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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/types.h>
18 #include <linux/netfilter.h>
19 #include <linux/module.h>
20 #include <linux/sched.h>
21 #include <linux/skbuff.h>
22 #include <linux/proc_fs.h>
23 #include <linux/vmalloc.h>
24 #include <linux/stddef.h>
25 #include <linux/slab.h>
26 #include <linux/random.h>
27 #include <linux/jhash.h>
28 #include <linux/err.h>
29 #include <linux/percpu.h>
30 #include <linux/moduleparam.h>
31 #include <linux/notifier.h>
32 #include <linux/kernel.h>
33 #include <linux/netdevice.h>
34 #include <linux/socket.h>
36 #include <linux/nsproxy.h>
37 #include <linux/rculist_nulls.h>
39 #include <net/netfilter/nf_conntrack.h>
40 #include <net/netfilter/nf_conntrack_l3proto.h>
41 #include <net/netfilter/nf_conntrack_l4proto.h>
42 #include <net/netfilter/nf_conntrack_expect.h>
43 #include <net/netfilter/nf_conntrack_helper.h>
44 #include <net/netfilter/nf_conntrack_seqadj.h>
45 #include <net/netfilter/nf_conntrack_core.h>
46 #include <net/netfilter/nf_conntrack_extend.h>
47 #include <net/netfilter/nf_conntrack_acct.h>
48 #include <net/netfilter/nf_conntrack_ecache.h>
49 #include <net/netfilter/nf_conntrack_zones.h>
50 #include <net/netfilter/nf_conntrack_timestamp.h>
51 #include <net/netfilter/nf_conntrack_timeout.h>
52 #include <net/netfilter/nf_conntrack_labels.h>
53 #include <net/netfilter/nf_conntrack_synproxy.h>
54 #include <net/netfilter/nf_nat.h>
55 #include <net/netfilter/nf_nat_core.h>
56 #include <net/netfilter/nf_nat_helper.h>
57 #include <net/netns/hash.h>
59 #include "nf_internals.h"
61 #define NF_CONNTRACK_VERSION "0.5.0"
63 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
64 enum nf_nat_manip_type manip
,
65 const struct nlattr
*attr
) __read_mostly
;
66 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
68 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks
[CONNTRACK_LOCKS
];
69 EXPORT_SYMBOL_GPL(nf_conntrack_locks
);
71 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(nf_conntrack_expect_lock
);
72 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock
);
74 struct hlist_nulls_head
*nf_conntrack_hash __read_mostly
;
75 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
77 struct conntrack_gc_work
{
78 struct delayed_work dwork
;
85 static __read_mostly
struct kmem_cache
*nf_conntrack_cachep
;
86 static __read_mostly spinlock_t nf_conntrack_locks_all_lock
;
87 static __read_mostly
DEFINE_SPINLOCK(nf_conntrack_locks_all_lock
);
88 static __read_mostly
bool nf_conntrack_locks_all
;
90 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
91 #define GC_MAX_BUCKETS_DIV 128u
92 /* upper bound of full table scan */
93 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
94 /* desired ratio of entries found to be expired */
95 #define GC_EVICT_RATIO 50u
97 static struct conntrack_gc_work conntrack_gc_work
;
99 void nf_conntrack_lock(spinlock_t
*lock
) __acquires(lock
)
101 /* 1) Acquire the lock */
104 /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
105 * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
107 if (likely(smp_load_acquire(&nf_conntrack_locks_all
) == false))
110 /* fast path failed, unlock */
113 /* Slow path 1) get global lock */
114 spin_lock(&nf_conntrack_locks_all_lock
);
116 /* Slow path 2) get the lock we want */
119 /* Slow path 3) release the global lock */
120 spin_unlock(&nf_conntrack_locks_all_lock
);
122 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
124 static void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
126 h1
%= CONNTRACK_LOCKS
;
127 h2
%= CONNTRACK_LOCKS
;
128 spin_unlock(&nf_conntrack_locks
[h1
]);
130 spin_unlock(&nf_conntrack_locks
[h2
]);
133 /* return true if we need to recompute hashes (in case hash table was resized) */
134 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
135 unsigned int h2
, unsigned int sequence
)
137 h1
%= CONNTRACK_LOCKS
;
138 h2
%= CONNTRACK_LOCKS
;
140 nf_conntrack_lock(&nf_conntrack_locks
[h1
]);
142 spin_lock_nested(&nf_conntrack_locks
[h2
],
143 SINGLE_DEPTH_NESTING
);
145 nf_conntrack_lock(&nf_conntrack_locks
[h2
]);
146 spin_lock_nested(&nf_conntrack_locks
[h1
],
147 SINGLE_DEPTH_NESTING
);
149 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
150 nf_conntrack_double_unlock(h1
, h2
);
156 static void nf_conntrack_all_lock(void)
160 spin_lock(&nf_conntrack_locks_all_lock
);
162 nf_conntrack_locks_all
= true;
164 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++) {
165 spin_lock(&nf_conntrack_locks
[i
]);
167 /* This spin_unlock provides the "release" to ensure that
168 * nf_conntrack_locks_all==true is visible to everyone that
169 * acquired spin_lock(&nf_conntrack_locks[]).
171 spin_unlock(&nf_conntrack_locks
[i
]);
175 static void nf_conntrack_all_unlock(void)
177 /* All prior stores must be complete before we clear
178 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
179 * might observe the false value but not the entire
181 * It pairs with the smp_load_acquire() in nf_conntrack_lock()
183 smp_store_release(&nf_conntrack_locks_all
, false);
184 spin_unlock(&nf_conntrack_locks_all_lock
);
187 unsigned int nf_conntrack_htable_size __read_mostly
;
188 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
190 unsigned int nf_conntrack_max __read_mostly
;
191 seqcount_t nf_conntrack_generation __read_mostly
;
192 static unsigned int nf_conntrack_hash_rnd __read_mostly
;
194 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
,
195 const struct net
*net
)
200 get_random_once(&nf_conntrack_hash_rnd
, sizeof(nf_conntrack_hash_rnd
));
202 /* The direction must be ignored, so we hash everything up to the
203 * destination ports (which is a multiple of 4) and treat the last
204 * three bytes manually.
