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
));
512 nf_ct_match(const struct nf_conn
*ct1
, const struct nf_conn
*ct2
)
514 return nf_ct_tuple_equal(&ct1
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
515 &ct2
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
) &&
516 nf_ct_tuple_equal(&ct1
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
517 &ct2
->tuplehash
[IP_CT_DIR_REPLY
].tuple
) &&
518 nf_ct_zone_equal(ct1
, nf_ct_zone(ct2
), IP_CT_DIR_ORIGINAL
) &&
519 nf_ct_zone_equal(ct1
, nf_ct_zone(ct2
), IP_CT_DIR_REPLY
) &&
520 net_eq(nf_ct_net(ct1
), nf_ct_net(ct2
));
523 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
524 static void nf_ct_gc_expired(struct nf_conn
*ct
)
526 if (!atomic_inc_not_zero(&ct
->ct_general
.use
))
529 if (nf_ct_should_gc(ct
))
537 * - Caller must take a reference on returned object
538 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
540 static struct nf_conntrack_tuple_hash
*
541 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
542 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
544 struct nf_conntrack_tuple_hash
*h
;
545 struct hlist_nulls_head
*ct_hash
;
546 struct hlist_nulls_node
*n
;
547 unsigned int bucket
, hsize
;
550 nf_conntrack_get_ht(&ct_hash
, &hsize
);
551 bucket
= reciprocal_scale(hash
, hsize
);
553 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[bucket
], hnnode
) {
556 ct
= nf_ct_tuplehash_to_ctrack(h
);
557 if (nf_ct_is_expired(ct
)) {
558 nf_ct_gc_expired(ct
);
562 if (nf_ct_is_dying(ct
))
565 if (nf_ct_key_equal(h
, tuple
, zone
, net
))
569 * if the nulls value we got at the end of this lookup is
570 * not the expected one, we must restart lookup.
571 * We probably met an item that was moved to another chain.
573 if (get_nulls_value(n
) != bucket
) {
574 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
581 /* Find a connection corresponding to a tuple. */
582 static struct nf_conntrack_tuple_hash
*
583 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
584 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
586 struct nf_conntrack_tuple_hash
*h
;
591 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
593 ct
= nf_ct_tuplehash_to_ctrack(h
);
594 if (unlikely(nf_ct_is_dying(ct
) ||
595 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
598 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
, net
))) {
609 struct nf_conntrack_tuple_hash
*
610 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
611 const struct nf_conntrack_tuple
*tuple
)
613 return __nf_conntrack_find_get(net
, zone
, tuple
,
614 hash_conntrack_raw(tuple
, net
));
616 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
618 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
620 unsigned int reply_hash
)
622 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
623 &nf_conntrack_hash
[hash
]);
624 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
625 &nf_conntrack_hash
[reply_hash
]);
629 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
631 const struct nf_conntrack_zone
*zone
;
632 struct net
*net
= nf_ct_net(ct
);
633 unsigned int hash
, reply_hash
;
634 struct nf_conntrack_tuple_hash
*h
;
635 struct hlist_nulls_node
*n
;
636 unsigned int sequence
;
638 zone
= nf_ct_zone(ct
);
642 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
643 hash
= hash_conntrack(net
,
644 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
645 reply_hash
= hash_conntrack(net
,
646 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
647 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
649 /* See if there's one in the list already, including reverse */
650 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
651 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
655 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
656 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
661 /* The caller holds a reference to this object */
662 atomic_set(&ct
->ct_general
.use
, 2);
663 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
664 nf_conntrack_double_unlock(hash
, reply_hash
);
665 NF_CT_STAT_INC(net
, insert
);
670 nf_conntrack_double_unlock(hash
, reply_hash
);
671 NF_CT_STAT_INC(net
, insert_failed
);
675 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
677 static inline void nf_ct_acct_update(struct nf_conn
*ct
,
678 enum ip_conntrack_info ctinfo
,
681 struct nf_conn_acct
*acct
;
683 acct
= nf_conn_acct_find(ct
);
685 struct nf_conn_counter
*counter
= acct
->counter
;
687 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
688 atomic64_add(len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
692 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
693 const struct nf_conn
*loser_ct
)
695 struct nf_conn_acct
*acct
;
697 acct
= nf_conn_acct_find(loser_ct
);
699 struct nf_conn_counter
*counter
= acct
->counter
;
702 /* u32 should be fine since we must have seen one packet. */
703 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
704 nf_ct_acct_update(ct
, ctinfo
, bytes
);
708 /* Resolve race on insertion if this protocol allows this. */
709 static int nf_ct_resolve_clash(struct net
*net
, struct sk_buff
*skb
,
710 enum ip_conntrack_info ctinfo
,
711 struct nf_conntrack_tuple_hash
*h
)
713 /* This is the conntrack entry already in hashes that won race. */
714 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
715 const struct nf_conntrack_l4proto
