
.ct
.hash
;
1210 nf_conntrack_htable_size
= hashsize
;
1211 init_net
.ct
.hash_vmalloc
= vmalloced
;
1212 init_net
.ct
.hash
= hash
;
1213 nf_conntrack_hash_rnd
= rnd
;
1214 spin_unlock_bh(&nf_conntrack_lock
);
1216 nf_ct_free_hashtable(old_hash
, old_vmalloced
, old_size
);
1219 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1221 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1222 &nf_conntrack_htable_size
, 0600);
1224 static int nf_conntrack_init_init_net(void)
1229 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1230 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1231 if (!nf_conntrack_htable_size
) {
1232 nf_conntrack_htable_size
1233 = (((num_physpages
<< PAGE_SHIFT
) / 16384)
1234 / sizeof(struct hlist_head
));
1235 if (num_physpages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1236 nf_conntrack_htable_size
= 16384;
1237 if (nf_conntrack_htable_size
< 32)
1238 nf_conntrack_htable_size
= 32;
1240 /* Use a max. factor of four by default to get the same max as
1241 * with the old struct list_heads. When a table size is given
1242 * we use the old value of 8 to avoid reducing the max.
1246 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1248 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1249 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1252 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1253 sizeof(struct nf_conn
),
1254 0, SLAB_DESTROY_BY_RCU
, NULL
);
1255 if (!nf_conntrack_cachep
) {
1256 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1261 ret
= nf_conntrack_proto_init();
1265 ret
= nf_conntrack_helper_init();
1272 nf_conntrack_proto_fini();
1274 kmem_cache_destroy(nf_conntrack_cachep
);
1280 * We need to use special "null" values, not used in hash table
1282 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1283 #define DYING_NULLS_VAL ((1<<30)+1)
1285 static int nf_conntrack_init_net(struct net
*net
)
1289 atomic_set(&net
->ct
.count
, 0);
1290 INIT_HLIST_NULLS_HEAD(&net
->ct
.unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1291 INIT_HLIST_NULLS_HEAD(&net
->ct
.dying
, DYING_NULLS_VAL
);
1292 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1293 if (!net
->ct
.stat
) {
1297 net
->ct
.hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
,
1298 &net
->ct
.hash_vmalloc
, 1);
1299 if (!net
->ct
.hash
) {
1301 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1304 ret
= nf_conntrack_expect_init(net
);
1307 ret
= nf_conntrack_acct_init(net
);
1310 ret
= nf_conntrack_ecache_init(net
);
1314 /* Set up fake conntrack:
1315 - to never be deleted, not in any hashes */
1316 #ifdef CONFIG_NET_NS
1317 nf_conntrack_untracked
.ct_net
= &init_net
;
1319 atomic_set(&nf_conntrack_untracked
.ct_general
.use
, 1);
1320 /* - and look it like as a confirmed connection */
1321 set_bit(IPS_CONFIRMED_BIT
, &nf_conntrack_untracked
.status
);
1326 nf_conntrack_acct_fini(net
);
1328 nf_conntrack_expect_fini(net
);
1330 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.hash_vmalloc
,
1331 nf_conntrack_htable_size
);
1333 free_percpu(net
->ct
.stat
);
1338 int nf_conntrack_init(struct net
*net
)
1342 if (net_eq(net
, &init_net
)) {
1343 ret
= nf_conntrack_init_init_net();
1347 ret
= nf_conntrack_init_net(net
);
1351 if (net_eq(net
, &init_net
)) {
1352 /* For use by REJECT target */
1353 rcu_assign_pointer(ip_ct_attach
, nf_conntrack_attach
);
1354 rcu_assign_pointer(nf_ct_destroy
, destroy_conntrack
);
1359 if (net_eq(net
, &init_net
))
1360 nf_conntrack_cleanup_init_net();