206 seed
= nf_conntrack_hash_rnd
^ net_hash_mix(net
);
207 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
208 return jhash2((u32
*)tuple
, n
, seed
^
209 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
210 tuple
->dst
.protonum
));
213 static u32
scale_hash(u32 hash
)
215 return reciprocal_scale(hash
, nf_conntrack_htable_size
);
218 static u32
__hash_conntrack(const struct net
*net
,
219 const struct nf_conntrack_tuple
*tuple
,
222 return reciprocal_scale(hash_conntrack_raw(tuple
, net
), size
);
225 static u32
hash_conntrack(const struct net
*net
,
226 const struct nf_conntrack_tuple
*tuple
)
228 return scale_hash(hash_conntrack_raw(tuple
, net
));
232 nf_ct_get_tuple(const struct sk_buff
*skb
,
234 unsigned int dataoff
,
238 struct nf_conntrack_tuple
*tuple
,
239 const struct nf_conntrack_l3proto
*l3proto
,
240 const struct nf_conntrack_l4proto
*l4proto
)
242 memset(tuple
, 0, sizeof(*tuple
));
244 tuple
->src
.l3num
= l3num
;
245 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
248 tuple
->dst
.protonum
= protonum
;
249 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
251 return l4proto
->pkt_to_tuple(skb
, dataoff
, net
, tuple
);
253 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
255 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
257 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
259 const struct nf_conntrack_l3proto
*l3proto
;
260 const struct nf_conntrack_l4proto
*l4proto
;
261 unsigned int protoff
;
267 l3proto
= __nf_ct_l3proto_find(l3num
);
268 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
269 if (ret
!= NF_ACCEPT
) {
274 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
276 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
,
282 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
285 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
286 const struct nf_conntrack_tuple
*orig
,
287 const struct nf_conntrack_l3proto
*l3proto
,
288 const struct nf_conntrack_l4proto
*l4proto
)
290 memset(inverse
, 0, sizeof(*inverse
));
292 inverse
->src
.l3num
= orig
->src
.l3num
;
293 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
296 inverse
->dst
.dir
= !orig
->dst
.dir
;
298 inverse
->dst
.protonum
= orig
->dst
.protonum
;
299 return l4proto
->invert_tuple(inverse
, orig
);
301 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
304 clean_from_lists(struct nf_conn
*ct
)
306 pr_debug("clean_from_lists(%p)\n", ct
);
307 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
308 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
310 /* Destroy all pending expectations */
311 nf_ct_remove_expectations(ct
);
314 /* must be called with local_bh_disable */
315 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
317 struct ct_pcpu
*pcpu
;
319 /* add this conntrack to the (per cpu) dying list */
320 ct
->cpu
= smp_processor_id();
321 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
323 spin_lock(&pcpu
->lock
);
324 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
326 spin_unlock(&pcpu
->lock
);
329 /* must be called with local_bh_disable */
330 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
332 struct ct_pcpu
*pcpu
;
334 /* add this conntrack to the (per cpu) unconfirmed list */
335 ct
->cpu
= smp_processor_id();
336 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
338 spin_lock(&pcpu
->lock
);
339 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
341 spin_unlock(&pcpu
->lock
);
344 /* must be called with local_bh_disable */
345 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
347 struct ct_pcpu
*pcpu
;
349 /* We overload first tuple to link into unconfirmed or dying list.*/
350 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
352 spin_lock(&pcpu
->lock
);
353 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
354 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
355 spin_unlock(&pcpu
->lock
);
358 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
360 /* Released via destroy_conntrack() */
361 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
362 const struct nf_conntrack_zone
*zone
,
365 struct nf_conn
*tmpl
, *p
;
367 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
) {
368 tmpl
= kzalloc(sizeof(*tmpl
) + NFCT_INFOMASK
, flags
);
373 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
375 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
376 tmpl
->proto
.tmpl_padto
= (char *)tmpl
- (char *)p
;
379 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
384 tmpl
->status
= IPS_TEMPLATE
;
385 write_pnet(&tmpl
->ct_net
, net
);
386 nf_ct_zone_add(tmpl
, zone
);
387 atomic_set(&tmpl
->ct_general
.use
, 0);
391 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
393 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
395 nf_ct_ext_destroy(tmpl
);
396 nf_ct_ext_free(tmpl
);
398 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
)
399 kfree((char *)tmpl
- tmpl
->proto
.tmpl_padto
);
403 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
406 destroy_conntrack(struct nf_conntrack
*nfct
)
408 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
409 const struct nf_conntrack_l4proto
*l4proto
;
411 pr_debug("destroy_conntrack(%p)\n", ct
);
412 WARN_ON(atomic_read(&nfct
->use
) != 0);
414 if (unlikely(nf_ct_is_template(ct
))) {
418 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
419 if (l4proto
->destroy
)
420 l4proto
->destroy(ct
);
423 /* Expectations will have been removed in clean_from_lists,
424 * except TFTP can create an expectation on the first packet,
425 * before connection is in the list, so we need to clean here,
428 nf_ct_remove_expectations(ct
);
430 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
435 nf_ct_put(ct
->master
);
437 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
438 nf_conntrack_free(ct
);
441 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
443 struct net
*net
= nf_ct_net(ct
);
444 unsigned int hash
, reply_hash
;
445 unsigned int sequence
;
447 nf_ct_helper_destroy(ct
);
451 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
452 hash
= hash_conntrack(net
,
453 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
454 reply_hash
= hash_conntrack(net
,
455 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
456 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
458 clean_from_lists(ct
);
459 nf_conntrack_double_unlock(hash
, reply_hash
);
461 nf_ct_add_to_dying_list(ct
);
466 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
468 struct nf_conn_tstamp
*tstamp
;
470 if (test_and_set_bit(IPS_DYING_BIT
, &ct
->status
))
473 tstamp
= nf_conn_tstamp_find(ct
);
474 if (tstamp
&& tstamp
->stop
== 0)
475 tstamp
->stop
= ktime_get_real_ns();
477 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
478 portid
, report
) < 0) {
479 /* destroy event was not delivered. nf_ct_put will
480 * be done by event cache worker on redelivery.
482 nf_ct_delete_from_lists(ct
);
483 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
487 nf_conntrack_ecache_work(nf_ct_net(ct
));
488 nf_ct_delete_from_lists(ct
);
492 EXPORT_SYMBOL_GPL(nf_ct_delete
);
495 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
496 const struct nf_conntrack_tuple
*tuple
,
497 const struct nf_conntrack_zone
*zone
,
498 const struct net
*net
)
500 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
502 /* A conntrack can be recreated with the equal tuple,
503 * so we need to check that the conntrack is confirmed
505 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
506 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
507 nf_ct_is_confirmed(ct
) &&
508 net_eq(net
, nf_ct_net(ct
));
511 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
512 static void nf_ct_gc_expired(struct nf_conn
*ct
)
514 if (!atomic_inc_not_zero(&ct
->ct_general
.use
))
517 if (nf_ct_should_gc(ct
))
525 * - Caller must take a reference on returned object
526 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
528 static struct nf_conntrack_tuple_hash
*
529 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
530 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
532 struct nf_conntrack_tuple_hash
*h
;
533 struct hlist_nulls_head
*ct_hash
;
534 struct hlist_nulls_node
*n
;
535 unsigned int bucket
, hsize
;
538 nf_conntrack_get_ht(&ct_hash
, &hsize
);
539 bucket
= reciprocal_scale(hash
, hsize
);
541 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[bucket
], hnnode
) {
544 ct
= nf_ct_tuplehash_to_ctrack(h
);
545 if (nf_ct_is_expired(ct
)) {
546 nf_ct_gc_expired(ct
);
550 if (nf_ct_is_dying(ct
))
553 if (nf_ct_key_equal(h
, tuple
, zone
, net
))
557 * if the nulls value we got at the end of this lookup is
558 * not the expected one, we must restart lookup.
