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>
8 * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
15 #include <linux/types.h>
16 #include <linux/netfilter.h>
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/skbuff.h>
20 #include <linux/proc_fs.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stddef.h>
23 #include <linux/slab.h>
24 #include <linux/random.h>
25 #include <linux/jhash.h>
26 #include <linux/err.h>
27 #include <linux/percpu.h>
28 #include <linux/moduleparam.h>
29 #include <linux/notifier.h>
30 #include <linux/kernel.h>
31 #include <linux/netdevice.h>
32 #include <linux/socket.h>
34 #include <linux/nsproxy.h>
35 #include <linux/rculist_nulls.h>
37 #include <net/netfilter/nf_conntrack.h>
38 #include <net/netfilter/nf_conntrack_l3proto.h>
39 #include <net/netfilter/nf_conntrack_l4proto.h>
40 #include <net/netfilter/nf_conntrack_expect.h>
41 #include <net/netfilter/nf_conntrack_helper.h>
42 #include <net/netfilter/nf_conntrack_core.h>
43 #include <net/netfilter/nf_conntrack_extend.h>
44 #include <net/netfilter/nf_conntrack_acct.h>
45 #include <net/netfilter/nf_conntrack_ecache.h>
46 #include <net/netfilter/nf_conntrack_zones.h>
47 #include <net/netfilter/nf_conntrack_timestamp.h>
48 #include <net/netfilter/nf_conntrack_timeout.h>
49 #include <net/netfilter/nf_conntrack_labels.h>
50 #include <net/netfilter/nf_nat.h>
51 #include <net/netfilter/nf_nat_core.h>
52 #include <net/netfilter/nf_nat_helper.h>
54 #define NF_CONNTRACK_VERSION "0.5.0"
56 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
57 enum nf_nat_manip_type manip
,
58 const struct nlattr
*attr
) __read_mostly
;
59 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
61 int (*nf_nat_seq_adjust_hook
)(struct sk_buff
*skb
,
63 enum ip_conntrack_info ctinfo
,
64 unsigned int protoff
);
65 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook
);
67 DEFINE_SPINLOCK(nf_conntrack_lock
);
68 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
70 unsigned int nf_conntrack_htable_size __read_mostly
;
71 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
73 unsigned int nf_conntrack_max __read_mostly
;
74 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
76 DEFINE_PER_CPU(struct nf_conn
, nf_conntrack_untracked
);
77 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked
);
79 unsigned int nf_conntrack_hash_rnd __read_mostly
;
80 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd
);
82 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
, u16 zone
)
86 /* The direction must be ignored, so we hash everything up to the
87 * destination ports (which is a multiple of 4) and treat the last
88 * three bytes manually.
90 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
91 return jhash2((u32
*)tuple
, n
, zone
^ nf_conntrack_hash_rnd
^
92 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
93 tuple
->dst
.protonum
));
96 static u32
__hash_bucket(u32 hash
, unsigned int size
)
98 return ((u64
)hash
* size
) >> 32;
101 static u32
hash_bucket(u32 hash
, const struct net
*net
)
103 return __hash_bucket(hash
, net
->ct
.htable_size
);
106 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
107 u16 zone
, unsigned int size
)
109 return __hash_bucket(hash_conntrack_raw(tuple
, zone
), size
);
112 static inline u_int32_t
hash_conntrack(const struct net
*net
, u16 zone
,
113 const struct nf_conntrack_tuple
*tuple
)
115 return __hash_conntrack(tuple
, zone
, net
->ct
.htable_size
);
119 nf_ct_get_tuple(const struct sk_buff
*skb
,
121 unsigned int dataoff
,
124 struct nf_conntrack_tuple
*tuple
,
125 const struct nf_conntrack_l3proto
*l3proto
,
126 const struct nf_conntrack_l4proto
*l4proto
)
128 memset(tuple
, 0, sizeof(*tuple
));
130 tuple
->src
.l3num
= l3num
;
131 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
134 tuple
->dst
.protonum
= protonum
;
135 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
137 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
139 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
141 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
142 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
144 struct nf_conntrack_l3proto
*l3proto
;
145 struct nf_conntrack_l4proto
*l4proto
;
146 unsigned int protoff
;
152 l3proto
= __nf_ct_l3proto_find(l3num
);
153 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
154 if (ret
!= NF_ACCEPT
) {
159 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
161 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
167 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
170 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
171 const struct nf_conntrack_tuple
*orig
,
172 const struct nf_conntrack_l3proto
*l3proto
,
173 const struct nf_conntrack_l4proto
*l4proto
)
175 memset(inverse
, 0, sizeof(*inverse
));
177 inverse
->src
.l3num
= orig
->src
.l3num
;
178 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
181 inverse
->dst
.dir
= !orig
->dst
.dir
;
183 inverse
->dst
.protonum
= orig
->dst
.protonum
;
184 return l4proto
->invert_tuple(inverse
, orig
);
186 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
189 clean_from_lists(struct nf_conn
*ct
)
191 pr_debug("clean_from_lists(%p)\n", ct
);
192 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
193 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
195 /* Destroy all pending expectations */
196 nf_ct_remove_expectations(ct
);
200 destroy_conntrack(struct nf_conntrack
*nfct
)
202 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
203 struct net
*net
= nf_ct_net(ct
);
204 struct nf_conntrack_l4proto
*l4proto
;
206 pr_debug("destroy_conntrack(%p)\n", ct
);
207 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
208 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
210 /* To make sure we don't get any weird locking issues here:
211 * destroy_conntrack() MUST NOT be called with a write lock
212 * to nf_conntrack_lock!!! -HW */
214 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
215 if (l4proto
&& l4proto
->destroy
)
216 l4proto
->destroy(ct
);
220 spin_lock_bh(&nf_conntrack_lock
);
221 /* Expectations will have been removed in clean_from_lists,
222 * except TFTP can create an expectation on the first packet,
223 * before connection is in the list, so we need to clean here,
225 nf_ct_remove_expectations(ct
);
227 /* We overload first tuple to link into unconfirmed or dying list.*/
228 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
229 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
231 NF_CT_STAT_INC(net
, delete);
232 spin_unlock_bh(&nf_conntrack_lock
);
235 nf_ct_put(ct
->master
);
237 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
238 nf_conntrack_free(ct
);
241 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
243 struct net
*net
= nf_ct_net(ct
);
245 nf_ct_helper_destroy(ct
);
246 spin_lock_bh(&nf_conntrack_lock
);
247 /* Inside lock so preempt is disabled on module removal path.