*l4proto
;
716 enum ip_conntrack_info oldinfo
;
717 struct nf_conn
*loser_ct
= nf_ct_get(skb
, &oldinfo
);
719 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
720 if (l4proto
->allow_clash
&&
721 !nf_ct_is_dying(ct
) &&
722 atomic_inc_not_zero(&ct
->ct_general
.use
)) {
723 if (((ct
->status
& IPS_NAT_DONE_MASK
) == 0) ||
724 nf_ct_match(ct
, loser_ct
)) {
725 nf_ct_acct_merge(ct
, ctinfo
, loser_ct
);
726 nf_conntrack_put(&loser_ct
->ct_general
);
727 nf_ct_set(skb
, ct
, oldinfo
);
732 NF_CT_STAT_INC(net
, drop
);
736 /* Confirm a connection given skb; places it in hash table */
738 __nf_conntrack_confirm(struct sk_buff
*skb
)
740 const struct nf_conntrack_zone
*zone
;
741 unsigned int hash
, reply_hash
;
742 struct nf_conntrack_tuple_hash
*h
;
744 struct nf_conn_help
*help
;
745 struct nf_conn_tstamp
*tstamp
;
746 struct hlist_nulls_node
*n
;
747 enum ip_conntrack_info ctinfo
;
749 unsigned int sequence
;
752 ct
= nf_ct_get(skb
, &ctinfo
);
755 /* ipt_REJECT uses nf_conntrack_attach to attach related
756 ICMP/TCP RST packets in other direction. Actual packet
757 which created connection will be IP_CT_NEW or for an
758 expected connection, IP_CT_RELATED. */
759 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
762 zone
= nf_ct_zone(ct
);
766 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
767 /* reuse the hash saved before */
768 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
769 hash
= scale_hash(hash
);
770 reply_hash
= hash_conntrack(net
,
771 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
773 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
775 /* We're not in hash table, and we refuse to set up related
776 * connections for unconfirmed conns. But packet copies and
777 * REJECT will give spurious warnings here.
780 /* Another skb with the same unconfirmed conntrack may
781 * win the race. This may happen for bridge(br_flood)
782 * or broadcast/multicast packets do skb_clone with
783 * unconfirmed conntrack.
785 if (unlikely(nf_ct_is_confirmed(ct
))) {
787 nf_conntrack_double_unlock(hash
, reply_hash
);
792 pr_debug("Confirming conntrack %p\n", ct
);
793 /* We have to check the DYING flag after unlink to prevent
794 * a race against nf_ct_get_next_corpse() possibly called from
795 * user context, else we insert an already 'dead' hash, blocking
796 * further use of that particular connection -JM.
798 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
800 if (unlikely(nf_ct_is_dying(ct
))) {
801 nf_ct_add_to_dying_list(ct
);
805 /* See if there's one in the list already, including reverse:
806 NAT could have grabbed it without realizing, since we're
807 not in the hash. If there is, we lost race. */
808 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
809 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
813 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
814 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
818 /* Timer relative to confirmation time, not original
819 setting time, otherwise we'd get timer wrap in
820 weird delay cases. */
821 ct
->timeout
+= nfct_time_stamp
;
822 atomic_inc(&ct
->ct_general
.use
);
823 ct
->status
|= IPS_CONFIRMED
;
825 /* set conntrack timestamp, if enabled. */
826 tstamp
= nf_conn_tstamp_find(ct
);
828 if (skb
->tstamp
== 0)
829 __net_timestamp(skb
);
831 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
833 /* Since the lookup is lockless, hash insertion must be done after
834 * starting the timer and setting the CONFIRMED bit. The RCU barriers
835 * guarantee that no other CPU can find the conntrack before the above
836 * stores are visible.
838 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
839 nf_conntrack_double_unlock(hash
, reply_hash
);
842 help
= nfct_help(ct
);
843 if (help
&& help
->helper
)
844 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
846 nf_conntrack_event_cache(master_ct(ct
) ?
847 IPCT_RELATED
: IPCT_NEW
, ct
);
851 nf_ct_add_to_dying_list(ct
);
852 ret
= nf_ct_resolve_clash(net
, skb
, ctinfo
, h
);
854 nf_conntrack_double_unlock(hash
, reply_hash
);
855 NF_CT_STAT_INC(net
, insert_failed
);
859 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
861 /* Returns true if a connection correspondings to the tuple (required
864 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
865 const struct nf_conn
*ignored_conntrack
)
867 struct net
*net
= nf_ct_net(ignored_conntrack
);
868 const struct nf_conntrack_zone
*zone
;
869 struct nf_conntrack_tuple_hash
*h
;
870 struct hlist_nulls_head
*ct_hash
;
871 unsigned int hash
, hsize
;
872 struct hlist_nulls_node
*n
;
875 zone
= nf_ct_zone(ignored_conntrack
);
879 nf_conntrack_get_ht(&ct_hash
, &hsize
);
880 hash
= __hash_conntrack(net
, tuple
, hsize
);
882 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
883 ct
= nf_ct_tuplehash_to_ctrack(h
);
885 if (ct
== ignored_conntrack
)
888 if (nf_ct_is_expired(ct
)) {
889 nf_ct_gc_expired(ct
);
893 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
894 /* Tuple is taken already, so caller will need to find
895 * a new source port to use.
898 * If the *original tuples* are identical, then both
899 * conntracks refer to the same flow.
900 * This is a rare situation, it can occur e.g. when
901 * more than one UDP packet is sent from same socket
902 * in different threads.
904 * Let nf_ct_resolve_clash() deal with this later.