559 * We probably met an item that was moved to another chain.
561 if (get_nulls_value(n
) != bucket
) {
562 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
569 /* Find a connection corresponding to a tuple. */
570 static struct nf_conntrack_tuple_hash
*
571 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
572 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
574 struct nf_conntrack_tuple_hash
*h
;
579 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
581 ct
= nf_ct_tuplehash_to_ctrack(h
);
582 if (unlikely(nf_ct_is_dying(ct
) ||
583 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
586 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
, net
))) {
597 struct nf_conntrack_tuple_hash
*
598 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
599 const struct nf_conntrack_tuple
*tuple
)
601 return __nf_conntrack_find_get(net
, zone
, tuple
,
602 hash_conntrack_raw(tuple
, net
));
604 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
606 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
608 unsigned int reply_hash
)
610 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
611 &nf_conntrack_hash
[hash
]);
612 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
613 &nf_conntrack_hash
[reply_hash
]);
617 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
619 const struct nf_conntrack_zone
*zone
;
620 struct net
*net
= nf_ct_net(ct
);
621 unsigned int hash
, reply_hash
;
622 struct nf_conntrack_tuple_hash
*h
;
623 struct hlist_nulls_node
*n
;
624 unsigned int sequence
;
626 zone
= nf_ct_zone(ct
);
630 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
631 hash
= hash_conntrack(net
,
632 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
633 reply_hash
= hash_conntrack(net
,
634 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
635 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
637 /* See if there's one in the list already, including reverse */
638 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
639 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
643 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
644 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
649 /* The caller holds a reference to this object */
650 atomic_set(&ct
->ct_general
.use
, 2);
651 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
652 nf_conntrack_double_unlock(hash
, reply_hash
);
653 NF_CT_STAT_INC(net
, insert
);
658 nf_conntrack_double_unlock(hash
, reply_hash
);
659 NF_CT_STAT_INC(net
, insert_failed
);
663 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
665 static inline void nf_ct_acct_update(struct nf_conn
*ct
,
666 enum ip_conntrack_info ctinfo
,
669 struct nf_conn_acct
*acct
;
671 acct
= nf_conn_acct_find(ct
);
673 struct nf_conn_counter
*counter
= acct
->counter
;
675 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
676 atomic64_add(len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
680 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
681 const struct nf_conn
*loser_ct
)
683 struct nf_conn_acct
*acct
;
685 acct
= nf_conn_acct_find(loser_ct
);
687 struct nf_conn_counter
*counter
= acct
->counter
;
690 /* u32 should be fine since we must have seen one packet. */
691 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
692 nf_ct_acct_update(ct
, ctinfo
, bytes
);
696 /* Resolve race on insertion if this protocol allows this. */
697 static int nf_ct_resolve_clash(struct net
*net
, struct sk_buff
*skb
,
698 enum ip_conntrack_info ctinfo
,
699 struct nf_conntrack_tuple_hash
*h
)
701 /* This is the conntrack entry already in hashes that won race. */
702 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
703 const struct nf_conntrack_l4proto
*l4proto
;
705 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
706 if (l4proto
->allow_clash
&&
707 ((ct
->status
& IPS_NAT_DONE_MASK
) == 0) &&
708 !nf_ct_is_dying(ct
) &&
709 atomic_inc_not_zero(&ct
->ct_general
.use
)) {
710 enum ip_conntrack_info oldinfo
;
711 struct nf_conn
*loser_ct
= nf_ct_get(skb
, &oldinfo
);
713 nf_ct_acct_merge(ct
, ctinfo
, loser_ct
);
714 nf_conntrack_put(&loser_ct
->ct_general
);
715 nf_ct_set(skb
, ct
, oldinfo
);
718 NF_CT_STAT_INC(net
, drop
);
722 /* Confirm a connection given skb; places it in hash table */
724 __nf_conntrack_confirm(struct sk_buff
*skb
)
726 const struct nf_conntrack_zone
*zone
;
727 unsigned int hash
, reply_hash
;
728 struct nf_conntrack_tuple_hash
*h
;
730 struct nf_conn_help
*help
;
731 struct nf_conn_tstamp
*tstamp
;
732 struct hlist_nulls_node
*n
;
733 enum ip_conntrack_info ctinfo
;
735 unsigned int sequence
;
738 ct
= nf_ct_get(skb
, &ctinfo
);
741 /* ipt_REJECT uses nf_conntrack_attach to attach related
742 ICMP/TCP RST packets in other direction. Actual packet
743 which created connection will be IP_CT_NEW or for an
744 expected connection, IP_CT_RELATED. */
745 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
748 zone
= nf_ct_zone(ct
);
752 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
753 /* reuse the hash saved before */
754 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
755 hash
= scale_hash(hash
);
756 reply_hash
= hash_conntrack(net
,
757 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
759 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
761 /* We're not in hash table, and we refuse to set up related
762 * connections for unconfirmed conns. But packet copies and
763 * REJECT will give spurious warnings here.
766 /* Another skb with the same unconfirmed conntrack may
767 * win the race. This may happen for bridge(br_flood)
768 * or broadcast/multicast packets do skb_clone with
769 * unconfirmed conntrack.
771 if (unlikely(nf_ct_is_confirmed(ct
))) {
773 nf_conntrack_double_unlock(hash
, reply_hash
);
778 pr_debug("Confirming conntrack %p\n", ct
);
779 /* We have to check the DYING flag after unlink to prevent
780 * a race against nf_ct_get_next_corpse() possibly called from
781 * user context, else we insert an already 'dead' hash, blocking
782 * further use of that particular connection -JM.