248 * Otherwise we can get spurious warnings. */
249 NF_CT_STAT_INC(net
, delete_list
);
250 clean_from_lists(ct
);
251 /* add this conntrack to the dying list */
252 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
254 spin_unlock_bh(&nf_conntrack_lock
);
257 static void death_by_event(unsigned long ul_conntrack
)
259 struct nf_conn
*ct
= (void *)ul_conntrack
;
260 struct net
*net
= nf_ct_net(ct
);
261 struct nf_conntrack_ecache
*ecache
= nf_ct_ecache_find(ct
);
263 BUG_ON(ecache
== NULL
);
265 if (nf_conntrack_event(IPCT_DESTROY
, ct
) < 0) {
266 /* bad luck, let's retry again */
267 ecache
->timeout
.expires
= jiffies
+
268 (prandom_u32() % net
->ct
.sysctl_events_retry_timeout
);
269 add_timer(&ecache
->timeout
);
272 /* we've got the event delivered, now it's dying */
273 set_bit(IPS_DYING_BIT
, &ct
->status
);
277 static void nf_ct_dying_timeout(struct nf_conn
*ct
)
279 struct net
*net
= nf_ct_net(ct
);
280 struct nf_conntrack_ecache
*ecache
= nf_ct_ecache_find(ct
);
282 BUG_ON(ecache
== NULL
);
284 /* set a new timer to retry event delivery */
285 setup_timer(&ecache
->timeout
, death_by_event
, (unsigned long)ct
);
286 ecache
->timeout
.expires
= jiffies
+
287 (prandom_u32() % net
->ct
.sysctl_events_retry_timeout
);
288 add_timer(&ecache
->timeout
);
291 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
293 struct nf_conn_tstamp
*tstamp
;
295 tstamp
= nf_conn_tstamp_find(ct
);
296 if (tstamp
&& tstamp
->stop
== 0)
297 tstamp
->stop
= ktime_to_ns(ktime_get_real());
299 if (!nf_ct_is_dying(ct
) &&
300 unlikely(nf_conntrack_event_report(IPCT_DESTROY
, ct
,
301 portid
, report
) < 0)) {
302 /* destroy event was not delivered */
303 nf_ct_delete_from_lists(ct
);
304 nf_ct_dying_timeout(ct
);
307 set_bit(IPS_DYING_BIT
, &ct
->status
);
308 nf_ct_delete_from_lists(ct
);
312 EXPORT_SYMBOL_GPL(nf_ct_delete
);
314 static void death_by_timeout(unsigned long ul_conntrack
)
316 nf_ct_delete((struct nf_conn
*)ul_conntrack
, 0, 0);
321 * - Caller must take a reference on returned object
322 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
324 * - Caller must lock nf_conntrack_lock before calling this function
326 static struct nf_conntrack_tuple_hash
*
327 ____nf_conntrack_find(struct net
*net
, u16 zone
,
328 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
330 struct nf_conntrack_tuple_hash
*h
;
331 struct hlist_nulls_node
*n
;
332 unsigned int bucket
= hash_bucket(hash
, net
);
334 /* Disable BHs the entire time since we normally need to disable them
335 * at least once for the stats anyway.
339 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[bucket
], hnnode
) {
340 if (nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
341 nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)) == zone
) {
342 NF_CT_STAT_INC(net
, found
);
346 NF_CT_STAT_INC(net
, searched
);
349 * if the nulls value we got at the end of this lookup is
350 * not the expected one, we must restart lookup.
351 * We probably met an item that was moved to another chain.