906 if (nf_ct_tuple_equal(&ignored_conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
907 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
))
910 NF_CT_STAT_INC_ATOMIC(net
, found
);
916 if (get_nulls_value(n
) != hash
) {
917 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
925 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
927 #define NF_CT_EVICTION_RANGE 8
929 /* There's a small race here where we may free a just-assured
930 connection. Too bad: we're in trouble anyway. */
931 static unsigned int early_drop_list(struct net
*net
,
932 struct hlist_nulls_head
*head
)
934 struct nf_conntrack_tuple_hash
*h
;
935 struct hlist_nulls_node
*n
;
936 unsigned int drops
= 0;
939 hlist_nulls_for_each_entry_rcu(h
, n
, head
, hnnode
) {
940 tmp
= nf_ct_tuplehash_to_ctrack(h
);
942 if (nf_ct_is_expired(tmp
)) {
943 nf_ct_gc_expired(tmp
);
947 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
948 !net_eq(nf_ct_net(tmp
), net
) ||
952 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
955 /* kill only if still in same netns -- might have moved due to
956 * SLAB_TYPESAFE_BY_RCU rules.
958 * We steal the timer reference. If that fails timer has
959 * already fired or someone else deleted it. Just drop ref
960 * and move to next entry.
962 if (net_eq(nf_ct_net(tmp
), net
) &&
963 nf_ct_is_confirmed(tmp
) &&
964 nf_ct_delete(tmp
, 0, 0))
973 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
975 unsigned int i
, bucket
;
977 for (i
= 0; i
< NF_CT_EVICTION_RANGE
; i
++) {
978 struct hlist_nulls_head
*ct_hash
;
979 unsigned int hsize
, drops
;
982 nf_conntrack_get_ht(&ct_hash
, &hsize
);
984 bucket
= reciprocal_scale(hash
, hsize
);
986 bucket
= (bucket
+ 1) % hsize
;
988 drops
= early_drop_list(net
, &ct_hash
[bucket
]);
992 NF_CT_STAT_ADD_ATOMIC(net
, early_drop
, drops
);
1000 static bool gc_worker_skip_ct(const struct nf_conn
*ct
)
1002 return !nf_ct_is_confirmed(ct
) || nf_ct_is_dying(ct
);
1005 static bool gc_worker_can_early_drop(const struct nf_conn
*ct
)
1007 const struct nf_conntrack_l4proto
*l4proto
;
1009 if (!test_bit(IPS_ASSURED_BIT
, &ct
->status
))
1012 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
1013 if (l4proto
->can_early_drop
&& l4proto
->can_early_drop(ct
))
1019 static void gc_worker(struct work_struct
*work
)
1021 unsigned int min_interval
= max(HZ
/ GC_MAX_BUCKETS_DIV
, 1u);
1022 unsigned int i
, goal
, buckets
= 0, expired_count
= 0;
1023 unsigned int nf_conntrack_max95
= 0;
1024 struct conntrack_gc_work
*gc_work
;
1025 unsigned int ratio
, scanned
= 0;
1026 unsigned long next_run
;
1028 gc_work
= container_of(work
, struct conntrack_gc_work
, dwork
.work
);
1030 goal
= nf_conntrack_htable_size
/ GC_MAX_BUCKETS_DIV
;
1031 i
= gc_work
->last_bucket
;
1032 if (gc_work
->early_drop
)
1033 nf_conntrack_max95
= nf_conntrack_max
/ 100u * 95u;
1036 struct nf_conntrack_tuple_hash
*h
;
1037 struct hlist_nulls_head
*ct_hash
;
1038 struct hlist_nulls_node
*n
;
1039 unsigned int hashsz
;
1040 struct nf_conn
*tmp
;
1045 nf_conntrack_get_ht(&ct_hash
, &hashsz
);
1049 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[i
], hnnode
) {
1052 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1055 if (nf_ct_is_expired(tmp
)) {
1056 nf_ct_gc_expired(tmp
);
1061 if (nf_conntrack_max95
== 0 || gc_worker_skip_ct(tmp
))
1064 net
= nf_ct_net(tmp
);
1065 if (atomic_read(&net
->ct
.count
) < nf_conntrack_max95
)
1068 /* need to take reference to avoid possible races */
1069 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1072 if (gc_worker_skip_ct(tmp
)) {
1077 if (gc_worker_can_early_drop(tmp
))
1083 /* could check get_nulls_value() here and restart if ct
1084 * was moved to another chain. But given gc is best-effort
1085 * we will just continue with next hash slot.
1088 cond_resched_rcu_qs();
1089 } while (++buckets
< goal
);
1091 if (gc_work
->exiting
)
1095 * Eviction will normally happen from the packet path, and not
1096 * from this gc worker.
1098 * This worker is only here to reap expired entries when system went
1099 * idle after a busy period.
1101 * The heuristics below are supposed to balance conflicting goals:
1103 * 1. Minimize time until we notice a stale entry
1104 * 2. Maximize scan intervals to not waste cycles
1106 * Normally, expire ratio will be close to 0.
1108 * As soon as a sizeable fraction of the entries have expired
1109 * increase scan frequency.