784 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
786 if (unlikely(nf_ct_is_dying(ct
))) {
787 nf_ct_add_to_dying_list(ct
);
791 /* See if there's one in the list already, including reverse:
792 NAT could have grabbed it without realizing, since we're
793 not in the hash. If there is, we lost race. */
794 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
795 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
799 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
800 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
804 /* Timer relative to confirmation time, not original
805 setting time, otherwise we'd get timer wrap in
806 weird delay cases. */
807 ct
->timeout
+= nfct_time_stamp
;
808 atomic_inc(&ct
->ct_general
.use
);
809 ct
->status
|= IPS_CONFIRMED
;
811 /* set conntrack timestamp, if enabled. */
812 tstamp
= nf_conn_tstamp_find(ct
);
814 if (skb
->tstamp
== 0)
815 __net_timestamp(skb
);
817 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
819 /* Since the lookup is lockless, hash insertion must be done after
820 * starting the timer and setting the CONFIRMED bit. The RCU barriers
821 * guarantee that no other CPU can find the conntrack before the above
822 * stores are visible.
824 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
825 nf_conntrack_double_unlock(hash
, reply_hash
);
828 help
= nfct_help(ct
);
829 if (help
&& help
->helper
)
830 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
832 nf_conntrack_event_cache(master_ct(ct
) ?
833 IPCT_RELATED
: IPCT_NEW
, ct
);
837 nf_ct_add_to_dying_list(ct
);
838 ret
= nf_ct_resolve_clash(net
, skb
, ctinfo
, h
);
840 nf_conntrack_double_unlock(hash
, reply_hash
);
841 NF_CT_STAT_INC(net
, insert_failed
);
845 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
847 /* Returns true if a connection correspondings to the tuple (required
850 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
851 const struct nf_conn
*ignored_conntrack
)
853 struct net
*net
= nf_ct_net(ignored_conntrack
);
854 const struct nf_conntrack_zone
*zone
;
855 struct nf_conntrack_tuple_hash
*h
;
856 struct hlist_nulls_head
*ct_hash
;
857 unsigned int hash
, hsize
;
858 struct hlist_nulls_node
*n
;
861 zone
= nf_ct_zone(ignored_conntrack
);
865 nf_conntrack_get_ht(&ct_hash
, &hsize
);
866 hash
= __hash_conntrack(net
, tuple
, hsize
);
868 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
869 ct
= nf_ct_tuplehash_to_ctrack(h
);
871 if (ct
== ignored_conntrack
)
874 if (nf_ct_is_expired(ct
)) {
875 nf_ct_gc_expired(ct
);
879 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
880 /* Tuple is taken already, so caller will need to find
881 * a new source port to use.
884 * If the *original tuples* are identical, then both
885 * conntracks refer to the same flow.
886 * This is a rare situation, it can occur e.g. when
887 * more than one UDP packet is sent from same socket
888 * in different threads.
890 * Let nf_ct_resolve_clash() deal with this later.
892 if (nf_ct_tuple_equal(&ignored_conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
893 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
))
896 NF_CT_STAT_INC_ATOMIC(net
, found
);
902 if (get_nulls_value(n
) != hash
) {
903 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
911 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
913 #define NF_CT_EVICTION_RANGE 8
915 /* There's a small race here where we may free a just-assured
916 connection. Too bad: we're in trouble anyway. */
917 static unsigned int early_drop_list(struct net
*net
,
918 struct hlist_nulls_head
*head
)
920 struct nf_conntrack_tuple_hash
*h
;
921 struct hlist_nulls_node
*n
;
922 unsigned int drops
= 0;
925 hlist_nulls_for_each_entry_rcu(h
, n
, head
, hnnode
) {
926 tmp
= nf_ct_tuplehash_to_ctrack(h
);
928 if (nf_ct_is_expired(tmp
)) {
929 nf_ct_gc_expired(tmp
);
933 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
934 !net_eq(nf_ct_net(tmp
), net
) ||
938 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
941 /* kill only if still in same netns -- might have moved due to
942 * SLAB_TYPESAFE_BY_RCU rules.
944 * We steal the timer reference. If that fails timer has
945 * already fired or someone else deleted it. Just drop ref
946 * and move to next entry.
948 if (net_eq(nf_ct_net(tmp
), net
) &&
949 nf_ct_is_confirmed(tmp
) &&
950 nf_ct_delete(tmp
, 0, 0))
959 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
961 unsigned int i
, bucket
;
963 for (i
= 0; i
< NF_CT_EVICTION_RANGE
; i
++) {
964 struct hlist_nulls_head
*ct_hash
;
965 unsigned int hsize
, drops
;
968 nf_conntrack_get_ht(&ct_hash
, &hsize
);
970 bucket
= reciprocal_scale(hash
, hsize
);
972 bucket
= (bucket
+ 1) % hsize
;
974 drops
= early_drop_list(net
, &ct_hash
[bucket
]);
978 NF_CT_STAT_ADD_ATOMIC(net
, early_drop
, drops
);
986 static bool gc_worker_skip_ct(const struct nf_conn
*ct
)
988 return !nf_ct_is_confirmed(ct
) || nf_ct_is_dying(ct
);
991 static bool gc_worker_can_early_drop(const struct nf_conn
*ct
)
993 const struct nf_conntrack_l4proto
*l4proto
;
995 if (!test_bit(IPS_ASSURED_BIT
, &ct
->status
))
998 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
999 if (l4proto
->can_early_drop
&& l4proto
->can_early_drop(ct
))
1005 static void gc_worker(struct work_struct
*work
)
1007 unsigned int min_interval
= max(HZ
/ GC_MAX_BUCKETS_DIV
, 1u);
1008 unsigned int i
, goal
, buckets
= 0, expired_count
= 0;
1009 unsigned int nf_conntrack_max95
= 0;
1010 struct conntrack_gc_work
*gc_work
;
1011 unsigned int ratio
, scanned
= 0;
1012 unsigned long next_run
;
1014 gc_work
= container_of(work
, struct conntrack_gc_work
, dwork
.work
);
1016 goal
= nf_conntrack_htable_size
/ GC_MAX_BUCKETS_DIV
;
1017 i
= gc_work
->last_bucket
;
1018 if (gc_work
->early_drop
)
1019 nf_conntrack_max95
= nf_conntrack_max
/ 100u * 95u;
1022 struct nf_conntrack_tuple_hash
*h
;
1023 struct hlist_nulls_head
*ct_hash
;
1024 struct hlist_nulls_node
*n
;
1025 unsigned int hashsz
;
1026 struct nf_conn
*tmp
;
1031 nf_conntrack_get_ht(&ct_hash
, &hashsz
);
1035 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[i
], hnnode
) {
1038 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1041 if (nf_ct_is_expired(tmp
)) {
1042 nf_ct_gc_expired(tmp
);
1047 if (nf_conntrack_max95
== 0 || gc_worker_skip_ct(tmp
))
1050 net
= nf_ct_net(tmp
);
1051 if (atomic_read(&net
->ct
.count
) < nf_conntrack_max95
)
1054 /* need to take reference to avoid possible races */
1055 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1058 if (gc_worker_skip_ct(tmp
)) {
1063 if (gc_worker_can_early_drop(tmp
))
1069 /* could check get_nulls_value() here and restart if ct
1070 * was moved to another chain. But given gc is best-effort
1071 * we will just continue with next hash slot.
