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 #define NF_CONNTRACK_VERSION "0.5.0"
61 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
62 enum nf_nat_manip_type manip
,
63 const struct nlattr
*attr
) __read_mostly
;
64 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
66 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks
[CONNTRACK_LOCKS
];
67 EXPORT_SYMBOL_GPL(nf_conntrack_locks
);
69 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(nf_conntrack_expect_lock
);
70 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock
);
72 struct hlist_nulls_head
*nf_conntrack_hash __read_mostly
;
73 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
75 struct conntrack_gc_work
{
76 struct delayed_work dwork
;
83 static __read_mostly
struct kmem_cache
*nf_conntrack_cachep
;
84 static __read_mostly spinlock_t nf_conntrack_locks_all_lock
;
85 static __read_mostly
DEFINE_SPINLOCK(nf_conntrack_locks_all_lock
);
86 static __read_mostly
bool nf_conntrack_locks_all
;
88 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
89 #define GC_MAX_BUCKETS_DIV 128u
90 /* upper bound of full table scan */
91 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
92 /* desired ratio of entries found to be expired */
93 #define GC_EVICT_RATIO 50u
95 static struct conntrack_gc_work conntrack_gc_work
;
97 void nf_conntrack_lock(spinlock_t
*lock
) __acquires(lock
)
99 /* 1) Acquire the lock */
102 /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
103 * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
105 if (likely(smp_load_acquire(&nf_conntrack_locks_all
) == false))
108 /* fast path failed, unlock */
111 /* Slow path 1) get global lock */
112 spin_lock(&nf_conntrack_locks_all_lock
);
114 /* Slow path 2) get the lock we want */
117 /* Slow path 3) release the global lock */
118 spin_unlock(&nf_conntrack_locks_all_lock
);
120 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
122 static void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
124 h1
%= CONNTRACK_LOCKS
;
125 h2
%= CONNTRACK_LOCKS
;
126 spin_unlock(&nf_conntrack_locks
[h1
]);
128 spin_unlock(&nf_conntrack_locks
[h2
]);
131 /* return true if we need to recompute hashes (in case hash table was resized) */
132 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
133 unsigned int h2
, unsigned int sequence
)
135 h1
%= CONNTRACK_LOCKS
;
136 h2
%= CONNTRACK_LOCKS
;
138 nf_conntrack_lock(&nf_conntrack_locks
[h1
]);
140 spin_lock_nested(&nf_conntrack_locks
[h2
],
141 SINGLE_DEPTH_NESTING
);
143 nf_conntrack_lock(&nf_conntrack_locks
[h2
]);
144 spin_lock_nested(&nf_conntrack_locks
[h1
],
145 SINGLE_DEPTH_NESTING
);
147 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
148 nf_conntrack_double_unlock(h1
, h2
);
154 static void nf_conntrack_all_lock(void)
158 spin_lock(&nf_conntrack_locks_all_lock
);
160 nf_conntrack_locks_all
= true;
162 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++) {
163 spin_lock(&nf_conntrack_locks
[i
]);
165 /* This spin_unlock provides the "release" to ensure that
166 * nf_conntrack_locks_all==true is visible to everyone that
167 * acquired spin_lock(&nf_conntrack_locks[]).
169 spin_unlock(&nf_conntrack_locks
[i
]);
173 static void nf_conntrack_all_unlock(void)
175 /* All prior stores must be complete before we clear
176 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
177 * might observe the false value but not the entire
179 * It pairs with the smp_load_acquire() in nf_conntrack_lock()
181 smp_store_release(&nf_conntrack_locks_all
, false);
182 spin_unlock(&nf_conntrack_locks_all_lock
);
185 unsigned int nf_conntrack_htable_size __read_mostly
;
186 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
188 unsigned int nf_conntrack_max __read_mostly
;
189 seqcount_t nf_conntrack_generation __read_mostly
;
190 static unsigned int nf_conntrack_hash_rnd __read_mostly
;
192 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
,
193 const struct net
*net
)
198 get_random_once(&nf_conntrack_hash_rnd
, sizeof(nf_conntrack_hash_rnd
));
200 /* The direction must be ignored, so we hash everything up to the
201 * destination ports (which is a multiple of 4) and treat the last
202 * three bytes manually.
204 seed
= nf_conntrack_hash_rnd
^ net_hash_mix(net
);
205 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
206 return jhash2((u32
*)tuple
, n
, seed
^
207 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
208 tuple
->dst
.protonum
));
211 static u32
scale_hash(u32 hash
)
213 return reciprocal_scale(hash
, nf_conntrack_htable_size
);
216 static u32
__hash_conntrack(const struct net
*net
,
217 const struct nf_conntrack_tuple
*tuple
,
220 return reciprocal_scale(hash_conntrack_raw(tuple
, net
), size
);
223 static u32
hash_conntrack(const struct net
*net
,
224 const struct nf_conntrack_tuple
*tuple
)
226 return scale_hash(hash_conntrack_raw(tuple
, net
));
230 nf_ct_get_tuple(const struct sk_buff
*skb
,
232 unsigned int dataoff
,
236 struct nf_conntrack_tuple
*tuple
,
237 const struct nf_conntrack_l3proto
*l3proto
,
238 const struct nf_conntrack_l4proto
*l4proto
)
240 memset(tuple
, 0, sizeof(*tuple
));
242 tuple
->src
.l3num
= l3num
;
243 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
246 tuple
->dst
.protonum
= protonum
;
247 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
249 return l4proto
->pkt_to_tuple(skb
, dataoff
, net
, tuple
);
251 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
253 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
255 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
257 struct nf_conntrack_l3proto
*l3proto
;
258 struct nf_conntrack_l4proto
*l4proto
;
259 unsigned int protoff
;
265 l3proto
= __nf_ct_l3proto_find(l3num
);
266 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
267 if (ret
!= NF_ACCEPT
) {
272 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
274 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
,
280 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
283 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
284 const struct nf_conntrack_tuple
*orig
,
285 const struct nf_conntrack_l3proto
*l3proto
,
286 const struct nf_conntrack_l4proto
*l4proto
)
288 memset(inverse
, 0, sizeof(*inverse
));
290 inverse
->src
.l3num
= orig
->src
.l3num
;
291 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
294 inverse
->dst
.dir
= !orig
->dst
.dir
;
296 inverse
->dst
.protonum
= orig
->dst
.protonum
;
297 return l4proto
->invert_tuple(inverse
, orig
);