353 if (get_nulls_value(n
) != bucket
) {
354 NF_CT_STAT_INC(net
, search_restart
);
362 struct nf_conntrack_tuple_hash
*
363 __nf_conntrack_find(struct net
*net
, u16 zone
,
364 const struct nf_conntrack_tuple
*tuple
)
366 return ____nf_conntrack_find(net
, zone
, tuple
,
367 hash_conntrack_raw(tuple
, zone
));
369 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
371 /* Find a connection corresponding to a tuple. */
372 static struct nf_conntrack_tuple_hash
*
373 __nf_conntrack_find_get(struct net
*net
, u16 zone
,
374 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
376 struct nf_conntrack_tuple_hash
*h
;
381 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
383 ct
= nf_ct_tuplehash_to_ctrack(h
);
384 if (unlikely(nf_ct_is_dying(ct
) ||
385 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
388 if (unlikely(!nf_ct_tuple_equal(tuple
, &h
->tuple
) ||
389 nf_ct_zone(ct
) != zone
)) {
400 struct nf_conntrack_tuple_hash
*
401 nf_conntrack_find_get(struct net
*net
, u16 zone
,
402 const struct nf_conntrack_tuple
*tuple
)
404 return __nf_conntrack_find_get(net
, zone
, tuple
,
405 hash_conntrack_raw(tuple
, zone
));
407 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
409 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
411 unsigned int repl_hash
)
413 struct net
*net
= nf_ct_net(ct
);
415 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
416 &net
->ct
.hash
[hash
]);
417 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
418 &net
->ct
.hash
[repl_hash
]);
422 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
424 struct net
*net
= nf_ct_net(ct
);
425 unsigned int hash
, repl_hash
;
426 struct nf_conntrack_tuple_hash
*h
;
427 struct hlist_nulls_node
*n
;
430 zone
= nf_ct_zone(ct
);
431 hash
= hash_conntrack(net
, zone
,
432 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
433 repl_hash
= hash_conntrack(net
, zone
,
434 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
436 spin_lock_bh(&nf_conntrack_lock
);
438 /* See if there's one in the list already, including reverse */
439 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
440 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
442 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
444 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnnode
)
445 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
447 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
450 add_timer(&ct
->timeout
);
451 nf_conntrack_get(&ct
->ct_general
);
452 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
453 NF_CT_STAT_INC(net
, insert
);
454 spin_unlock_bh(&nf_conntrack_lock
);
459 NF_CT_STAT_INC(net
, insert_failed
);
460 spin_unlock_bh(&nf_conntrack_lock
);
463 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
465 /* Confirm a connection given skb; places it in hash table */
467 __nf_conntrack_confirm(struct sk_buff
*skb
)
469 unsigned int hash
, repl_hash
;
470 struct nf_conntrack_tuple_hash
*h
;
472 struct nf_conn_help
*help
;
473 struct nf_conn_tstamp
*tstamp
;
474 struct hlist_nulls_node
*n
;
475 enum ip_conntrack_info ctinfo
;
479 ct
= nf_ct_get(skb
, &ctinfo
);
482 /* ipt_REJECT uses nf_conntrack_attach to attach related
483 ICMP/TCP RST packets in other direction. Actual packet
484 which created connection will be IP_CT_NEW or for an
485 expected connection, IP_CT_RELATED. */
486 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
489 zone
= nf_ct_zone(ct
);
490 /* reuse the hash saved before */
491 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
492 hash
= hash_bucket(hash
, net
);
493 repl_hash
= hash_conntrack(net
, zone
,
494 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
496 /* We're not in hash table, and we refuse to set up related
497 connections for unconfirmed conns. But packet copies and
498 REJECT will give spurious warnings here. */
499 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
501 /* No external references means no one else could have
503 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
504 pr_debug("Confirming conntrack %p\n", ct
);
506 spin_lock_bh(&nf_conntrack_lock
);
508 /* We have to check the DYING flag inside the lock to prevent
509 a race against nf_ct_get_next_corpse() possibly called from
510 user context, else we insert an already 'dead' hash, blocking
511 further use of that particular connection -JM */
513 if (unlikely(nf_ct_is_dying(ct
))) {
514 spin_unlock_bh(&nf_conntrack_lock
);
518 /* See if there's one in the list already, including reverse:
519 NAT could have grabbed it without realizing, since we're
520 not in the hash. If there is, we lost race. */
521 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
522 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
524 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
526 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnnode
)
527 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
529 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
532 /* Remove from unconfirmed list */
533 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
535 /* Timer relative to confirmation time, not original
536 setting time, otherwise we'd get timer wrap in
537 weird delay cases. */
538 ct
->timeout
.expires
+= jiffies
;
539 add_timer(&ct
->timeout
);
540 atomic_inc(&ct
->ct_general
.use
);
541 ct
->status
|= IPS_CONFIRMED
;
543 /* set conntrack timestamp, if enabled. */
544 tstamp
= nf_conn_tstamp_find(ct
);
546 if (skb
->tstamp
.tv64
== 0)
547 __net_timestamp(skb
);
549 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
551 /* Since the lookup is lockless, hash insertion must be done after
552 * starting the timer and setting the CONFIRMED bit. The RCU barriers
553 * guarantee that no other CPU can find the conntrack before the above
554 * stores are visible.