1074 cond_resched_rcu_qs();
1075 } while (++buckets
< goal
);
1077 if (gc_work
->exiting
)
1081 * Eviction will normally happen from the packet path, and not
1082 * from this gc worker.
1084 * This worker is only here to reap expired entries when system went
1085 * idle after a busy period.
1087 * The heuristics below are supposed to balance conflicting goals:
1089 * 1. Minimize time until we notice a stale entry
1090 * 2. Maximize scan intervals to not waste cycles
1092 * Normally, expire ratio will be close to 0.
1094 * As soon as a sizeable fraction of the entries have expired
1095 * increase scan frequency.
1097 ratio
= scanned
? expired_count
* 100 / scanned
: 0;
1098 if (ratio
> GC_EVICT_RATIO
) {
1099 gc_work
->next_gc_run
= min_interval
;
1101 unsigned int max
= GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
;
1103 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
) == 0);
1105 gc_work
->next_gc_run
+= min_interval
;
1106 if (gc_work
->next_gc_run
> max
)
1107 gc_work
->next_gc_run
= max
;
1110 next_run
= gc_work
->next_gc_run
;
1111 gc_work
->last_bucket
= i
;
1112 gc_work
->early_drop
= false;
1113 queue_delayed_work(system_long_wq
, &gc_work
->dwork
, next_run
);
1116 static void conntrack_gc_work_init(struct conntrack_gc_work
*gc_work
)
1118 INIT_DEFERRABLE_WORK(&gc_work
->dwork
, gc_worker
);
1119 gc_work
->next_gc_run
= HZ
;
1120 gc_work
->exiting
= false;
1123 static struct nf_conn
*
1124 __nf_conntrack_alloc(struct net
*net
,
1125 const struct nf_conntrack_zone
*zone
,
1126 const struct nf_conntrack_tuple
*orig
,
1127 const struct nf_conntrack_tuple
*repl
,
1128 gfp_t gfp
, u32 hash
)
1132 /* We don't want any race condition at early drop stage */
1133 atomic_inc(&net
->ct
.count
);
1135 if (nf_conntrack_max
&&
1136 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
1137 if (!early_drop(net
, hash
)) {
1138 if (!conntrack_gc_work
.early_drop
)
1139 conntrack_gc_work
.early_drop
= true;
1140 atomic_dec(&net
->ct
.count
);
1141 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1142 return ERR_PTR(-ENOMEM
);
1147 * Do not use kmem_cache_zalloc(), as this cache uses
1148 * SLAB_TYPESAFE_BY_RCU.
1150 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
1154 spin_lock_init(&ct
->lock
);
1155 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
1156 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
1157 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
1158 /* save hash for reusing when confirming */
1159 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
1161 write_pnet(&ct
->ct_net
, net
);
1162 memset(&ct
->__nfct_init_offset
[0], 0,
1163 offsetof(struct nf_conn
, proto
) -
1164 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
1166 nf_ct_zone_add(ct
, zone
);
1168 /* Because we use RCU lookups, we set ct_general.use to zero before
1169 * this is inserted in any list.
1171 atomic_set(&ct
->ct_general
.use
, 0);
1174 atomic_dec(&net
->ct
.count
);
1175 return ERR_PTR(-ENOMEM
);
1178 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
1179 const struct nf_conntrack_zone
*zone
,
1180 const struct nf_conntrack_tuple
*orig
,
1181 const struct nf_conntrack_tuple
*repl
,
1184 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
1186 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
1188 void nf_conntrack_free(struct nf_conn
*ct
)
1190 struct net
*net
= nf_ct_net(ct
);
1192 /* A freed object has refcnt == 0, that's
1193 * the golden rule for SLAB_TYPESAFE_BY_RCU
1195 WARN_ON(atomic_read(&ct
->ct_general
.use
) != 0);
1197 nf_ct_ext_destroy(ct
);
1199 kmem_cache_free(nf_conntrack_cachep
, ct
);
1200 smp_mb__before_atomic();
1201 atomic_dec(&net
->ct
.count
);
1203 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
1206 /* Allocate a new conntrack: we return -ENOMEM if classification
1207 failed due to stress. Otherwise it really is unclassifiable. */
1208 static noinline
struct nf_conntrack_tuple_hash
*
1209 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
1210 const struct nf_conntrack_tuple
*tuple
,
1211 const struct nf_conntrack_l3proto
*l3proto
,
1212 const struct nf_conntrack_l4proto
*l4proto
,
1213 struct sk_buff
*skb
,
1214 unsigned int dataoff
, u32 hash
)
1217 struct nf_conn_help
*help
;
1218 struct nf_conntrack_tuple repl_tuple
;
1219 struct nf_conntrack_ecache
*ecache
;
1220 struct nf_conntrack_expect
*exp
= NULL
;
1221 const struct nf_conntrack_zone
*zone
;
1222 struct nf_conn_timeout
*timeout_ext
;
1223 struct nf_conntrack_zone tmp
;
1224 unsigned int *timeouts
;
1226 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
1227 pr_debug("Can't invert tuple.\n");
1231 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1232 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
1235 return (struct nf_conntrack_tuple_hash
*)ct
;
1237 if (!nf_ct_add_synproxy(ct
, tmpl
)) {
1238 nf_conntrack_free(ct
);
1239 return ERR_PTR(-ENOMEM
);
1242 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1244 timeouts
= nf_ct_timeout_data(timeout_ext
);
1245 if (unlikely(!timeouts
))
1246 timeouts
= l4proto
->get_timeouts(net
);
1248 timeouts
= l4proto
->get_timeouts(net
);
1251 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
1252 nf_conntrack_free(ct
);
1253 pr_debug("can't track with proto module\n");
1258 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1261 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1262 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1263 nf_ct_labels_ext_add(ct
);
1265 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1266 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1267 ecache
? ecache
->expmask
: 0,
1271 if (net
->ct
.expect_count
) {
1272 spin_lock(&nf_conntrack_expect_lock
);
1273 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1275 pr_debug("expectation arrives ct=%p exp=%p\n",
1277 /* Welcome, Mr. Bond. We've been expecting you... */
1278 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1279 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1280 ct
->master
= exp
->master
;
1282 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
1285 rcu_assign_pointer(help
->helper
, exp
->helper
);
1288 #ifdef CONFIG_NF_CONNTRACK_MARK
1289 ct
->mark
= exp
->master
->mark
;
1291 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1292 ct
->secmark
= exp
->master
->secmark
;
1294 NF_CT_STAT_INC(net
, expect_new
);
1296 spin_unlock(&nf_conntrack_expect_lock
);
1299 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1301 /* Now it is inserted into the unconfirmed list, bump refcount */
1302 nf_conntrack_get(&ct
->ct_general
);
1303 nf_ct_add_to_unconfirmed_list(ct
);
1309 exp
->expectfn(ct
, exp
);
1310 nf_ct_expect_put(exp
);
1313 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1316 /* On success, returns 0, sets skb->_nfct | ctinfo */
1318 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1319 struct sk_buff
*skb
,
1320 unsigned int dataoff
,
1323 const struct nf_conntrack_l3proto
*l3proto
,
1324 const struct nf_conntrack_l4proto
*l4proto
)
1326 const struct nf_conntrack_zone
*zone
;
1327 struct nf_conntrack_tuple tuple
;
1328 struct nf_conntrack_tuple_hash
*h
;
1329 enum ip_conntrack_info ctinfo
;
1330 struct nf_conntrack_zone tmp