299 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
302 clean_from_lists(struct nf_conn
*ct
)
304 pr_debug("clean_from_lists(%p)\n", ct
);
305 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
306 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
308 /* Destroy all pending expectations */
309 nf_ct_remove_expectations(ct
);
312 /* must be called with local_bh_disable */
313 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
315 struct ct_pcpu
*pcpu
;
317 /* add this conntrack to the (per cpu) dying list */
318 ct
->cpu
= smp_processor_id();
319 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
321 spin_lock(&pcpu
->lock
);
322 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
324 spin_unlock(&pcpu
->lock
);
327 /* must be called with local_bh_disable */
328 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
330 struct ct_pcpu
*pcpu
;
332 /* add this conntrack to the (per cpu) unconfirmed list */
333 ct
->cpu
= smp_processor_id();
334 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
336 spin_lock(&pcpu
->lock
);
337 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
339 spin_unlock(&pcpu
->lock
);
342 /* must be called with local_bh_disable */
343 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
345 struct ct_pcpu
*pcpu
;
347 /* We overload first tuple to link into unconfirmed or dying list.*/
348 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
350 spin_lock(&pcpu
->lock
);
351 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
352 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
353 spin_unlock(&pcpu
->lock
);
356 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
358 /* Released via destroy_conntrack() */
359 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
360 const struct nf_conntrack_zone
*zone
,
363 struct nf_conn
*tmpl
, *p
;
365 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
) {
366 tmpl
= kzalloc(sizeof(*tmpl
) + NFCT_INFOMASK
, flags
);
371 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
373 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
374 tmpl
->proto
.tmpl_padto
= (char *)tmpl
- (char *)p
;
377 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
382 tmpl
->status
= IPS_TEMPLATE
;
383 write_pnet(&tmpl
->ct_net
, net
);
384 nf_ct_zone_add(tmpl
, zone
);
385 atomic_set(&tmpl
->ct_general
.use
, 0);
389 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
391 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
393 nf_ct_ext_destroy(tmpl
);
394 nf_ct_ext_free(tmpl
);
396 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
)
397 kfree((char *)tmpl
- tmpl
->proto
.tmpl_padto
);
401 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
404 destroy_conntrack(struct nf_conntrack
*nfct
)
406 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
407 struct nf_conntrack_l4proto
*l4proto
;
409 pr_debug("destroy_conntrack(%p)\n", ct
);
410 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
412 if (unlikely(nf_ct_is_template(ct
))) {
417 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
418 if (l4proto
->destroy
)
419 l4proto
->destroy(ct
);
424 /* Expectations will have been removed in clean_from_lists,
425 * except TFTP can create an expectation on the first packet,
426 * before connection is in the list, so we need to clean here,
429 nf_ct_remove_expectations(ct
);
431 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
436 nf_ct_put(ct
->master
);
438 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
439 nf_conntrack_free(ct
);
442 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
444 struct net
*net
= nf_ct_net(ct
);
445 unsigned int hash
, reply_hash
;
446 unsigned int sequence
;
448 nf_ct_helper_destroy(ct
);
452 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
453 hash
= hash_conntrack(net
,
454 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
455 reply_hash
= hash_conntrack(net
,
456 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
457 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
459 clean_from_lists(ct
);
460 nf_conntrack_double_unlock(hash
, reply_hash
);
462 nf_ct_add_to_dying_list(ct
);
467 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
469 struct nf_conn_tstamp
*tstamp
;
471 if (test_and_set_bit(IPS_DYING_BIT
, &ct
->status
))
474 tstamp
= nf_conn_tstamp_find(ct
);
475 if (tstamp
&& tstamp
->stop
== 0)
476 tstamp
->stop
= ktime_get_real_ns();
478 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
479 portid
, report
) < 0) {
480 /* destroy event was not delivered. nf_ct_put will
481 * be done by event cache worker on redelivery.
483 nf_ct_delete_from_lists(ct
);
484 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
488 nf_conntrack_ecache_work(nf_ct_net(ct
));
489 nf_ct_delete_from_lists(ct
);
493 EXPORT_SYMBOL_GPL(nf_ct_delete
);
496 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
497 const struct nf_conntrack_tuple
*tuple
,
498 const struct nf_conntrack_zone
*zone
,
499 const struct net
*net
)
501 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
503 /* A conntrack can be recreated with the equal tuple,
504 * so we need to check that the conntrack is confirmed
506 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
507 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
508 nf_ct_is_confirmed(ct
) &&
509 net_eq(net
, nf_ct_net(ct
));
512 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
513 static void nf_ct_gc_expired(struct nf_conn
*ct
)
515 if (!atomic_inc_not_zero(&ct
->ct_general
.use
))
518 if (nf_ct_should_gc(ct
))
526 * - Caller must take a reference on returned object
527 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
529 static struct nf_conntrack_tuple_hash
*
530 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
531 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
533 struct nf_conntrack_tuple_hash
*h
;
534 struct hlist_nulls_head
*ct_hash
;
535 struct hlist_nulls_node
*n
;
536 unsigned int bucket
, hsize
;
539 nf_conntrack_get_ht(&ct_hash
, &hsize
);
540 bucket
= reciprocal_scale(hash
, hsize
);
542 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[bucket
], hnnode
) {
545 ct
= nf_ct_tuplehash_to_ctrack(h
);
546 if (nf_ct_is_expired(ct
)) {
547 nf_ct_gc_expired(ct
);
551 if (nf_ct_is_dying(ct
))
554 if (nf_ct_key_equal(h
, tuple
, zone
, net
))
558 * if the nulls value we got at the end of this lookup is
559 * not the expected one, we must restart lookup.
560 * We probably met an item that was moved to another chain.