1111 ratio
= scanned
? expired_count
* 100 / scanned
: 0;
1112 if (ratio
> GC_EVICT_RATIO
) {
1113 gc_work
->next_gc_run
= min_interval
;
1115 unsigned int max
= GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
;
1117 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
) == 0);
1119 gc_work
->next_gc_run
+= min_interval
;
1120 if (gc_work
->next_gc_run
> max
)
1121 gc_work
->next_gc_run
= max
;
1124 next_run
= gc_work
->next_gc_run
;
1125 gc_work
->last_bucket
= i
;
1126 gc_work
->early_drop
= false;
1127 queue_delayed_work(system_power_efficient_wq
, &gc_work
->dwork
, next_run
);
1130 static void conntrack_gc_work_init(struct conntrack_gc_work
*gc_work
)
1132 INIT_DEFERRABLE_WORK(&gc_work
->dwork
, gc_worker
);
1133 gc_work
->next_gc_run
= HZ
;
1134 gc_work
->exiting
= false;
1137 static struct nf_conn
*
1138 __nf_conntrack_alloc(struct net
*net
,
1139 const struct nf_conntrack_zone
*zone
,
1140 const struct nf_conntrack_tuple
*orig
,
1141 const struct nf_conntrack_tuple
*repl
,
1142 gfp_t gfp
, u32 hash
)
1146 /* We don't want any race condition at early drop stage */
1147 atomic_inc(&net
->ct
.count
);
1149 if (nf_conntrack_max
&&
1150 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
1151 if (!early_drop(net
, hash
)) {
1152 if (!conntrack_gc_work
.early_drop
)
1153 conntrack_gc_work
.early_drop
= true;
1154 atomic_dec(&net
->ct
.count
);
1155 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1156 return ERR_PTR(-ENOMEM
);
1161 * Do not use kmem_cache_zalloc(), as this cache uses
1162 * SLAB_TYPESAFE_BY_RCU.
1164 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
1168 spin_lock_init(&ct
->lock
);
1169 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
1170 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
1171 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
1172 /* save hash for reusing when confirming */
1173 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
1175 write_pnet(&ct
->ct_net
, net
);
1176 memset(&ct
->__nfct_init_offset
[0], 0,
1177 offsetof(struct nf_conn
, proto
) -
1178 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
1180 nf_ct_zone_add(ct
, zone
);
1182 /* Because we use RCU lookups, we set ct_general.use to zero before
1183 * this is inserted in any list.
1185 atomic_set(&ct
->ct_general
.use
, 0);
1188 atomic_dec(&net
->ct
.count
);
1189 return ERR_PTR(-ENOMEM
);
1192 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
1193 const struct nf_conntrack_zone
*zone
,
1194 const struct nf_conntrack_tuple
*orig
,
1195 const struct nf_conntrack_tuple
*repl
,
1198 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
1200 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
1202 void nf_conntrack_free(struct nf_conn
*ct
)
1204 struct net
*net
= nf_ct_net(ct
);
1206 /* A freed object has refcnt == 0, that's
1207 * the golden rule for SLAB_TYPESAFE_BY_RCU
1209 WARN_ON(atomic_read(&ct
->ct_general
.use
) != 0);
1211 nf_ct_ext_destroy(ct
);
1213 kmem_cache_free(nf_conntrack_cachep
, ct
);
1214 smp_mb__before_atomic();
1215 atomic_dec(&net
->ct
.count
);
1217 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
1220 /* Allocate a new conntrack: we return -ENOMEM if classification
1221 failed due to stress. Otherwise it really is unclassifiable. */
1222 static noinline
struct nf_conntrack_tuple_hash
*
1223 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
1224 const struct nf_conntrack_tuple
*tuple
,
1225 const struct nf_conntrack_l3proto
*l3proto
,
1226 const struct nf_conntrack_l4proto
*l4proto
,
1227 struct sk_buff
*skb
,
1228 unsigned int dataoff
, u32 hash
)
1231 struct nf_conn_help
*help
;
1232 struct nf_conntrack_tuple repl_tuple
;
1233 struct nf_conntrack_ecache
*ecache
;
1234 struct nf_conntrack_expect
*exp
= NULL
;
1235 const struct nf_conntrack_zone
*zone
;
1236 struct nf_conn_timeout
*timeout_ext
;
1237 struct nf_conntrack_zone tmp
;
1238 unsigned int *timeouts
;
1240 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
1241 pr_debug("Can't invert tuple.\n");
1245 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1246 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
1249 return (struct nf_conntrack_tuple_hash
*)ct
;
1251 if (!nf_ct_add_synproxy(ct
, tmpl
)) {
1252 nf_conntrack_free(ct
);
1253 return ERR_PTR(-ENOMEM
);
1256 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1258 timeouts
= nf_ct_timeout_data(timeout_ext
);
1259 if (unlikely(!timeouts
))
1260 timeouts
= l4proto
->get_timeouts(net
);
1262 timeouts
= l4proto
->get_timeouts(net
);
1265 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
1266 nf_conntrack_free(ct
);
1267 pr_debug("can't track with proto module\n");
1272 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1275 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1276 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1277 nf_ct_labels_ext_add(ct
);
1279 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1280 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1281 ecache
? ecache
->expmask
: 0,
1285 if (net
->ct
.expect_count
) {
1286 spin_lock(&nf_conntrack_expect_lock
);
1287 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1289 pr_debug("expectation arrives ct=%p exp=%p\n",
1291 /* Welcome, Mr. Bond. We've been expecting you... */
1292 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1293 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1294 ct
->master
= exp
->master
;
1296 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
1299 rcu_assign_pointer(help
->helper
, exp
->helper
);
1302 #ifdef CONFIG_NF_CONNTRACK_MARK
1303 ct
->mark
= exp
->master
->mark
;