;
1334 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1335 dataoff
, l3num
, protonum
, net
, &tuple
, l3proto
,
1337 pr_debug("Can't get tuple\n");
1341 /* look for tuple match */
1342 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1343 hash
= hash_conntrack_raw(&tuple
, net
);
1344 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1346 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1347 skb
, dataoff
, hash
);
1353 ct
= nf_ct_tuplehash_to_ctrack(h
);
1355 /* It exists; we have (non-exclusive) reference. */
1356 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1357 ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1359 /* Once we've had two way comms, always ESTABLISHED. */
1360 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1361 pr_debug("normal packet for %p\n", ct
);
1362 ctinfo
= IP_CT_ESTABLISHED
;
1363 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1364 pr_debug("related packet for %p\n", ct
);
1365 ctinfo
= IP_CT_RELATED
;
1367 pr_debug("new packet for %p\n", ct
);
1371 nf_ct_set(skb
, ct
, ctinfo
);
1376 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1377 struct sk_buff
*skb
)
1379 const struct nf_conntrack_l3proto
*l3proto
;
1380 const struct nf_conntrack_l4proto
*l4proto
;
1381 struct nf_conn
*ct
, *tmpl
;
1382 enum ip_conntrack_info ctinfo
;
1383 unsigned int *timeouts
;
1384 unsigned int dataoff
;
1388 tmpl
= nf_ct_get(skb
, &ctinfo
);
1389 if (tmpl
|| ctinfo
== IP_CT_UNTRACKED
) {
1390 /* Previously seen (loopback or untracked)? Ignore. */
1391 if ((tmpl
&& !nf_ct_is_template(tmpl
)) ||
1392 ctinfo
== IP_CT_UNTRACKED
) {
1393 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1399 /* rcu_read_lock()ed by nf_hook_thresh */
1400 l3proto
= __nf_ct_l3proto_find(pf
);
1401 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1402 &dataoff
, &protonum
);
1404 pr_debug("not prepared to track yet or error occurred\n");
1405 NF_CT_STAT_INC_ATOMIC(net
, error
);
1406 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1411 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1413 /* It may be an special packet, error, unclean...
1414 * inverse of the return code tells to the netfilter
1415 * core what to do with the packet. */
1416 if (l4proto
->error
!= NULL
) {
1417 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, pf
, hooknum
);
1419 NF_CT_STAT_INC_ATOMIC(net
, error
);
1420 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1424 /* ICMP[v6] protocol trackers may assign one conntrack. */
1429 ret
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1432 /* Too stressed to deal. */
1433 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1438 ct
= nf_ct_get(skb
, &ctinfo
);
1440 /* Not valid part of a connection */
1441 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1446 /* Decide what timeout policy we want to apply to this flow. */
1447 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1449 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, timeouts
);
1451 /* Invalid: inverse of the return code tells
1452 * the netfilter core what to do */
1453 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1454 nf_conntrack_put(&ct
->ct_general
);
1456 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1457 if (ret
== -NF_DROP
)
1458 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1459 /* Special case: TCP tracker reports an attempt to reopen a
1460 * closed/aborted connection. We have to go back and create a
1463 if (ret
== -NF_REPEAT
)
1469 if (ctinfo
== IP_CT_ESTABLISHED_REPLY
&&
1470 !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1471 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1478 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1480 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1481 const struct nf_conntrack_tuple
*orig
)
1486 ret
= nf_ct_invert_tuple(inverse
, orig
,
1487 __nf_ct_l3proto_find(orig
->src
.l3num
),
1488 __nf_ct_l4proto_find(orig
->src
.l3num
,
1489 orig
->dst
.protonum
));
1493 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1495 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1496 implicitly racy: see __nf_conntrack_confirm */
1497 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1498 const struct nf_conntrack_tuple
*newreply
)
1500 struct nf_conn_help
*help
= nfct_help(ct
);
1502 /* Should be unconfirmed, so not in hash table yet */
1503 WARN_ON(nf_ct_is_confirmed(ct
));
1505 pr_debug("Altering reply tuple of %p to ", ct
);
1506 nf_ct_dump_tuple(newreply
);
1508 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1509 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1513 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1516 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1518 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1519 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1520 enum ip_conntrack_info ctinfo
,
1521 const struct sk_buff
*skb
,
1522 unsigned long extra_jiffies
,
1527 /* Only update if this is not a fixed timeout */
1528 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1531 /* If not in hash table, timer will not be active yet */
1532 if (nf_ct_is_confirmed(ct
))
1533 extra_jiffies
+= nfct_time_stamp
;
1535 ct
->timeout
= extra_jiffies
;
1538 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1540 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1542 bool nf_ct_kill_acct(struct nf_conn
*ct
,
1543 enum ip_conntrack_info ctinfo
,
1544 const struct sk_buff
*skb
)
1546 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1548 return nf_ct_delete(ct
, 0, 0);
1550 EXPORT_SYMBOL_GPL(nf_ct_kill_acct
);
1552 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1554 #include <linux/netfilter/nfnetlink.h>
1555 #include <linux/netfilter/nfnetlink_conntrack.h>
1556 #include <linux/mutex.h>
1558 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1559 * in ip_conntrack_core, since we don't want the protocols to autoload
1560 * or depend on ctnetlink */
1561 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1562 const struct nf_conntrack_tuple
*tuple
)
1564 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1565 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1566 goto nla_put_failure
;
1572 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1574 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1575 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1576 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1578 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1580 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1581 struct nf_conntrack_tuple
*t
)
1583 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1586 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1587 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1591 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1593 int nf_ct_port_nlattr_tuple_size(void)
1595 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1597 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1600 /* Used by ipt_REJECT and ip6t_REJECT. */
1601 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1604 enum ip_conntrack_info ctinfo
;
1606 /* This ICMP is in reverse direction to the packet which caused it */