556 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
557 NF_CT_STAT_INC(net
, insert
);
558 spin_unlock_bh(&nf_conntrack_lock
);
560 help
= nfct_help(ct
);
561 if (help
&& help
->helper
)
562 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
564 nf_conntrack_event_cache(master_ct(ct
) ?
565 IPCT_RELATED
: IPCT_NEW
, ct
);
569 NF_CT_STAT_INC(net
, insert_failed
);
570 spin_unlock_bh(&nf_conntrack_lock
);
573 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
575 /* Returns true if a connection correspondings to the tuple (required
578 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
579 const struct nf_conn
*ignored_conntrack
)
581 struct net
*net
= nf_ct_net(ignored_conntrack
);
582 struct nf_conntrack_tuple_hash
*h
;
583 struct hlist_nulls_node
*n
;
585 u16 zone
= nf_ct_zone(ignored_conntrack
);
586 unsigned int hash
= hash_conntrack(net
, zone
, tuple
);
588 /* Disable BHs the entire time since we need to disable them at
589 * least once for the stats anyway.
592 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
593 ct
= nf_ct_tuplehash_to_ctrack(h
);
594 if (ct
!= ignored_conntrack
&&
595 nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
596 nf_ct_zone(ct
) == zone
) {
597 NF_CT_STAT_INC(net
, found
);
598 rcu_read_unlock_bh();
601 NF_CT_STAT_INC(net
, searched
);
603 rcu_read_unlock_bh();
607 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
609 #define NF_CT_EVICTION_RANGE 8
611 /* There's a small race here where we may free a just-assured
612 connection. Too bad: we're in trouble anyway. */
613 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
615 /* Use oldest entry, which is roughly LRU */
616 struct nf_conntrack_tuple_hash
*h
;
617 struct nf_conn
*ct
= NULL
, *tmp
;
618 struct hlist_nulls_node
*n
;
619 unsigned int i
, cnt
= 0;
623 for (i
= 0; i
< net
->ct
.htable_size
; i
++) {
624 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
626 tmp
= nf_ct_tuplehash_to_ctrack(h
);
627 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
633 if (likely(!nf_ct_is_dying(ct
) &&
634 atomic_inc_not_zero(&ct
->ct_general
.use
)))
640 if (cnt
>= NF_CT_EVICTION_RANGE
)
643 hash
= (hash
+ 1) % net
->ct
.htable_size
;
650 if (del_timer(&ct
->timeout
)) {
651 if (nf_ct_delete(ct
, 0, 0)) {
653 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
660 void init_nf_conntrack_hash_rnd(void)
665 * Why not initialize nf_conntrack_rnd in a "init()" function ?
666 * Because there isn't enough entropy when system initializing,
667 * and we initialize it as late as possible.
670 get_random_bytes(&rand
, sizeof(rand
));
672 cmpxchg(&nf_conntrack_hash_rnd
, 0, rand
);
675 static struct nf_conn
*
676 __nf_conntrack_alloc(struct net
*net
, u16 zone
,
677 const struct nf_conntrack_tuple
*orig
,
678 const struct nf_conntrack_tuple
*repl
,
683 if (unlikely(!nf_conntrack_hash_rnd
)) {
684 init_nf_conntrack_hash_rnd();
685 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
686 hash
= hash_conntrack_raw(orig
, zone
);
689 /* We don't want any race condition at early drop stage */
690 atomic_inc(&net
->ct
.count
);
692 if (nf_conntrack_max
&&
693 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
694 if (!early_drop(net
, hash_bucket(hash
, net
))) {
695 atomic_dec(&net
->ct
.count
);
696 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
697 return ERR_PTR(-ENOMEM
);
702 * Do not use kmem_cache_zalloc(), as this cache uses
703 * SLAB_DESTROY_BY_RCU.