1305 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1306 ct
->secmark
= exp
->master
->secmark
;
1308 NF_CT_STAT_INC(net
, expect_new
);
1310 spin_unlock(&nf_conntrack_expect_lock
);
1313 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1315 /* Now it is inserted into the unconfirmed list, bump refcount */
1316 nf_conntrack_get(&ct
->ct_general
);
1317 nf_ct_add_to_unconfirmed_list(ct
);
1323 exp
->expectfn(ct
, exp
);
1324 nf_ct_expect_put(exp
);
1327 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1330 /* On success, returns 0, sets skb->_nfct | ctinfo */
1332 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1333 struct sk_buff
*skb
,
1334 unsigned int dataoff
,
1337 const struct nf_conntrack_l3proto
*l3proto
,
1338 const struct nf_conntrack_l4proto
*l4proto
)
1340 const struct nf_conntrack_zone
*zone
;
1341 struct nf_conntrack_tuple tuple
;
1342 struct nf_conntrack_tuple_hash
*h
;
1343 enum ip_conntrack_info ctinfo
;
1344 struct nf_conntrack_zone tmp
;
1348 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1349 dataoff
, l3num
, protonum
, net
, &tuple
, l3proto
,
1351 pr_debug("Can't get tuple\n");
1355 /* look for tuple match */
1356 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1357 hash
= hash_conntrack_raw(&tuple
, net
);
1358 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1360 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1361 skb
, dataoff
, hash
);
1367 ct
= nf_ct_tuplehash_to_ctrack(h
);
1369 /* It exists; we have (non-exclusive) reference. */
1370 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1371 ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1373 /* Once we've had two way comms, always ESTABLISHED. */
1374 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1375 pr_debug("normal packet for %p\n", ct
);
1376 ctinfo
= IP_CT_ESTABLISHED
;
1377 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1378 pr_debug("related packet for %p\n", ct
);
1379 ctinfo
= IP_CT_RELATED
;
1381 pr_debug("new packet for %p\n", ct
);
1385 nf_ct_set(skb
, ct
, ctinfo
);
1390 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1391 struct sk_buff
*skb
)
1393 const struct nf_conntrack_l3proto
*l3proto
;
1394 const struct nf_conntrack_l4proto
*l4proto
;
1395 struct nf_conn
*ct
, *tmpl
;
1396 enum ip_conntrack_info ctinfo
;
1397 unsigned int *timeouts
;
1398 unsigned int dataoff
;
1402 tmpl
= nf_ct_get(skb
, &ctinfo
);
1403 if (tmpl
|| ctinfo
== IP_CT_UNTRACKED
) {
1404 /* Previously seen (loopback or untracked)? Ignore. */
1405 if ((tmpl
&& !nf_ct_is_template(tmpl
)) ||
1406 ctinfo
== IP_CT_UNTRACKED
) {
1407 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1413 /* rcu_read_lock()ed by nf_hook_thresh */
1414 l3proto
= __nf_ct_l3proto_find(pf
);
1415 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1416 &dataoff
, &protonum
);
1418 pr_debug("not prepared to track yet or error occurred\n");
1419 NF_CT_STAT_INC_ATOMIC(net
, error
);
1420 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1425 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1427 /* It may be an special packet, error, unclean...
1428 * inverse of the return code tells to the netfilter
1429 * core what to do with the packet. */
1430 if (l4proto
->error
!= NULL
) {
1431 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, pf
, hooknum
);
1433 NF_CT_STAT_INC_ATOMIC(net
, error
);
1434 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1438 /* ICMP[v6] protocol trackers may assign one conntrack. */
1443 ret
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1446 /* Too stressed to deal. */
1447 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1452 ct
= nf_ct_get(skb
, &ctinfo
);
1454 /* Not valid part of a connection */
1455 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1460 /* Decide what timeout policy we want to apply to this flow. */
1461 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1463 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, timeouts
);
1465 /* Invalid: inverse of the return code tells
1466 * the netfilter core what to do */
1467 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1468 nf_conntrack_put(&ct
->ct_general
);
1470 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1471 if (ret
== -NF_DROP
)
1472 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1473 /* Special case: TCP tracker reports an attempt to reopen a
1474 * closed/aborted connection. We have to go back and create a
1477 if (ret
== -NF_REPEAT
)
1483 if (ctinfo
== IP_CT_ESTABLISHED_REPLY
&&
1484 !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1485 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1492 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1494 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1495 const struct nf_conntrack_tuple
*orig
)
1500 ret
= nf_ct_invert_tuple(inverse
, orig
,
1501 __nf_ct_l3proto_find(orig
->src
.l3num
),
1502 __nf_ct_l4proto_find(orig
->src
.l3num
,
1503 orig
->dst
.protonum
));
1507 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1509 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1510 implicitly racy: see __nf_conntrack_confirm */
1511 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1512 const struct nf_conntrack_tuple
*newreply
)
1514 struct nf_conn_help
*help
= nfct_help(ct
);
1516 /* Should be unconfirmed, so not in hash table yet */
1517 WARN_ON(nf_ct_is_confirmed(ct
));
1519 pr_debug("Altering reply tuple of %p to ", ct
);
1520 nf_ct_dump_tuple(newreply
);