562 if (get_nulls_value(n
) != bucket
) {
563 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
570 /* Find a connection corresponding to a tuple. */
571 static struct nf_conntrack_tuple_hash
*
572 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
573 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
575 struct nf_conntrack_tuple_hash
*h
;
580 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
582 ct
= nf_ct_tuplehash_to_ctrack(h
);
583 if (unlikely(nf_ct_is_dying(ct
) ||
584 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
587 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
, net
))) {
598 struct nf_conntrack_tuple_hash
*
599 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
600 const struct nf_conntrack_tuple
*tuple
)
602 return __nf_conntrack_find_get(net
, zone
, tuple
,
603 hash_conntrack_raw(tuple
, net
));
605 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
607 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
609 unsigned int reply_hash
)
611 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
612 &nf_conntrack_hash
[hash
]);
613 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
614 &nf_conntrack_hash
[reply_hash
]);
618 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
620 const struct nf_conntrack_zone
*zone
;
621 struct net
*net
= nf_ct_net(ct
);
622 unsigned int hash
, reply_hash
;
623 struct nf_conntrack_tuple_hash
*h
;
624 struct hlist_nulls_node
*n
;
625 unsigned int sequence
;
627 zone
= nf_ct_zone(ct
);
631 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
632 hash
= hash_conntrack(net
,
633 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
634 reply_hash
= hash_conntrack(net
,
635 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
636 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
638 /* See if there's one in the list already, including reverse */
639 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
640 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
644 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
645 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
650 /* The caller holds a reference to this object */
651 atomic_set(&ct
->ct_general
.use
, 2);
652 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
653 nf_conntrack_double_unlock(hash
, reply_hash
);
654 NF_CT_STAT_INC(net
, insert
);
659 nf_conntrack_double_unlock(hash
, reply_hash
);
660 NF_CT_STAT_INC(net
, insert_failed
);
664 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
666 static inline void nf_ct_acct_update(struct nf_conn
*ct
,
667 enum ip_conntrack_info ctinfo
,
670 struct nf_conn_acct
*acct
;
672 acct
= nf_conn_acct_find(ct
);
674 struct nf_conn_counter
*counter
= acct
->counter
;
676 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
677 atomic64_add(len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
681 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
682 const struct nf_conn
*loser_ct
)
684 struct nf_conn_acct
*acct
;
686 acct
= nf_conn_acct_find(loser_ct
);
688 struct nf_conn_counter
*counter
= acct
->counter
;
691 /* u32 should be fine since we must have seen one packet. */
692 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
693 nf_ct_acct_update(ct
, ctinfo
, bytes
);
697 /* Resolve race on insertion if this protocol allows this. */
698 static int nf_ct_resolve_clash(struct net
*net
, struct sk_buff
*skb
,
699 enum ip_conntrack_info ctinfo
,
700 struct nf_conntrack_tuple_hash
*h
)
702 /* This is the conntrack entry already in hashes that won race. */
703 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
704 struct nf_conntrack_l4proto
*l4proto
;
706 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
707 if (l4proto
->allow_clash
&&
708 ((ct
->status
& IPS_NAT_DONE_MASK
) == 0) &&
709 !nf_ct_is_dying(ct
) &&
710 atomic_inc_not_zero(&ct
->ct_general
.use
)) {
711 enum ip_conntrack_info oldinfo
;
712 struct nf_conn
*loser_ct
= nf_ct_get(skb
, &oldinfo
);
714 nf_ct_acct_merge(ct
, ctinfo
, loser_ct
);
715 nf_conntrack_put(&loser_ct
->ct_general
);
716 nf_ct_set(skb
, ct
, oldinfo
);
719 NF_CT_STAT_INC(net
, drop
);
723 /* Confirm a connection given skb; places it in hash table */
725 __nf_conntrack_confirm(struct sk_buff
*skb
)
727 const struct nf_conntrack_zone
*zone
;
728 unsigned int hash
, reply_hash
;
729 struct nf_conntrack_tuple_hash
*h
;
731 struct nf_conn_help
*help
;
732 struct nf_conn_tstamp
*tstamp
;
733 struct hlist_nulls_node
*n
;
734 enum ip_conntrack_info ctinfo
;
736 unsigned int sequence
;
739 ct
= nf_ct_get(skb
, &ctinfo
);
742 /* ipt_REJECT uses nf_conntrack_attach to attach related
743 ICMP/TCP RST packets in other direction. Actual packet
744 which created connection will be IP_CT_NEW or for an
745 expected connection, IP_CT_RELATED. */
746 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
749 zone
= nf_ct_zone(ct
);
753 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
754 /* reuse the hash saved before */
755 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
756 hash
= scale_hash(hash
);
757 reply_hash
= hash_conntrack(net
,
758 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
760 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
762 /* We're not in hash table, and we refuse to set up related
763 * connections for unconfirmed conns. But packet copies and
764 * REJECT will give spurious warnings here.
766 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
768 /* No external references means no one else could have
771 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
772 pr_debug("Confirming conntrack %p\n", ct
);
773 /* We have to check the DYING flag after unlink to prevent
774 * a race against nf_ct_get_next_corpse() possibly called from
775 * user context, else we insert an already 'dead' hash, blocking
776 * further use of that particular connection -JM.
778 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
780 if (unlikely(nf_ct_is_dying(ct
))) {
781 nf_ct_add_to_dying_list(ct
);
785 /* See if there's one in the list already, including reverse:
786 NAT could have grabbed it without realizing, since we're
787 not in the hash. If there is, we lost race. */
788 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
789 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
793 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
794 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
798 /* Timer relative to confirmation time, not original
799 setting time, otherwise we'd get timer wrap in
800 weird delay cases. */
801 ct
->timeout
+= nfct_time_stamp
;
802 atomic_inc(&ct
->ct_general
.use
);
803 ct
->status
|= IPS_CONFIRMED
;
805 /* set conntrack timestamp, if enabled. */
806 tstamp
= nf_conn_tstamp_find(ct
);
808 if (skb
->tstamp
== 0)
809 __net_timestamp(skb
);
811 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
813 /* Since the lookup is lockless, hash insertion must be done after
814 * starting the timer and setting the CONFIRMED bit. The RCU barriers
815 * guarantee that no other CPU can find the conntrack before the above
816 * stores are visible.
818 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
819 nf_conntrack_double_unlock(hash
, reply_hash
);
822 help
= nfct_help(ct
);
823 if (help
&& help
->helper
)
824 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
826 nf_conntrack_event_cache(master_ct(ct
) ?