705 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
707 atomic_dec(&net
->ct
.count
);
708 return ERR_PTR(-ENOMEM
);
711 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
712 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
714 memset(&ct
->tuplehash
[IP_CT_DIR_MAX
], 0,
715 offsetof(struct nf_conn
, proto
) -
716 offsetof(struct nf_conn
, tuplehash
[IP_CT_DIR_MAX
]));
717 spin_lock_init(&ct
->lock
);
718 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
719 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
720 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
721 /* save hash for reusing when confirming */
722 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
723 /* Don't set timer yet: wait for confirmation */
724 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
725 write_pnet(&ct
->ct_net
, net
);
726 #ifdef CONFIG_NF_CONNTRACK_ZONES
728 struct nf_conntrack_zone
*nf_ct_zone
;
730 nf_ct_zone
= nf_ct_ext_add(ct
, NF_CT_EXT_ZONE
, GFP_ATOMIC
);
733 nf_ct_zone
->id
= zone
;
737 * changes to lookup keys must be done before setting refcnt to 1
740 atomic_set(&ct
->ct_general
.use
, 1);
743 #ifdef CONFIG_NF_CONNTRACK_ZONES
745 atomic_dec(&net
->ct
.count
);
746 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
747 return ERR_PTR(-ENOMEM
);
751 struct nf_conn
*nf_conntrack_alloc(struct net
*net
, u16 zone
,
752 const struct nf_conntrack_tuple
*orig
,
753 const struct nf_conntrack_tuple
*repl
,
756 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
758 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
760 void nf_conntrack_free(struct nf_conn
*ct
)
762 struct net
*net
= nf_ct_net(ct
);
764 nf_ct_ext_destroy(ct
);
765 atomic_dec(&net
->ct
.count
);
767 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
769 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
772 /* Allocate a new conntrack: we return -ENOMEM if classification
773 failed due to stress. Otherwise it really is unclassifiable. */
774 static struct nf_conntrack_tuple_hash
*
775 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
776 const struct nf_conntrack_tuple
*tuple
,
777 struct nf_conntrack_l3proto
*l3proto
,
778 struct nf_conntrack_l4proto
*l4proto
,
780 unsigned int dataoff
, u32 hash
)
783 struct nf_conn_help
*help
;
784 struct nf_conntrack_tuple repl_tuple
;
785 struct nf_conntrack_ecache
*ecache
;
786 struct nf_conntrack_expect
*exp
;
787 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
788 struct nf_conn_timeout
*timeout_ext
;
789 unsigned int *timeouts
;
791 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
792 pr_debug("Can't invert tuple.\n");
796 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
799 return (struct nf_conntrack_tuple_hash
*)ct
;
801 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
803 timeouts
= NF_CT_TIMEOUT_EXT_DATA(timeout_ext
);
805 timeouts
= l4proto
->get_timeouts(net
);
807 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
808 nf_conntrack_free(ct
);
809 pr_debug("init conntrack: can't track with proto module\n");
814 nf_ct_timeout_ext_add(ct
, timeout_ext
->timeout
, GFP_ATOMIC
);
816 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
817 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
818 nf_ct_labels_ext_add(ct
);
820 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
821 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
822 ecache
? ecache
->expmask
: 0,
825 spin_lock_bh(&nf_conntrack_lock
);
826 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
828 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
830 /* Welcome, Mr. Bond. We've been expecting you... */
831 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
832 ct
->master
= exp
->master
;
834 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
837 rcu_assign_pointer(help
->helper
, exp
->helper
);
840 #ifdef CONFIG_NF_CONNTRACK_MARK
841 ct
->mark
= exp
->master
->mark
;
843 #ifdef CONFIG_NF_CONNTRACK_SECMARK
844 ct
->secmark
= exp
->master
->secmark
;
846 nf_conntrack_get(&ct
->master
->ct_general
);
847 NF_CT_STAT_INC(net
, expect_new
);
849 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
850 NF_CT_STAT_INC(net
, new);
853 /* Overload tuple linked list to put us in unconfirmed list. */
854 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
855 &net
->ct
.unconfirmed
);
857 spin_unlock_bh(&nf_conntrack_lock
);
861 exp
->expectfn(ct
, exp
);
862 nf_ct_expect_put(exp
);
865 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
868 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
869 static inline struct nf_conn
*
870 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
872 unsigned int dataoff
,
875 struct nf_conntrack_l3proto
*l3proto
,
876 struct nf_conntrack_l4proto
*l4proto
,
878 enum ip_conntrack_info
*ctinfo
)
880 struct nf_conntrack_tuple tuple
;
881 struct nf_conntrack_tuple_hash
*h
;
883 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
886 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
887 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
889 pr_debug("resolve_normal_ct: Can't get tuple\n");
893 /* look for tuple match */
894 hash
= hash_conntrack_raw(&tuple
, zone
);
895 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
897 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
904 ct
= nf_ct_tuplehash_to_ctrack(h
);
906 /* It exists; we have (non-exclusive) reference. */
907 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
908 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
909 /* Please set reply bit if this packet OK */
912 /* Once we've had two way comms, always ESTABLISHED. */
913 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
914 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
915 *ctinfo
= IP_CT_ESTABLISHED
;
916 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
917 pr_debug("nf_conntrack_in: related packet for %p\n",
919 *ctinfo
= IP_CT_RELATED
;
921 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
926 skb
->nfct
= &ct
->ct_general
;
927 skb
->nfctinfo
= *ctinfo
;
932 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
935 struct nf_conn
*ct
, *tmpl
= NULL
;
936 enum ip_conntrack_info ctinfo
;
937 struct nf_conntrack_l3proto
*l3proto
;
938 struct nf_conntrack_l4proto
*l4proto
;
939 unsigned int *timeouts
;
940 unsigned int dataoff
;
946 /* Previously seen (loopback or untracked)? Ignore. */
947 tmpl
= (struct nf_conn
*)skb
->nfct
;
948 if (!nf_ct_is_template(tmpl
)) {
949 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
955 /* rcu_read_lock()ed by nf_hook_slow */
956 l3proto
= __nf_ct_l3proto_find(pf
);
957 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
958 &dataoff
, &protonum
);
960 pr_debug("not prepared to track yet or error occurred\n");
961 NF_CT_STAT_INC_ATOMIC(net
, error
);
962 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
967 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
969 /* It may be an special packet, error, unclean...