1607 ct
= nf_ct_get(skb
, &ctinfo
);
1608 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1609 ctinfo
= IP_CT_RELATED_REPLY
;
1611 ctinfo
= IP_CT_RELATED
;
1613 /* Attach to new skbuff, and increment count */
1614 nf_ct_set(nskb
, ct
, ctinfo
);
1615 nf_conntrack_get(skb_nfct(nskb
));
1618 /* Bring out ya dead! */
1619 static struct nf_conn
*
1620 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
1621 void *data
, unsigned int *bucket
)
1623 struct nf_conntrack_tuple_hash
*h
;
1625 struct hlist_nulls_node
*n
;
1628 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1629 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1631 nf_conntrack_lock(lockp
);
1632 if (*bucket
< nf_conntrack_htable_size
) {
1633 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
1634 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1636 ct
= nf_ct_tuplehash_to_ctrack(h
);
1648 atomic_inc(&ct
->ct_general
.use
);
1654 static void nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
),
1655 void *data
, u32 portid
, int report
)
1657 unsigned int bucket
= 0, sequence
;
1663 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
1665 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
1666 /* Time to push up daises... */
1668 nf_ct_delete(ct
, portid
, report
);
1673 if (!read_seqcount_retry(&nf_conntrack_generation
, sequence
))
1680 int (*iter
)(struct nf_conn
*i
, void *data
);
1685 static int iter_net_only(struct nf_conn
*i
, void *data
)
1687 struct iter_data
*d
= data
;
1689 if (!net_eq(d
->net
, nf_ct_net(i
)))
1692 return d
->iter(i
, d
->data
);
1696 __nf_ct_unconfirmed_destroy(struct net
*net
)
1700 for_each_possible_cpu(cpu
) {
1701 struct nf_conntrack_tuple_hash
*h
;
1702 struct hlist_nulls_node
*n
;
1703 struct ct_pcpu
*pcpu
;
1705 pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1707 spin_lock_bh(&pcpu
->lock
);
1708 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1711 ct
= nf_ct_tuplehash_to_ctrack(h
);
1713 /* we cannot call iter() on unconfirmed list, the
1714 * owning cpu can reallocate ct->ext at any time.
1716 set_bit(IPS_DYING_BIT
, &ct
->status
);
1718 spin_unlock_bh(&pcpu
->lock
);
1723 void nf_ct_unconfirmed_destroy(struct net
*net
)
1727 if (atomic_read(&net
->ct
.count
) > 0) {
1728 __nf_ct_unconfirmed_destroy(net
);
1729 nf_queue_nf_hook_drop(net
);
1733 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy
);
1735 void nf_ct_iterate_cleanup_net(struct net
*net
,
1736 int (*iter
)(struct nf_conn
*i
, void *data
),
1737 void *data
, u32 portid
, int report
)
1743 if (atomic_read(&net
->ct
.count
) == 0)
1750 nf_ct_iterate_cleanup(iter_net_only
, &d
, portid
, report
);
1752 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net
);
1755 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
1756 * @iter: callback to invoke for each conntrack
1757 * @data: data to pass to @iter
1759 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
1760 * unconfirmed list as dying (so they will not be inserted into
1763 * Can only be called in module exit path.
1766 nf_ct_iterate_destroy(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
1772 if (atomic_read(&net
->ct
.count
) == 0)
1774 __nf_ct_unconfirmed_destroy(net
);
1775 nf_queue_nf_hook_drop(net
);
1779 /* Need to wait for netns cleanup worker to finish, if its
1780 * running -- it might have deleted a net namespace from
1781 * the global list, so our __nf_ct_unconfirmed_destroy() might
1782 * not have affected all namespaces.
1786 /* a conntrack could have been unlinked from unconfirmed list
1787 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
1788 * This makes sure its inserted into conntrack table.
1792 nf_ct_iterate_cleanup(iter
, data
, 0, 0);
1794 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy
);
1796 static int kill_all(struct nf_conn
*i
, void *data
)
1798 return net_eq(nf_ct_net(i
), data
);
1801 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1803 if (is_vmalloc_addr(hash
))
1806 free_pages((unsigned long)hash
,
1807 get_order(sizeof(struct hlist_head
) * size
));
1809 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1811 void nf_conntrack_cleanup_start(void)
1813 conntrack_gc_work
.exiting
= true;
1814 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1817 void nf_conntrack_cleanup_end(void)
1819 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1821 cancel_delayed_work_sync(&conntrack_gc_work
.dwork
);
1822 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1824 nf_conntrack_proto_fini();
1825 nf_conntrack_seqadj_fini();
1826 nf_conntrack_labels_fini();
1827 nf_conntrack_helper_fini();
1828 nf_conntrack_timeout_fini();
1829 nf_conntrack_ecache_fini();
1830 nf_conntrack_tstamp_fini();
1831 nf_conntrack_acct_fini();
1832 nf_conntrack_expect_fini();
1834 kmem_cache_destroy(nf_conntrack_cachep
);
1838 * Mishearing the voices in his head, our hero wonders how he's
1839 * supposed to kill the mall.
1841 void nf_conntrack_cleanup_net(struct net
*net
)
1845 list_add(&net
->exit_list
, &single
);
1846 nf_conntrack_cleanup_net_list(&single
);
1849 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1855 * This makes sure all current packets have passed through
1856 * netfilter framework. Roll on, two-stage module
1862 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1863 nf_ct_iterate_cleanup(kill_all
, net
, 0, 0);
1864 if (atomic_read(&net
->ct
.count
) != 0)
1869 goto i_see_dead_people
;
1872 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1873 nf_conntrack_proto_pernet_fini(net
);
1874 nf_conntrack_helper_pernet_fini(net
);
1875 nf_conntrack_ecache_pernet_fini(net
);
1876 nf_conntrack_tstamp_pernet_fini(net
);
1877 nf_conntrack_acct_pernet_fini(net
);
1878 nf_conntrack_expect_pernet_fini(net
);
1879 free_percpu(net
->ct
.stat
);
1880 free_percpu(net
->ct
.pcpu_lists
);
1884 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1886 struct hlist_nulls_head
*hash
;
1887 unsigned int nr_slots
, i
;
1890 if (*sizep
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
1893 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1894 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1896 if (nr_slots
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
1899 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1900 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1906 for (i
= 0; i
< nr_slots
; i
++)
1907 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1911 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1913 int nf_conntrack_hash_resize(unsigned int hashsize
)
1916 unsigned int old_size
;
1917 struct hlist_nulls_head
*hash
, *old_hash
;
1918 struct nf_conntrack_tuple_hash
*h
;
1924 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1928 old_size
= nf_conntrack_htable_size
;
1929 if (old_size
== hashsize
) {
1930 nf_ct_free_hashtable(hash
, hashsize
);
1935 nf_conntrack_all_lock();
1936 write_seqcount_begin(&nf_conntrack_generation
);
1938 /* Lookups in the old hash might happen in parallel, which means we
1939 * might get false negatives during connection lookup. New connections
1940 * created because of a false negative won't make it into the hash
1941 * though since that required taking the locks.