1522 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1523 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1527 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1530 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1532 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1533 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1534 enum ip_conntrack_info ctinfo
,
1535 const struct sk_buff
*skb
,
1536 unsigned long extra_jiffies
,
1541 /* Only update if this is not a fixed timeout */
1542 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1545 /* If not in hash table, timer will not be active yet */
1546 if (nf_ct_is_confirmed(ct
))
1547 extra_jiffies
+= nfct_time_stamp
;
1549 ct
->timeout
= extra_jiffies
;
1552 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1554 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1556 bool nf_ct_kill_acct(struct nf_conn
*ct
,
1557 enum ip_conntrack_info ctinfo
,
1558 const struct sk_buff
*skb
)
1560 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1562 return nf_ct_delete(ct
, 0, 0);
1564 EXPORT_SYMBOL_GPL(nf_ct_kill_acct
);
1566 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1568 #include <linux/netfilter/nfnetlink.h>
1569 #include <linux/netfilter/nfnetlink_conntrack.h>
1570 #include <linux/mutex.h>
1572 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1573 * in ip_conntrack_core, since we don't want the protocols to autoload
1574 * or depend on ctnetlink */
1575 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1576 const struct nf_conntrack_tuple
*tuple
)
1578 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1579 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1580 goto nla_put_failure
;
1586 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1588 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1589 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1590 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1592 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1594 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1595 struct nf_conntrack_tuple
*t
)
1597 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1600 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1601 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1605 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1607 unsigned int nf_ct_port_nlattr_tuple_size(void)
1609 static unsigned int size __read_mostly
;
1612 size
= nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1616 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1619 /* Used by ipt_REJECT and ip6t_REJECT. */
1620 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1623 enum ip_conntrack_info ctinfo
;
1625 /* This ICMP is in reverse direction to the packet which caused it */
1626 ct
= nf_ct_get(skb
, &ctinfo
);
1627 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1628 ctinfo
= IP_CT_RELATED_REPLY
;
1630 ctinfo
= IP_CT_RELATED
;
1632 /* Attach to new skbuff, and increment count */
1633 nf_ct_set(nskb
, ct
, ctinfo
);
1634 nf_conntrack_get(skb_nfct(nskb
));
1637 /* Bring out ya dead! */
1638 static struct nf_conn
*
1639 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
1640 void *data
, unsigned int *bucket
)
1642 struct nf_conntrack_tuple_hash
*h
;
1644 struct hlist_nulls_node
*n
;
1647 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1648 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1650 nf_conntrack_lock(lockp
);
1651 if (*bucket
< nf_conntrack_htable_size
) {
1652 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
1653 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1655 ct
= nf_ct_tuplehash_to_ctrack(h
);
1667 atomic_inc(&ct
->ct_general
.use
);
1673 static void nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
),
1674 void *data
, u32 portid
, int report
)
1676 unsigned int bucket
= 0, sequence
;
1682 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
1684 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
1685 /* Time to push up daises... */
1687 nf_ct_delete(ct
, portid
, report
);
1692 if (!read_seqcount_retry(&nf_conntrack_generation
, sequence
))
1699 int (*iter
)(struct nf_conn
*i
, void *data
);
1704 static int iter_net_only(struct nf_conn
*i
, void *data
)
1706 struct iter_data
*d
= data
;
1708 if (!net_eq(d
->net
, nf_ct_net(i
)))
1711 return d
->iter(i
, d
->data
);
1715 __nf_ct_unconfirmed_destroy(struct net
*net
)
1719 for_each_possible_cpu(cpu
) {
1720 struct nf_conntrack_tuple_hash
*h
;
1721 struct hlist_nulls_node
*n
;
1722 struct ct_pcpu
*pcpu
;
1724 pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1726 spin_lock_bh(&pcpu
->lock
);
1727 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1730 ct
= nf_ct_tuplehash_to_ctrack(h
);
1732 /* we cannot call iter() on unconfirmed list, the
1733 * owning cpu can reallocate ct->ext at any time.
1735 set_bit(IPS_DYING_BIT
, &ct
->status
);
1737 spin_unlock_bh(&pcpu
->lock
);
1742 void nf_ct_unconfirmed_destroy(struct net
*net
)
1746 if (atomic_read(&net
->ct
.count
) > 0) {
1747 __nf_ct_unconfirmed_destroy(net
);
1748 nf_queue_nf_hook_drop(net
);
1752 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy
);
1754 void nf_ct_iterate_cleanup_net(struct net
*net
,
1755 int (*iter
)(struct nf_conn
*i
, void *data
),
1756 void *data
, u32 portid
, int report
)
1762 if (atomic_read(&net
->ct
.count
) == 0)
1769 nf_ct_iterate_cleanup(iter_net_only
, &d
, portid
, report
);
1771 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net
);
1774 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
1775 * @iter: callback to invoke for each conntrack
1776 * @data: data to pass to @iter
1778 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
1779 * unconfirmed list as dying (so they will not be inserted into
1782 * Can only be called in module exit path.