827 IPCT_RELATED
: IPCT_NEW
, ct
);
831 nf_ct_add_to_dying_list(ct
);
832 ret
= nf_ct_resolve_clash(net
, skb
, ctinfo
, h
);
834 nf_conntrack_double_unlock(hash
, reply_hash
);
835 NF_CT_STAT_INC(net
, insert_failed
);
839 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
841 /* Returns true if a connection correspondings to the tuple (required
844 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
845 const struct nf_conn
*ignored_conntrack
)
847 struct net
*net
= nf_ct_net(ignored_conntrack
);
848 const struct nf_conntrack_zone
*zone
;
849 struct nf_conntrack_tuple_hash
*h
;
850 struct hlist_nulls_head
*ct_hash
;
851 unsigned int hash
, hsize
;
852 struct hlist_nulls_node
*n
;
855 zone
= nf_ct_zone(ignored_conntrack
);
859 nf_conntrack_get_ht(&ct_hash
, &hsize
);
860 hash
= __hash_conntrack(net
, tuple
, hsize
);
862 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
863 ct
= nf_ct_tuplehash_to_ctrack(h
);
865 if (ct
== ignored_conntrack
)
868 if (nf_ct_is_expired(ct
)) {
869 nf_ct_gc_expired(ct
);
873 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
874 NF_CT_STAT_INC_ATOMIC(net
, found
);
880 if (get_nulls_value(n
) != hash
) {
881 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
889 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
891 #define NF_CT_EVICTION_RANGE 8
893 /* There's a small race here where we may free a just-assured
894 connection. Too bad: we're in trouble anyway. */
895 static unsigned int early_drop_list(struct net
*net
,
896 struct hlist_nulls_head
*head
)
898 struct nf_conntrack_tuple_hash
*h
;
899 struct hlist_nulls_node
*n
;
900 unsigned int drops
= 0;
903 hlist_nulls_for_each_entry_rcu(h
, n
, head
, hnnode
) {
904 tmp
= nf_ct_tuplehash_to_ctrack(h
);
906 if (nf_ct_is_expired(tmp
)) {
907 nf_ct_gc_expired(tmp
);
911 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
912 !net_eq(nf_ct_net(tmp
), net
) ||
916 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
919 /* kill only if still in same netns -- might have moved due to
920 * SLAB_TYPESAFE_BY_RCU rules.
922 * We steal the timer reference. If that fails timer has
923 * already fired or someone else deleted it. Just drop ref
924 * and move to next entry.
926 if (net_eq(nf_ct_net(tmp
), net
) &&
927 nf_ct_is_confirmed(tmp
) &&
928 nf_ct_delete(tmp
, 0, 0))
937 static noinline
int early_drop(struct net
*net
, unsigned int _hash
)
941 for (i
= 0; i
< NF_CT_EVICTION_RANGE
; i
++) {
942 struct hlist_nulls_head
*ct_hash
;
943 unsigned int hash
, hsize
, drops
;
946 nf_conntrack_get_ht(&ct_hash
, &hsize
);
947 hash
= reciprocal_scale(_hash
++, hsize
);
949 drops
= early_drop_list(net
, &ct_hash
[hash
]);
953 NF_CT_STAT_ADD_ATOMIC(net
, early_drop
, drops
);
961 static bool gc_worker_skip_ct(const struct nf_conn
*ct
)
963 return !nf_ct_is_confirmed(ct
) || nf_ct_is_dying(ct
);
966 static bool gc_worker_can_early_drop(const struct nf_conn
*ct
)
968 const struct nf_conntrack_l4proto
*l4proto
;
970 if (!test_bit(IPS_ASSURED_BIT
, &ct
->status
))
973 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
974 if (l4proto
->can_early_drop
&& l4proto
->can_early_drop(ct
))
980 static void gc_worker(struct work_struct
*work
)
982 unsigned int min_interval
= max(HZ
/ GC_MAX_BUCKETS_DIV
, 1u);
983 unsigned int i
, goal
, buckets
= 0, expired_count
= 0;
984 unsigned int nf_conntrack_max95
= 0;
985 struct conntrack_gc_work
*gc_work
;
986 unsigned int ratio
, scanned
= 0;
987 unsigned long next_run
;
989 gc_work
= container_of(work
, struct conntrack_gc_work
, dwork
.work
);
991 goal
= nf_conntrack_htable_size
/ GC_MAX_BUCKETS_DIV
;
992 i
= gc_work
->last_bucket
;
993 if (gc_work
->early_drop
)
994 nf_conntrack_max95
= nf_conntrack_max
/ 100u * 95u;
997 struct nf_conntrack_tuple_hash
*h
;
998 struct hlist_nulls_head
*ct_hash
;
999 struct hlist_nulls_node
*n
;
1000 unsigned int hashsz
;
1001 struct nf_conn
*tmp
;
1006 nf_conntrack_get_ht(&ct_hash
, &hashsz
);
1010 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[i
], hnnode
) {
1013 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1016 if (nf_ct_is_expired(tmp
)) {
1017 nf_ct_gc_expired(tmp
);
1022 if (nf_conntrack_max95
== 0 || gc_worker_skip_ct(tmp
))
1025 net
= nf_ct_net(tmp
);
1026 if (atomic_read(&net
->ct
.count
) < nf_conntrack_max95
)
1029 /* need to take reference to avoid possible races */
1030 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1033 if (gc_worker_skip_ct(tmp
)) {
1038 if (gc_worker_can_early_drop(tmp
))
1044 /* could check get_nulls_value() here and restart if ct
1045 * was moved to another chain. But given gc is best-effort
1046 * we will just continue with next hash slot.
1049 cond_resched_rcu_qs();
1050 } while (++buckets
< goal
);
1052 if (gc_work
->exiting
)
1056 * Eviction will normally happen from the packet path, and not
1057 * from this gc worker.
1059 * This worker is only here to reap expired entries when system went
1060 * idle after a busy period.
1062 * The heuristics below are supposed to balance conflicting goals:
1064 * 1. Minimize time until we notice a stale entry
1065 * 2. Maximize scan intervals to not waste cycles
1067 * Normally, expire ratio will be close to 0.
1069 * As soon as a sizeable fraction of the entries have expired
1070 * increase scan frequency.