1944 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1945 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
1946 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
1947 struct nf_conntrack_tuple_hash
, hnnode
);
1948 ct
= nf_ct_tuplehash_to_ctrack(h
);
1949 hlist_nulls_del_rcu(&h
->hnnode
);
1950 bucket
= __hash_conntrack(nf_ct_net(ct
),
1951 &h
->tuple
, hashsize
);
1952 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1955 old_size
= nf_conntrack_htable_size
;
1956 old_hash
= nf_conntrack_hash
;
1958 nf_conntrack_hash
= hash
;
1959 nf_conntrack_htable_size
= hashsize
;
1961 write_seqcount_end(&nf_conntrack_generation
);
1962 nf_conntrack_all_unlock();
1966 nf_ct_free_hashtable(old_hash
, old_size
);
1970 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1972 unsigned int hashsize
;
1975 if (current
->nsproxy
->net_ns
!= &init_net
)
1978 /* On boot, we can set this without any fancy locking. */
1979 if (!nf_conntrack_hash
)
1980 return param_set_uint(val
, kp
);
1982 rc
= kstrtouint(val
, 0, &hashsize
);
1986 return nf_conntrack_hash_resize(hashsize
);
1988 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1990 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1991 &nf_conntrack_htable_size
, 0600);
1993 static __always_inline
unsigned int total_extension_size(void)
1995 /* remember to add new extensions below */
1996 BUILD_BUG_ON(NF_CT_EXT_NUM
> 9);
1998 return sizeof(struct nf_ct_ext
) +
1999 sizeof(struct nf_conn_help
)
2000 #if IS_ENABLED(CONFIG_NF_NAT)
2001 + sizeof(struct nf_conn_nat
)
2003 + sizeof(struct nf_conn_seqadj
)
2004 + sizeof(struct nf_conn_acct
)
2005 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2006 + sizeof(struct nf_conntrack_ecache
)
2008 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2009 + sizeof(struct nf_conn_tstamp
)
2011 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2012 + sizeof(struct nf_conn_timeout
)
2014 #ifdef CONFIG_NF_CONNTRACK_LABELS
2015 + sizeof(struct nf_conn_labels
)
2017 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2018 + sizeof(struct nf_conn_synproxy
)
2023 int nf_conntrack_init_start(void)
2029 /* struct nf_ct_ext uses u8 to store offsets/size */
2030 BUILD_BUG_ON(total_extension_size() > 255u);
2032 seqcount_init(&nf_conntrack_generation
);
2034 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
2035 spin_lock_init(&nf_conntrack_locks
[i
]);
2037 if (!nf_conntrack_htable_size
) {
2038 /* Idea from tcp.c: use 1/16384 of memory.
2039 * On i386: 32MB machine has 512 buckets.
2040 * >= 1GB machines have 16384 buckets.
2041 * >= 4GB machines have 65536 buckets.
2043 nf_conntrack_htable_size
2044 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
2045 / sizeof(struct hlist_head
));
2046 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
2047 nf_conntrack_htable_size
= 65536;
2048 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
2049 nf_conntrack_htable_size
= 16384;
2050 if (nf_conntrack_htable_size
< 32)
2051 nf_conntrack_htable_size
= 32;
2053 /* Use a max. factor of four by default to get the same max as
2054 * with the old struct list_heads. When a table size is given
2055 * we use the old value of 8 to avoid reducing the max.
2060 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
2061 if (!nf_conntrack_hash
)
2064 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
2066 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
2067 sizeof(struct nf_conn
),
2069 SLAB_TYPESAFE_BY_RCU
| SLAB_HWCACHE_ALIGN
, NULL
);
2070 if (!nf_conntrack_cachep
)
2073 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
2074 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
2077 ret
= nf_conntrack_expect_init();
2081 ret
= nf_conntrack_acct_init();
2085 ret
= nf_conntrack_tstamp_init();
2089 ret
= nf_conntrack_ecache_init();
2093 ret
= nf_conntrack_timeout_init();
2097 ret
= nf_conntrack_helper_init();
2101 ret
= nf_conntrack_labels_init();
2105 ret
= nf_conntrack_seqadj_init();
2109 ret
= nf_conntrack_proto_init();
2113 conntrack_gc_work_init(&conntrack_gc_work
);
2114 queue_delayed_work(system_long_wq
, &conntrack_gc_work
.dwork
, HZ
);
2119 nf_conntrack_seqadj_fini();
2121 nf_conntrack_labels_fini();
2123 nf_conntrack_helper_fini();
2125 nf_conntrack_timeout_fini();
2127 nf_conntrack_ecache_fini();
2129 nf_conntrack_tstamp_fini();
2131 nf_conntrack_acct_fini();
2133 nf_conntrack_expect_fini();
2135 kmem_cache_destroy(nf_conntrack_cachep
);
2137 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
2141 void nf_conntrack_init_end(void)
2143 /* For use by REJECT target */
2144 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
2145 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
2149 * We need to use special "null" values, not used in hash table
2151 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2152 #define DYING_NULLS_VAL ((1<<30)+1)
2153 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
2155 int nf_conntrack_init_net(struct net
*net
)
2160 BUILD_BUG_ON(IP_CT_UNTRACKED
== IP_CT_NUMBER
);
2161 atomic_set(&net
->ct
.count
, 0);
2163 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
2164 if (!net
->ct
.pcpu_lists
)
2167 for_each_possible_cpu(cpu
) {
2168 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2170 spin_lock_init(&pcpu
->lock
);
2171 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
2172 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
2175 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
2177 goto err_pcpu_lists
;
2179 ret
= nf_conntrack_expect_pernet_init(net
);
2182 ret
= nf_conntrack_acct_pernet_init(net
);
2185 ret
= nf_conntrack_tstamp_pernet_init(net
);
2188 ret
= nf_conntrack_ecache_pernet_init(net
);
2191 ret
= nf_conntrack_helper_pernet_init(net
);
2194 ret
= nf_conntrack_proto_pernet_init(net
);
2200 nf_conntrack_helper_pernet_fini(net
);
2202 nf_conntrack_ecache_pernet_fini(net
);
2204 nf_conntrack_tstamp_pernet_fini(net
);
2206 nf_conntrack_acct_pernet_fini(net
);
2208 nf_conntrack_expect_pernet_fini(net
);
2210 free_percpu(net
->ct
.stat
);
2212 free_percpu(net
->ct
.pcpu_lists
);