970 * inverse of the return code tells to the netfilter
971 * core what to do with the packet. */
972 if (l4proto
->error
!= NULL
) {
973 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
976 NF_CT_STAT_INC_ATOMIC(net
, error
);
977 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
981 /* ICMP[v6] protocol trackers may assign one conntrack. */
986 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
987 l3proto
, l4proto
, &set_reply
, &ctinfo
);
989 /* Not valid part of a connection */
990 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
996 /* Too stressed to deal. */
997 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1002 NF_CT_ASSERT(skb
->nfct
);
1004 /* Decide what timeout policy we want to apply to this flow. */
1005 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1007 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1009 /* Invalid: inverse of the return code tells
1010 * the netfilter core what to do */
1011 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1012 nf_conntrack_put(skb
->nfct
);
1014 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1015 if (ret
== -NF_DROP
)
1016 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1021 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1022 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1025 /* Special case: we have to repeat this hook, assign the
1026 * template again to this packet. We assume that this packet
1027 * has no conntrack assigned. This is used by nf_ct_tcp. */
1028 if (ret
== NF_REPEAT
)
1029 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
1036 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1038 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1039 const struct nf_conntrack_tuple
*orig
)
1044 ret
= nf_ct_invert_tuple(inverse
, orig
,
1045 __nf_ct_l3proto_find(orig
->src
.l3num
),
1046 __nf_ct_l4proto_find(orig
->src
.l3num
,
1047 orig
->dst
.protonum
));
1051 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1053 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1054 implicitly racy: see __nf_conntrack_confirm */
1055 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1056 const struct nf_conntrack_tuple
*newreply
)
1058 struct nf_conn_help
*help
= nfct_help(ct
);
1060 /* Should be unconfirmed, so not in hash table yet */
1061 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1063 pr_debug("Altering reply tuple of %p to ", ct
);
1064 nf_ct_dump_tuple(newreply
);
1066 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1067 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1071 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1074 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1076 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1077 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1078 enum ip_conntrack_info ctinfo
,
1079 const struct sk_buff
*skb
,
1080 unsigned long extra_jiffies
,
1083 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1086 /* Only update if this is not a fixed timeout */
1087 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1090 /* If not in hash table, timer will not be active yet */
1091 if (!nf_ct_is_confirmed(ct
)) {
1092 ct
->timeout
.expires
= extra_jiffies
;
1094 unsigned long newtime
= jiffies
+ extra_jiffies
;
1096 /* Only update the timeout if the new timeout is at least
1097 HZ jiffies from the old timeout. Need del_timer for race
1098 avoidance (may already be dying). */
1099 if (newtime
- ct
->timeout
.expires
>= HZ
)
1100 mod_timer_pending(&ct
->timeout
, newtime
);
1105 struct nf_conn_counter
*acct
;
1107 acct
= nf_conn_acct_find(ct
);
1109 atomic64_inc(&acct
[CTINFO2DIR(ctinfo
)].packets
);
1110 atomic64_add(skb
->len
, &acct
[CTINFO2DIR(ctinfo
)].bytes
);
1114 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1116 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1117 enum ip_conntrack_info ctinfo
,
1118 const struct sk_buff
*skb
,
1122 struct nf_conn_counter
*acct
;
1124 acct
= nf_conn_acct_find(ct
);
1126 atomic64_inc(&acct
[CTINFO2DIR(ctinfo
)].packets
);
1127 atomic64_add(skb
->len
- skb_network_offset(skb
),
1128 &acct
[CTINFO2DIR(ctinfo
)].bytes
);
1132 if (del_timer(&ct
->timeout
)) {
1133 ct
->timeout
.function((unsigned long)ct
);
1138 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1140 #ifdef CONFIG_NF_CONNTRACK_ZONES
1141 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1142 .len
= sizeof(struct nf_conntrack_zone
),
1143 .align
= __alignof__(struct nf_conntrack_zone
),
1144 .id
= NF_CT_EXT_ZONE
,
1148 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1150 #include <linux/netfilter/nfnetlink.h>
1151 #include <linux/netfilter/nfnetlink_conntrack.h>
1152 #include <linux/mutex.h>
1154 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1155 * in ip_conntrack_core, since we don't want the protocols to autoload
1156 * or depend on ctnetlink */
1157 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1158 const struct nf_conntrack_tuple
*tuple
)
1160 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1161 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1162 goto nla_put_failure
;
1168 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1170 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1171 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1172 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1174 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1176 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1177 struct nf_conntrack_tuple
*t
)
1179 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1182 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1183 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1187 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1189 int nf_ct_port_nlattr_tuple_size(void)
1191 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1193 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1196 /* Used by ipt_REJECT and ip6t_REJECT. */