1072 ratio
= scanned
? expired_count
* 100 / scanned
: 0;
1073 if (ratio
> GC_EVICT_RATIO
) {
1074 gc_work
->next_gc_run
= min_interval
;
1076 unsigned int max
= GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
;
1078 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
) == 0);
1080 gc_work
->next_gc_run
+= min_interval
;
1081 if (gc_work
->next_gc_run
> max
)
1082 gc_work
->next_gc_run
= max
;
1085 next_run
= gc_work
->next_gc_run
;
1086 gc_work
->last_bucket
= i
;
1087 gc_work
->early_drop
= false;
1088 queue_delayed_work(system_long_wq
, &gc_work
->dwork
, next_run
);
1091 static void conntrack_gc_work_init(struct conntrack_gc_work
*gc_work
)
1093 INIT_DELAYED_WORK(&gc_work
->dwork
, gc_worker
);
1094 gc_work
->next_gc_run
= HZ
;
1095 gc_work
->exiting
= false;
1098 static struct nf_conn
*
1099 __nf_conntrack_alloc(struct net
*net
,
1100 const struct nf_conntrack_zone
*zone
,
1101 const struct nf_conntrack_tuple
*orig
,
1102 const struct nf_conntrack_tuple
*repl
,
1103 gfp_t gfp
, u32 hash
)
1107 /* We don't want any race condition at early drop stage */
1108 atomic_inc(&net
->ct
.count
);
1110 if (nf_conntrack_max
&&
1111 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
1112 if (!early_drop(net
, hash
)) {
1113 if (!conntrack_gc_work
.early_drop
)
1114 conntrack_gc_work
.early_drop
= true;
1115 atomic_dec(&net
->ct
.count
);
1116 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1117 return ERR_PTR(-ENOMEM
);
1122 * Do not use kmem_cache_zalloc(), as this cache uses
1123 * SLAB_TYPESAFE_BY_RCU.
1125 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
1129 spin_lock_init(&ct
->lock
);
1130 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
1131 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
1132 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
1133 /* save hash for reusing when confirming */
1134 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
1136 write_pnet(&ct
->ct_net
, net
);
1137 memset(&ct
->__nfct_init_offset
[0], 0,
1138 offsetof(struct nf_conn
, proto
) -
1139 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
1141 nf_ct_zone_add(ct
, zone
);
1143 /* Because we use RCU lookups, we set ct_general.use to zero before
1144 * this is inserted in any list.
1146 atomic_set(&ct
->ct_general
.use
, 0);
1149 atomic_dec(&net
->ct
.count
);
1150 return ERR_PTR(-ENOMEM
);
1153 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
1154 const struct nf_conntrack_zone
*zone
,
1155 const struct nf_conntrack_tuple
*orig
,
1156 const struct nf_conntrack_tuple
*repl
,
1159 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
1161 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
1163 void nf_conntrack_free(struct nf_conn
*ct
)
1165 struct net
*net
= nf_ct_net(ct
);
1167 /* A freed object has refcnt == 0, that's
1168 * the golden rule for SLAB_TYPESAFE_BY_RCU
1170 NF_CT_ASSERT(atomic_read(&ct
->ct_general
.use
) == 0);
1172 nf_ct_ext_destroy(ct
);
1174 kmem_cache_free(nf_conntrack_cachep
, ct
);
1175 smp_mb__before_atomic();
1176 atomic_dec(&net
->ct
.count
);
1178 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
1181 /* Allocate a new conntrack: we return -ENOMEM if classification
1182 failed due to stress. Otherwise it really is unclassifiable. */
1183 static noinline
struct nf_conntrack_tuple_hash
*
1184 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
1185 const struct nf_conntrack_tuple
*tuple
,
1186 struct nf_conntrack_l3proto
*l3proto
,
1187 struct nf_conntrack_l4proto
*l4proto
,
1188 struct sk_buff
*skb
,
1189 unsigned int dataoff
, u32 hash
)
1192 struct nf_conn_help
*help
;
1193 struct nf_conntrack_tuple repl_tuple
;
1194 struct nf_conntrack_ecache
*ecache
;
1195 struct nf_conntrack_expect
*exp
= NULL
;
1196 const struct nf_conntrack_zone
*zone
;
1197 struct nf_conn_timeout
*timeout_ext
;
1198 struct nf_conntrack_zone tmp
;
1199 unsigned int *timeouts
;
1201 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
1202 pr_debug("Can't invert tuple.\n");
1206 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1207 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
1210 return (struct nf_conntrack_tuple_hash
*)ct
;
1212 if (!nf_ct_add_synproxy(ct
, tmpl
)) {
1213 nf_conntrack_free(ct
);
1214 return ERR_PTR(-ENOMEM
);
1217 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1219 timeouts
= nf_ct_timeout_data(timeout_ext
);
1220 if (unlikely(!timeouts
))
1221 timeouts
= l4proto
->get_timeouts(net
);
1223 timeouts
= l4proto
->get_timeouts(net
);
1226 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
1227 nf_conntrack_free(ct
);
1228 pr_debug("can't track with proto module\n");
1233 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1236 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1237 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1238 nf_ct_labels_ext_add(ct
);
1240 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1241 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1242 ecache
? ecache
->expmask
: 0,
1246 if (net
->ct
.expect_count
) {
1247 spin_lock(&nf_conntrack_expect_lock
);
1248 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1250 pr_debug("expectation arrives ct=%p exp=%p\n",
1252 /* Welcome, Mr. Bond. We've been expecting you... */
1253 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1254 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1255 ct
->master
= exp
->master
;
1257 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
1260 rcu_assign_pointer(help
->helper
, exp
->helper
);
1263 #ifdef CONFIG_NF_CONNTRACK_MARK
1264 ct
->mark
= exp
->master
->mark
;
1266 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1267 ct
->secmark
= exp
->master
->secmark
;
1269 NF_CT_STAT_INC(net
, expect_new
);
1271 spin_unlock(&nf_conntrack_expect_lock
);
1274 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1276 /* Now it is inserted into the unconfirmed list, bump refcount */
1277 nf_conntrack_get(&ct
->ct_general
);
1278 nf_ct_add_to_unconfirmed_list(ct
);
1284 exp
->expectfn(ct
, exp
);
1285 nf_ct_expect_put(exp
);
1288 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1291 /* On success, returns 0, sets skb->_nfct | ctinfo */
1293 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1294 struct sk_buff
*skb
,
1295 unsigned int dataoff
,
1298 struct nf_conntrack_l3proto
*l3proto
,
1299 struct nf_conntrack_l4proto
*l4proto
)
1301 const struct nf_conntrack_zone
*zone
;
1302 struct nf_conntrack_tuple tuple
;
1303 struct nf_conntrack_tuple_hash
*h
;
1304 enum ip_conntrack_info ctinfo
;
1305 struct nf_conntrack_zone tmp
;
1309 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1310 dataoff
, l3num
, protonum
, net
, &tuple
, l3proto
,
1312 pr_debug("Can't get tuple\n");
1316 /* look for tuple match */
1317 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1318 hash
= hash_conntrack_raw(&tuple
, net
);
1319 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1321 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1322 skb
, dataoff
, hash
);
1328 ct
= nf_ct_tuplehash_to_ctrack(h
);
1330 /* It exists; we have (non-exclusive) reference. */
1331 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1332 ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1334 /* Once we've had two way comms, always ESTABLISHED. */
1335 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1336 pr_debug("normal packet for %p\n", ct
);
1337 ctinfo
= IP_CT_ESTABLISHED
;
1338 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1339 pr_debug("related packet for %p\n", ct
);
1340 ctinfo
= IP_CT_RELATED
;
1342 pr_debug("new packet for %p\n", ct
);
1346 nf_ct_set(skb
, ct
, ctinfo
);
1351 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1352 struct sk_buff
*skb
)
1354 struct nf_conn
*ct
, *tmpl
;
1355 enum ip_conntrack_info ctinfo
;
1356 struct nf_conntrack_l3proto
*l3proto
;
1357 struct nf_conntrack_l4proto
*l4proto
;
1358 unsigned int *timeouts
;