1785 nf_ct_iterate_destroy(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
1791 if (atomic_read(&net
->ct
.count
) == 0)
1793 __nf_ct_unconfirmed_destroy(net
);
1794 nf_queue_nf_hook_drop(net
);
1798 /* Need to wait for netns cleanup worker to finish, if its
1799 * running -- it might have deleted a net namespace from
1800 * the global list, so our __nf_ct_unconfirmed_destroy() might
1801 * not have affected all namespaces.
1805 /* a conntrack could have been unlinked from unconfirmed list
1806 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
1807 * This makes sure its inserted into conntrack table.
1811 nf_ct_iterate_cleanup(iter
, data
, 0, 0);
1813 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy
);
1815 static int kill_all(struct nf_conn
*i
, void *data
)
1817 return net_eq(nf_ct_net(i
), data
);
1820 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1822 if (is_vmalloc_addr(hash
))
1825 free_pages((unsigned long)hash
,
1826 get_order(sizeof(struct hlist_head
) * size
));
1828 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1830 void nf_conntrack_cleanup_start(void)
1832 conntrack_gc_work
.exiting
= true;
1833 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1836 void nf_conntrack_cleanup_end(void)
1838 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1840 cancel_delayed_work_sync(&conntrack_gc_work
.dwork
);
1841 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1843 nf_conntrack_proto_fini();
1844 nf_conntrack_seqadj_fini();
1845 nf_conntrack_labels_fini();
1846 nf_conntrack_helper_fini();
1847 nf_conntrack_timeout_fini();
1848 nf_conntrack_ecache_fini();
1849 nf_conntrack_tstamp_fini();
1850 nf_conntrack_acct_fini();
1851 nf_conntrack_expect_fini();
1853 kmem_cache_destroy(nf_conntrack_cachep
);
1857 * Mishearing the voices in his head, our hero wonders how he's
1858 * supposed to kill the mall.
1860 void nf_conntrack_cleanup_net(struct net
*net
)
1864 list_add(&net
->exit_list
, &single
);
1865 nf_conntrack_cleanup_net_list(&single
);
1868 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1874 * This makes sure all current packets have passed through
1875 * netfilter framework. Roll on, two-stage module
1881 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1882 nf_ct_iterate_cleanup(kill_all
, net
, 0, 0);
1883 if (atomic_read(&net
->ct
.count
) != 0)
1888 goto i_see_dead_people
;
1891 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1892 nf_conntrack_proto_pernet_fini(net
);
1893 nf_conntrack_helper_pernet_fini(net
);
1894 nf_conntrack_ecache_pernet_fini(net
);
1895 nf_conntrack_tstamp_pernet_fini(net
);
1896 nf_conntrack_acct_pernet_fini(net
);
1897 nf_conntrack_expect_pernet_fini(net
);
1898 free_percpu(net
->ct
.stat
);
1899 free_percpu(net
->ct
.pcpu_lists
);
1903 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1905 struct hlist_nulls_head
*hash
;
1906 unsigned int nr_slots
, i
;
1909 if (*sizep
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
1912 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1913 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1915 if (nr_slots
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
1918 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1919 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1925 for (i
= 0; i
< nr_slots
; i
++)
1926 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1930 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1932 int nf_conntrack_hash_resize(unsigned int hashsize
)
1935 unsigned int old_size
;
1936 struct hlist_nulls_head
*hash
, *old_hash
;
1937 struct nf_conntrack_tuple_hash
*h
;
1943 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1947 old_size
= nf_conntrack_htable_size
;
1948 if (old_size
== hashsize
) {
1949 nf_ct_free_hashtable(hash
, hashsize
);
1954 nf_conntrack_all_lock();
1955 write_seqcount_begin(&nf_conntrack_generation
);
1957 /* Lookups in the old hash might happen in parallel, which means we
1958 * might get false negatives during connection lookup. New connections
1959 * created because of a false negative won't make it into the hash
1960 * though since that required taking the locks.