1197 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1200 enum ip_conntrack_info ctinfo
;
1202 /* This ICMP is in reverse direction to the packet which caused it */
1203 ct
= nf_ct_get(skb
, &ctinfo
);
1204 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1205 ctinfo
= IP_CT_RELATED_REPLY
;
1207 ctinfo
= IP_CT_RELATED
;
1209 /* Attach to new skbuff, and increment count */
1210 nskb
->nfct
= &ct
->ct_general
;
1211 nskb
->nfctinfo
= ctinfo
;
1212 nf_conntrack_get(nskb
->nfct
);
1215 /* Bring out ya dead! */
1216 static struct nf_conn
*
1217 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1218 void *data
, unsigned int *bucket
)
1220 struct nf_conntrack_tuple_hash
*h
;
1222 struct hlist_nulls_node
*n
;
1224 spin_lock_bh(&nf_conntrack_lock
);
1225 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1226 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1227 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1229 ct
= nf_ct_tuplehash_to_ctrack(h
);
1234 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnnode
) {
1235 ct
= nf_ct_tuplehash_to_ctrack(h
);
1237 set_bit(IPS_DYING_BIT
, &ct
->status
);
1239 spin_unlock_bh(&nf_conntrack_lock
);
1242 atomic_inc(&ct
->ct_general
.use
);
1243 spin_unlock_bh(&nf_conntrack_lock
);
1247 void nf_ct_iterate_cleanup(struct net
*net
,
1248 int (*iter
)(struct nf_conn
*i
, void *data
),
1249 void *data
, u32 portid
, int report
)
1252 unsigned int bucket
= 0;
1254 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1255 /* Time to push up daises... */
1256 if (del_timer(&ct
->timeout
))
1257 nf_ct_delete(ct
, portid
, report
);
1259 /* ... else the timer will get him soon. */
1264 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1266 static int kill_all(struct nf_conn
*i
, void *data
)
1271 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1273 if (is_vmalloc_addr(hash
))
1276 free_pages((unsigned long)hash
,
1277 get_order(sizeof(struct hlist_head
) * size
));
1279 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1281 void nf_conntrack_flush_report(struct net
*net
, u32 portid
, int report
)
1283 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, portid
, report
);
1285 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report
);
1287 static void nf_ct_release_dying_list(struct net
*net
)
1289 struct nf_conntrack_tuple_hash
*h
;
1291 struct hlist_nulls_node
*n
;
1293 spin_lock_bh(&nf_conntrack_lock
);
1294 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.dying
, hnnode
) {
1295 ct
= nf_ct_tuplehash_to_ctrack(h
);
1296 /* never fails to remove them, no listeners at this point */
1299 spin_unlock_bh(&nf_conntrack_lock
);
1302 static int untrack_refs(void)
1306 for_each_possible_cpu(cpu
) {
1307 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1309 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1314 void nf_conntrack_cleanup_start(void)
1316 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1319 void nf_conntrack_cleanup_end(void)
1321 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1322 while (untrack_refs() > 0)
1325 #ifdef CONFIG_NF_CONNTRACK_ZONES
1326 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1328 nf_conntrack_proto_fini();
1329 nf_conntrack_labels_fini();
1330 nf_conntrack_helper_fini();
1331 nf_conntrack_timeout_fini();
1332 nf_conntrack_ecache_fini();
1333 nf_conntrack_tstamp_fini();
1334 nf_conntrack_acct_fini();
1335 nf_conntrack_expect_fini();
1339 * Mishearing the voices in his head, our hero wonders how he's
1340 * supposed to kill the mall.
1342 void nf_conntrack_cleanup_net(struct net
*net
)
1346 list_add(&net
->exit_list
, &single
);
1347 nf_conntrack_cleanup_net_list(&single
);
1350 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1356 * This makes sure all current packets have passed through
1357 * netfilter framework. Roll on, two-stage module
1363 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1364 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1365 nf_ct_release_dying_list(net
);
1366 if (atomic_read(&net
->ct
.count
) != 0)
1371 goto i_see_dead_people
;
1374 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1375 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1376 nf_conntrack_proto_pernet_fini(net
);
1377 nf_conntrack_helper_pernet_fini(net
);
1378 nf_conntrack_ecache_pernet_fini(net
);
1379 nf_conntrack_tstamp_pernet_fini(net
);
1380 nf_conntrack_acct_pernet_fini(net
);
1381 nf_conntrack_expect_pernet_fini(net
);
1382 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1383 kfree(net
->ct
.slabname
);
1384 free_percpu(net
->ct
.stat
);
1388 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1390 struct hlist_nulls_head
*hash
;
1391 unsigned int nr_slots
, i
;
1394 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1395 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1396 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1397 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1400 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1405 for (i
= 0; i
< nr_slots
; i
++)
1406 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1410 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1412 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1415 unsigned int hashsize
, old_size
;
1416 struct hlist_nulls_head
*hash
, *old_hash
;
1417 struct nf_conntrack_tuple_hash
*h
;
1420 if (current
->nsproxy
->net_ns
!= &init_net
)
1423 /* On boot, we can set this without any fancy locking. */
1424 if (!nf_conntrack_htable_size
)
1425 return param_set_uint(val
, kp
);
1427 rc
= kstrtouint(val
, 0, &hashsize
);
1433 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1437 /* Lookups in the old hash might happen in parallel, which means we
1438 * might get false negatives during connection lookup. New connections
1439 * created because of a false negative won't make it into the hash
1440 * though since that required taking the lock.