1359 unsigned int dataoff
;
1363 tmpl
= nf_ct_get(skb
, &ctinfo
);
1364 if (tmpl
|| ctinfo
== IP_CT_UNTRACKED
) {
1365 /* Previously seen (loopback or untracked)? Ignore. */
1366 if ((tmpl
&& !nf_ct_is_template(tmpl
)) ||
1367 ctinfo
== IP_CT_UNTRACKED
) {
1368 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1374 /* rcu_read_lock()ed by nf_hook_thresh */
1375 l3proto
= __nf_ct_l3proto_find(pf
);
1376 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1377 &dataoff
, &protonum
);
1379 pr_debug("not prepared to track yet or error occurred\n");
1380 NF_CT_STAT_INC_ATOMIC(net
, error
);
1381 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1386 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1388 /* It may be an special packet, error, unclean...
1389 * inverse of the return code tells to the netfilter
1390 * core what to do with the packet. */
1391 if (l4proto
->error
!= NULL
) {
1392 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, pf
, hooknum
);
1394 NF_CT_STAT_INC_ATOMIC(net
, error
);
1395 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1399 /* ICMP[v6] protocol trackers may assign one conntrack. */
1404 ret
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1407 /* Too stressed to deal. */
1408 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1413 ct
= nf_ct_get(skb
, &ctinfo
);
1415 /* Not valid part of a connection */
1416 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1421 /* Decide what timeout policy we want to apply to this flow. */
1422 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1424 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1426 /* Invalid: inverse of the return code tells
1427 * the netfilter core what to do */
1428 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1429 nf_conntrack_put(&ct
->ct_general
);
1431 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1432 if (ret
== -NF_DROP
)
1433 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1434 /* Special case: TCP tracker reports an attempt to reopen a
1435 * closed/aborted connection. We have to go back and create a
1438 if (ret
== -NF_REPEAT
)
1444 if (ctinfo
== IP_CT_ESTABLISHED_REPLY
&&
1445 !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1446 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1453 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1455 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1456 const struct nf_conntrack_tuple
*orig
)
1461 ret
= nf_ct_invert_tuple(inverse
, orig
,
1462 __nf_ct_l3proto_find(orig
->src
.l3num
),
1463 __nf_ct_l4proto_find(orig
->src
.l3num
,
1464 orig
->dst
.protonum
));
1468 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1470 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1471 implicitly racy: see __nf_conntrack_confirm */
1472 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1473 const struct nf_conntrack_tuple
*newreply
)
1475 struct nf_conn_help
*help
= nfct_help(ct
);
1477 /* Should be unconfirmed, so not in hash table yet */
1478 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1480 pr_debug("Altering reply tuple of %p to ", ct
);
1481 nf_ct_dump_tuple(newreply
);
1483 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1484 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1488 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1491 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1493 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1494 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1495 enum ip_conntrack_info ctinfo
,
1496 const struct sk_buff
*skb
,
1497 unsigned long extra_jiffies
,
1502 /* Only update if this is not a fixed timeout */
1503 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1506 /* If not in hash table, timer will not be active yet */
1507 if (nf_ct_is_confirmed(ct
))
1508 extra_jiffies
+= nfct_time_stamp
;
1510 ct
->timeout
= extra_jiffies
;
1513 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1515 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1517 bool nf_ct_kill_acct(struct nf_conn
*ct
,
1518 enum ip_conntrack_info ctinfo
,
1519 const struct sk_buff
*skb
)
1521 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1523 return nf_ct_delete(ct
, 0, 0);
1525 EXPORT_SYMBOL_GPL(nf_ct_kill_acct
);
1527 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1529 #include <linux/netfilter/nfnetlink.h>
1530 #include <linux/netfilter/nfnetlink_conntrack.h>
1531 #include <linux/mutex.h>
1533 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1534 * in ip_conntrack_core, since we don't want the protocols to autoload
1535 * or depend on ctnetlink */
1536 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1537 const struct nf_conntrack_tuple
*tuple
)
1539 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1540 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1541 goto nla_put_failure
;
1547 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1549 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1550 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1551 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1553 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1555 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1556 struct nf_conntrack_tuple
*t
)
1558 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1561 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1562 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1566 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1568 int nf_ct_port_nlattr_tuple_size(void)
1570 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1572 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1575 /* Used by ipt_REJECT and ip6t_REJECT. */
1576 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1579 enum ip_conntrack_info ctinfo
;
1581 /* This ICMP is in reverse direction to the packet which caused it */
1582 ct
= nf_ct_get(skb
, &ctinfo
);
1583 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1584 ctinfo
= IP_CT_RELATED_REPLY
;
1586 ctinfo
= IP_CT_RELATED
;
1588 /* Attach to new skbuff, and increment count */
1589 nf_ct_set(nskb
, ct
, ctinfo
);
1590 nf_conntrack_get(skb_nfct(nskb
));
1593 /* Bring out ya dead! */
1594 static struct nf_conn
*
1595 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
1596 void *data
, unsigned int *bucket
)
1598 struct nf_conntrack_tuple_hash
*h
;
1600 struct hlist_nulls_node
*n
;
1603 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1604 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1606 nf_conntrack_lock(lockp
);
1607 if (*bucket
< nf_conntrack_htable_size
) {
1608 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
1609 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1611 ct
= nf_ct_tuplehash_to_ctrack(h
);
1623 atomic_inc(&ct
->ct_general
.use
);
1629 static void nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
),
1630 void *data
, u32 portid
, int report
)
1632 unsigned int bucket
= 0, sequence
;
1638 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
1640 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
1641 /* Time to push up daises... */
1643 nf_ct_delete(ct
, portid
, report
);
1648 if (!read_seqcount_retry(&nf_conntrack_generation
, sequence
))
1655 int (*iter
)(struct nf_conn
*i
, void *data
);
1660 static int iter_net_only(struct nf_conn
*i
, void *data
)
1662 struct iter_data
*d
= data
;
1664 if (!net_eq(d
->net
, nf_ct_net(i
)))
1667 return d
->iter(i
, d
->data
);
1671 __nf_ct_unconfirmed_destroy(struct net
*net
)
1675 for_each_possible_cpu(cpu
) {
1676 struct nf_conntrack_tuple_hash
*h
;
1677 struct hlist_nulls_node
*n
;
1678 struct ct_pcpu
*pcpu
;
1680 pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1682 spin_lock_bh(&pcpu
->lock
);
1683 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1686 ct
= nf_ct_tuplehash_to_ctrack(h
);
1688 /* we cannot call iter() on unconfirmed list, the
1689 * owning cpu can reallocate ct->ext at any time.