1963 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1964 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
1965 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
1966 struct nf_conntrack_tuple_hash
, hnnode
);
1967 ct
= nf_ct_tuplehash_to_ctrack(h
);
1968 hlist_nulls_del_rcu(&h
->hnnode
);
1969 bucket
= __hash_conntrack(nf_ct_net(ct
),
1970 &h
->tuple
, hashsize
);
1971 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1974 old_size
= nf_conntrack_htable_size
;
1975 old_hash
= nf_conntrack_hash
;
1977 nf_conntrack_hash
= hash
;
1978 nf_conntrack_htable_size
= hashsize
;
1980 write_seqcount_end(&nf_conntrack_generation
);
1981 nf_conntrack_all_unlock();
1985 nf_ct_free_hashtable(old_hash
, old_size
);
1989 int nf_conntrack_set_hashsize(const char *val
, const struct kernel_param
*kp
)
1991 unsigned int hashsize
;
1994 if (current
->nsproxy
->net_ns
!= &init_net
)
1997 /* On boot, we can set this without any fancy locking. */
1998 if (!nf_conntrack_hash
)
1999 return param_set_uint(val
, kp
);
2001 rc
= kstrtouint(val
, 0, &hashsize
);
2005 return nf_conntrack_hash_resize(hashsize
);
2007 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
2009 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
2010 &nf_conntrack_htable_size
, 0600);
2012 static __always_inline
unsigned int total_extension_size(void)
2014 /* remember to add new extensions below */
2015 BUILD_BUG_ON(NF_CT_EXT_NUM
> 9);
2017 return sizeof(struct nf_ct_ext
) +
2018 sizeof(struct nf_conn_help
)
2019 #if IS_ENABLED(CONFIG_NF_NAT)
2020 + sizeof(struct nf_conn_nat
)
2022 + sizeof(struct nf_conn_seqadj
)
2023 + sizeof(struct nf_conn_acct
)
2024 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2025 + sizeof(struct nf_conntrack_ecache
)
2027 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2028 + sizeof(struct nf_conn_tstamp
)
2030 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2031 + sizeof(struct nf_conn_timeout
)
2033 #ifdef CONFIG_NF_CONNTRACK_LABELS
2034 + sizeof(struct nf_conn_labels
)
2036 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2037 + sizeof(struct nf_conn_synproxy
)
2042 int nf_conntrack_init_start(void)
2048 /* struct nf_ct_ext uses u8 to store offsets/size */
2049 BUILD_BUG_ON(total_extension_size() > 255u);
2051 seqcount_init(&nf_conntrack_generation
);
2053 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
2054 spin_lock_init(&nf_conntrack_locks
[i
]);
2056 if (!nf_conntrack_htable_size
) {
2057 /* Idea from tcp.c: use 1/16384 of memory.
2058 * On i386: 32MB machine has 512 buckets.
2059 * >= 1GB machines have 16384 buckets.
2060 * >= 4GB machines have 65536 buckets.
2062 nf_conntrack_htable_size
2063 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
2064 / sizeof(struct hlist_head
));
2065 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
2066 nf_conntrack_htable_size
= 65536;
2067 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
2068 nf_conntrack_htable_size
= 16384;
2069 if (nf_conntrack_htable_size
< 32)
2070 nf_conntrack_htable_size
= 32;
2072 /* Use a max. factor of four by default to get the same max as
2073 * with the old struct list_heads. When a table size is given
2074 * we use the old value of 8 to avoid reducing the max.
2079 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
2080 if (!nf_conntrack_hash
)
2083 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
2085 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
2086 sizeof(struct nf_conn
),
2088 SLAB_TYPESAFE_BY_RCU
| SLAB_HWCACHE_ALIGN
, NULL
);
2089 if (!nf_conntrack_cachep
)
2092 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
2093 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
2096 ret
= nf_conntrack_expect_init();
2100 ret
= nf_conntrack_acct_init();
2104 ret
= nf_conntrack_tstamp_init();
2108 ret
= nf_conntrack_ecache_init();
2112 ret
= nf_conntrack_timeout_init();
2116 ret
= nf_conntrack_helper_init();
2120 ret
= nf_conntrack_labels_init();
2124 ret
= nf_conntrack_seqadj_init();
2128 ret
= nf_conntrack_proto_init();
2132 conntrack_gc_work_init(&conntrack_gc_work
);
2133 queue_delayed_work(system_power_efficient_wq
, &conntrack_gc_work
.dwork
, HZ
);
2138 nf_conntrack_seqadj_fini();
2140 nf_conntrack_labels_fini();
2142 nf_conntrack_helper_fini();
2144 nf_conntrack_timeout_fini();
2146 nf_conntrack_ecache_fini();
2148 nf_conntrack_tstamp_fini();
2150 nf_conntrack_acct_fini();
2152 nf_conntrack_expect_fini();
2154 kmem_cache_destroy(nf_conntrack_cachep
);
2156 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
2160 void nf_conntrack_init_end(void)
2162 /* For use by REJECT target */
2163 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
2164 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
2168 * We need to use special "null" values, not used in hash table
2170 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2171 #define DYING_NULLS_VAL ((1<<30)+1)
2172 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
2174 int nf_conntrack_init_net(struct net
*net
)
2179 BUILD_BUG_ON(IP_CT_UNTRACKED
== IP_CT_NUMBER
);
2180 atomic_set(&net
->ct
.count
, 0);
2182 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
2183 if (!net
->ct
.pcpu_lists
)
2186 for_each_possible_cpu(cpu
) {
2187 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2189 spin_lock_init(&pcpu
->lock
);
2190 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
2191 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
2194 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
2196 goto err_pcpu_lists
;
2198 ret
= nf_conntrack_expect_pernet_init(net
);
2201 ret
= nf_conntrack_acct_pernet_init(net
);
2204 ret
= nf_conntrack_tstamp_pernet_init(net
);
2207 ret
= nf_conntrack_ecache_pernet_init(net
);
2210 ret
= nf_conntrack_helper_pernet_init(net
);
2213 ret
= nf_conntrack_proto_pernet_init(net
);
2219 nf_conntrack_helper_pernet_fini(net
);
2221 nf_conntrack_ecache_pernet_fini(net
);
2223 nf_conntrack_tstamp_pernet_fini(net
);
2225 nf_conntrack_acct_pernet_fini(net
);
2227 nf_conntrack_expect_pernet_fini(net
);
2229 free_percpu(net
->ct
.stat
);
2231 free_percpu(net
->ct
.pcpu_lists
);