1442 spin_lock_bh(&nf_conntrack_lock
);
1443 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1444 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1445 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1446 struct nf_conntrack_tuple_hash
, hnnode
);
1447 ct
= nf_ct_tuplehash_to_ctrack(h
);
1448 hlist_nulls_del_rcu(&h
->hnnode
);
1449 bucket
= __hash_conntrack(&h
->tuple
, nf_ct_zone(ct
),
1451 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1454 old_size
= init_net
.ct
.htable_size
;
1455 old_hash
= init_net
.ct
.hash
;
1457 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1458 init_net
.ct
.hash
= hash
;
1459 spin_unlock_bh(&nf_conntrack_lock
);
1461 nf_ct_free_hashtable(old_hash
, old_size
);
1464 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1466 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1467 &nf_conntrack_htable_size
, 0600);
1469 void nf_ct_untracked_status_or(unsigned long bits
)
1473 for_each_possible_cpu(cpu
)
1474 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1476 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1478 int nf_conntrack_init_start(void)
1483 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1484 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1485 if (!nf_conntrack_htable_size
) {
1486 nf_conntrack_htable_size
1487 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1488 / sizeof(struct hlist_head
));
1489 if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1490 nf_conntrack_htable_size
= 16384;
1491 if (nf_conntrack_htable_size
< 32)
1492 nf_conntrack_htable_size
= 32;
1494 /* Use a max. factor of four by default to get the same max as
1495 * with the old struct list_heads. When a table size is given
1496 * we use the old value of 8 to avoid reducing the max.
1500 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1502 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1503 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1506 ret
= nf_conntrack_expect_init();
1510 ret
= nf_conntrack_acct_init();
1514 ret
= nf_conntrack_tstamp_init();
1518 ret
= nf_conntrack_ecache_init();
1522 ret
= nf_conntrack_timeout_init();
1526 ret
= nf_conntrack_helper_init();
1530 ret
= nf_conntrack_labels_init();
1534 #ifdef CONFIG_NF_CONNTRACK_ZONES
1535 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1539 ret
= nf_conntrack_proto_init();
1543 /* Set up fake conntrack: to never be deleted, not in any hashes */
1544 for_each_possible_cpu(cpu
) {
1545 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1546 write_pnet(&ct
->ct_net
, &init_net
);
1547 atomic_set(&ct
->ct_general
.use
, 1);
1549 /* - and look it like as a confirmed connection */
1550 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1554 #ifdef CONFIG_NF_CONNTRACK_ZONES
1555 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1558 nf_conntrack_labels_fini();
1560 nf_conntrack_helper_fini();
1562 nf_conntrack_timeout_fini();
1564 nf_conntrack_ecache_fini();
1566 nf_conntrack_tstamp_fini();
1568 nf_conntrack_acct_fini();
1570 nf_conntrack_expect_fini();
1575 void nf_conntrack_init_end(void)
1577 /* For use by REJECT target */
1578 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1579 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1581 /* Howto get NAT offsets */
1582 RCU_INIT_POINTER(nf_ct_nat_offset
, NULL
);
1586 * We need to use special "null" values, not used in hash table
1588 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1589 #define DYING_NULLS_VAL ((1<<30)+1)
1590 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1592 int nf_conntrack_init_net(struct net
*net
)
1596 atomic_set(&net
->ct
.count
, 0);
1597 INIT_HLIST_NULLS_HEAD(&net
->ct
.unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1598 INIT_HLIST_NULLS_HEAD(&net
->ct
.dying
, DYING_NULLS_VAL
);
1599 INIT_HLIST_NULLS_HEAD(&net
->ct
.tmpl
, TEMPLATE_NULLS_VAL
);
1600 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1601 if (!net
->ct
.stat
) {
1606 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1607 if (!net
->ct
.slabname
) {
1612 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1613 sizeof(struct nf_conn
), 0,
1614 SLAB_DESTROY_BY_RCU
, NULL
);
1615 if (!net
->ct
.nf_conntrack_cachep
) {
1616 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1621 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1622 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
, 1);
1623 if (!net
->ct
.hash
) {
1625 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1628 ret
= nf_conntrack_expect_pernet_init(net
);
1631 ret
= nf_conntrack_acct_pernet_init(net
);
1634 ret
= nf_conntrack_tstamp_pernet_init(net
);
1637 ret
= nf_conntrack_ecache_pernet_init(net
);
1640 ret
= nf_conntrack_helper_pernet_init(net
);
1643 ret
= nf_conntrack_proto_pernet_init(net
);
1649 nf_conntrack_helper_pernet_fini(net
);
1651 nf_conntrack_ecache_pernet_fini(net
);
1653 nf_conntrack_tstamp_pernet_fini(net
);
1655 nf_conntrack_acct_pernet_fini(net
);
1657 nf_conntrack_expect_pernet_fini(net
);
1659 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1661 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1663 kfree(net
->ct
.slabname
);
1665 free_percpu(net
->ct
.stat
);
1670 s32 (*nf_ct_nat_offset
)(const struct nf_conn
*ct
,
1671 enum ip_conntrack_dir dir
,
1673 EXPORT_SYMBOL_GPL(nf_ct_nat_offset
);