1691 set_bit(IPS_DYING_BIT
, &ct
->status
);
1693 spin_unlock_bh(&pcpu
->lock
);
1698 void nf_ct_iterate_cleanup_net(struct net
*net
,
1699 int (*iter
)(struct nf_conn
*i
, void *data
),
1700 void *data
, u32 portid
, int report
)
1706 if (atomic_read(&net
->ct
.count
) == 0)
1709 __nf_ct_unconfirmed_destroy(net
);
1717 nf_ct_iterate_cleanup(iter_net_only
, &d
, portid
, report
);
1719 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net
);
1722 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
1723 * @iter: callback to invoke for each conntrack
1724 * @data: data to pass to @iter
1726 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
1727 * unconfirmed list as dying (so they will not be inserted into
1730 * Can only be called in module exit path.
1733 nf_ct_iterate_destroy(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
1739 if (atomic_read(&net
->ct
.count
) == 0)
1741 __nf_ct_unconfirmed_destroy(net
);
1745 /* Need to wait for netns cleanup worker to finish, if its
1746 * running -- it might have deleted a net namespace from
1747 * the global list, so our __nf_ct_unconfirmed_destroy() might
1748 * not have affected all namespaces.
1752 /* a conntrack could have been unlinked from unconfirmed list
1753 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
1754 * This makes sure its inserted into conntrack table.
1758 nf_ct_iterate_cleanup(iter
, data
, 0, 0);
1760 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy
);
1762 static int kill_all(struct nf_conn
*i
, void *data
)
1764 return net_eq(nf_ct_net(i
), data
);
1767 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1769 if (is_vmalloc_addr(hash
))
1772 free_pages((unsigned long)hash
,
1773 get_order(sizeof(struct hlist_head
) * size
));
1775 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1777 void nf_conntrack_cleanup_start(void)
1779 conntrack_gc_work
.exiting
= true;
1780 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1783 void nf_conntrack_cleanup_end(void)
1785 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1787 cancel_delayed_work_sync(&conntrack_gc_work
.dwork
);
1788 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1790 nf_conntrack_proto_fini();
1791 nf_conntrack_seqadj_fini();
1792 nf_conntrack_labels_fini();
1793 nf_conntrack_helper_fini();
1794 nf_conntrack_timeout_fini();
1795 nf_conntrack_ecache_fini();
1796 nf_conntrack_tstamp_fini();
1797 nf_conntrack_acct_fini();
1798 nf_conntrack_expect_fini();
1800 kmem_cache_destroy(nf_conntrack_cachep
);
1804 * Mishearing the voices in his head, our hero wonders how he's
1805 * supposed to kill the mall.
1807 void nf_conntrack_cleanup_net(struct net
*net
)
1811 list_add(&net
->exit_list
, &single
);
1812 nf_conntrack_cleanup_net_list(&single
);
1815 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1821 * This makes sure all current packets have passed through
1822 * netfilter framework. Roll on, two-stage module
1828 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1829 nf_ct_iterate_cleanup(kill_all
, net
, 0, 0);
1830 if (atomic_read(&net
->ct
.count
) != 0)
1835 goto i_see_dead_people
;
1838 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1839 nf_conntrack_proto_pernet_fini(net
);
1840 nf_conntrack_helper_pernet_fini(net
);
1841 nf_conntrack_ecache_pernet_fini(net
);
1842 nf_conntrack_tstamp_pernet_fini(net
);
1843 nf_conntrack_acct_pernet_fini(net
);
1844 nf_conntrack_expect_pernet_fini(net
);
1845 free_percpu(net
->ct
.stat
);
1846 free_percpu(net
->ct
.pcpu_lists
);
1850 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1852 struct hlist_nulls_head
*hash
;
1853 unsigned int nr_slots
, i
;
1856 if (*sizep
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
1859 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1860 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1862 if (nr_slots
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
1865 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1866 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1872 for (i
= 0; i
< nr_slots
; i
++)
1873 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1877 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1879 int nf_conntrack_hash_resize(unsigned int hashsize
)
1882 unsigned int old_size
;
1883 struct hlist_nulls_head
*hash
, *old_hash
;
1884 struct nf_conntrack_tuple_hash
*h
;
1890 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1894 old_size
= nf_conntrack_htable_size
;
1895 if (old_size
== hashsize
) {
1896 nf_ct_free_hashtable(hash
, hashsize
);
1901 nf_conntrack_all_lock();
1902 write_seqcount_begin(&nf_conntrack_generation
);
1904 /* Lookups in the old hash might happen in parallel, which means we
1905 * might get false negatives during connection lookup. New connections
1906 * created because of a false negative won't make it into the hash
1907 * though since that required taking the locks.
1910 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1911 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
1912 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
1913 struct nf_conntrack_tuple_hash
, hnnode
);
1914 ct
= nf_ct_tuplehash_to_ctrack(h
);
1915 hlist_nulls_del_rcu(&h
->hnnode
);
1916 bucket
= __hash_conntrack(nf_ct_net(ct
),
1917 &h
->tuple
, hashsize
);
1918 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1921 old_size
= nf_conntrack_htable_size
;
1922 old_hash
= nf_conntrack_hash
;
1924 nf_conntrack_hash
= hash
;
1925 nf_conntrack_htable_size
= hashsize
;
1927 write_seqcount_end(&nf_conntrack_generation
);
1928 nf_conntrack_all_unlock();
1932 nf_ct_free_hashtable(old_hash
, old_size
);
1936 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1938 unsigned int hashsize
;
1941 if (current
->nsproxy
->net_ns
!= &init_net
)
1944 /* On boot, we can set this without any fancy locking. */
1945 if (!nf_conntrack_htable_size
)
1946 return param_set_uint(val
, kp
);
1948 rc
= kstrtouint(val
, 0, &hashsize
);
1952 return nf_conntrack_hash_resize(hashsize
);
1954 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1956 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1957 &nf_conntrack_htable_size
, 0600);
1959 static __always_inline
unsigned int total_extension_size(void)
1961 /* remember to add new extensions below */
1962 BUILD_BUG_ON(NF_CT_EXT_NUM
> 9);
1964 return sizeof(struct nf_ct_ext
) +
1965 sizeof(struct nf_conn_help
)
1966 #if IS_ENABLED(CONFIG_NF_NAT)
1967 + sizeof(struct nf_conn_nat
)
1969 + sizeof(struct nf_conn_seqadj
)
1970 + sizeof(struct nf_conn_acct
)
1971 #ifdef CONFIG_NF_CONNTRACK_EVENTS
1972 + sizeof(struct nf_conntrack_ecache
)
1974 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
1975 + sizeof(struct nf_conn_tstamp
)
1977 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1978 + sizeof(struct nf_conn_timeout
)
1980 #ifdef CONFIG_NF_CONNTRACK_LABELS
1981 + sizeof(struct nf_conn_labels
)
1983 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
1984 + sizeof(struct nf_conn_synproxy
)
1989 int nf_conntrack_init_start(void)
1995 /* struct nf_ct_ext uses u8 to store offsets/size */
1996 BUILD_BUG_ON(total_extension_size() > 255u);
1998 seqcount_init(&nf_conntrack_generation
);
2000 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
2001 spin_lock_init(&nf_conntrack_locks
[i
]);
2003 if (!nf_conntrack_htable_size
) {
2004 /* Idea from tcp.c: use 1/16384 of memory.
2005 * On i386: 32MB machine has 512 buckets.
2006 * >= 1GB machines have 16384 buckets.
2007 * >= 4GB machines have 65536 buckets.
2009 nf_conntrack_htable_size
2010 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
2011 / sizeof(struct hlist_head
));
2012 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
2013 nf_conntrack_htable_size
= 65536;
2014 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
2015 nf_conntrack_htable_size
= 16384;
2016 if (nf_conntrack_htable_size
< 32)
2017 nf_conntrack_htable_size
= 32;
2019 /* Use a max. factor of four by default to get the same max as
2020 * with the old struct list_heads. When a table size is given
2021 * we use the old value of 8 to avoid reducing the max.
2026 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
2027 if (!nf_conntrack_hash
)
2030 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
2032 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
2033 sizeof(struct nf_conn
),
2035 SLAB_TYPESAFE_BY_RCU
| SLAB_HWCACHE_ALIGN
, NULL
);
2036 if (!nf_conntrack_cachep
)
2039 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
2040 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
2043 ret
= nf_conntrack_expect_init();
2047 ret
= nf_conntrack_acct_init();
2051 ret
= nf_conntrack_tstamp_init();
2055 ret
= nf_conntrack_ecache_init();
2059 ret
= nf_conntrack_timeout_init();
2063 ret
= nf_conntrack_helper_init();
2067 ret
= nf_conntrack_labels_init();
2071 ret
= nf_conntrack_seqadj_init();
2075 ret
= nf_conntrack_proto_init();
2079 conntrack_gc_work_init(&conntrack_gc_work
);
2080 queue_delayed_work(system_long_wq
, &conntrack_gc_work
.dwork
, HZ
);
2085 nf_conntrack_seqadj_fini();
2087 nf_conntrack_labels_fini();
2089 nf_conntrack_helper_fini();
2091 nf_conntrack_timeout_fini();
2093 nf_conntrack_ecache_fini();
2095 nf_conntrack_tstamp_fini();
2097 nf_conntrack_acct_fini();
2099 nf_conntrack_expect_fini();
2101 kmem_cache_destroy(nf_conntrack_cachep
);
2103 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
2107 void nf_conntrack_init_end(void)
2109 /* For use by REJECT target */
2110 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
2111 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
2115 * We need to use special "null" values, not used in hash table
2117 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2118 #define DYING_NULLS_VAL ((1<<30)+1)
2119 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
2121 int nf_conntrack_init_net(struct net
*net
)
2126 BUILD_BUG_ON(IP_CT_UNTRACKED
== IP_CT_NUMBER
);
2127 atomic_set(&net
->ct
.count
, 0);
2129 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
2130 if (!net
->ct
.pcpu_lists
)
2133 for_each_possible_cpu(cpu
) {
2134 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2136 spin_lock_init(&pcpu
->lock
);
2137 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
2138 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
2141 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
2143 goto err_pcpu_lists
;
2145 ret
= nf_conntrack_expect_pernet_init(net
);
2148 ret
= nf_conntrack_acct_pernet_init(net
);
2151 ret
= nf_conntrack_tstamp_pernet_init(net
);
2154 ret
= nf_conntrack_ecache_pernet_init(net
);
2157 ret
= nf_conntrack_helper_pernet_init(net
);
2160 ret
= nf_conntrack_proto_pernet_init(net
);
2166 nf_conntrack_helper_pernet_fini(net
);
2168 nf_conntrack_ecache_pernet_fini(net
);
2170 nf_conntrack_tstamp_pernet_fini(net
);
2172 nf_conntrack_acct_pernet_fini(net
);
2174 nf_conntrack_expect_pernet_fini(net
);
2176 free_percpu(net
->ct
.stat
);
2178 free_percpu(net
->ct
.pcpu_lists
);