1 /* Connection state tracking for netfilter. This is separated from,
2 but required by, the NAT layer; it can also be used by an iptables
5 /* (C) 1999-2001 Paul `Rusty' Russell
6 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
7 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
8 * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
15 #include <linux/types.h>
16 #include <linux/netfilter.h>
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/skbuff.h>
20 #include <linux/proc_fs.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stddef.h>
23 #include <linux/slab.h>
24 #include <linux/random.h>
25 #include <linux/jhash.h>
26 #include <linux/err.h>
27 #include <linux/percpu.h>
28 #include <linux/moduleparam.h>
29 #include <linux/notifier.h>
30 #include <linux/kernel.h>
31 #include <linux/netdevice.h>
32 #include <linux/socket.h>
34 #include <linux/nsproxy.h>
35 #include <linux/rculist_nulls.h>
37 #include <net/netfilter/nf_conntrack.h>
38 #include <net/netfilter/nf_conntrack_l3proto.h>
39 #include <net/netfilter/nf_conntrack_l4proto.h>
40 #include <net/netfilter/nf_conntrack_expect.h>
41 #include <net/netfilter/nf_conntrack_helper.h>
42 #include <net/netfilter/nf_conntrack_seqadj.h>
43 #include <net/netfilter/nf_conntrack_core.h>
44 #include <net/netfilter/nf_conntrack_extend.h>
45 #include <net/netfilter/nf_conntrack_acct.h>
46 #include <net/netfilter/nf_conntrack_ecache.h>
47 #include <net/netfilter/nf_conntrack_zones.h>
48 #include <net/netfilter/nf_conntrack_timestamp.h>
49 #include <net/netfilter/nf_conntrack_timeout.h>
50 #include <net/netfilter/nf_conntrack_labels.h>
51 #include <net/netfilter/nf_conntrack_synproxy.h>
52 #include <net/netfilter/nf_nat.h>
53 #include <net/netfilter/nf_nat_core.h>
54 #include <net/netfilter/nf_nat_helper.h>
56 #define NF_CONNTRACK_VERSION "0.5.0"
58 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
59 enum nf_nat_manip_type manip
,
60 const struct nlattr
*attr
) __read_mostly
;
61 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
63 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks
[CONNTRACK_LOCKS
];
64 EXPORT_SYMBOL_GPL(nf_conntrack_locks
);
66 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(nf_conntrack_expect_lock
);
67 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock
);
69 static __read_mostly spinlock_t nf_conntrack_locks_all_lock
;
70 static __read_mostly
bool nf_conntrack_locks_all
;
72 void nf_conntrack_lock(spinlock_t
*lock
) __acquires(lock
)
75 while (unlikely(nf_conntrack_locks_all
)) {
77 spin_lock(&nf_conntrack_locks_all_lock
);
78 spin_unlock(&nf_conntrack_locks_all_lock
);
82 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
84 static void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
86 h1
%= CONNTRACK_LOCKS
;
87 h2
%= CONNTRACK_LOCKS
;
88 spin_unlock(&nf_conntrack_locks
[h1
]);
90 spin_unlock(&nf_conntrack_locks
[h2
]);
93 /* return true if we need to recompute hashes (in case hash table was resized) */
94 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
95 unsigned int h2
, unsigned int sequence
)
97 h1
%= CONNTRACK_LOCKS
;
98 h2
%= CONNTRACK_LOCKS
;
100 nf_conntrack_lock(&nf_conntrack_locks
[h1
]);
102 spin_lock_nested(&nf_conntrack_locks
[h2
],
103 SINGLE_DEPTH_NESTING
);
105 nf_conntrack_lock(&nf_conntrack_locks
[h2
]);
106 spin_lock_nested(&nf_conntrack_locks
[h1
],
107 SINGLE_DEPTH_NESTING
);
109 if (read_seqcount_retry(&net
->ct
.generation
, sequence
)) {
110 nf_conntrack_double_unlock(h1
, h2
);
116 static void nf_conntrack_all_lock(void)
120 spin_lock(&nf_conntrack_locks_all_lock
);
121 nf_conntrack_locks_all
= true;
123 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++) {
124 spin_lock(&nf_conntrack_locks
[i
]);
125 spin_unlock(&nf_conntrack_locks
[i
]);
129 static void nf_conntrack_all_unlock(void)
131 nf_conntrack_locks_all
= false;
132 spin_unlock(&nf_conntrack_locks_all_lock
);
135 unsigned int nf_conntrack_htable_size __read_mostly
;
136 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
138 unsigned int nf_conntrack_max __read_mostly
;
139 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
141 DEFINE_PER_CPU(struct nf_conn
, nf_conntrack_untracked
);
142 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked
);
144 unsigned int nf_conntrack_hash_rnd __read_mostly
;
145 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd
);
147 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
)
151 /* The direction must be ignored, so we hash everything up to the
152 * destination ports (which is a multiple of 4) and treat the last
153 * three bytes manually.
155 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
156 return jhash2((u32
*)tuple
, n
, nf_conntrack_hash_rnd
^
157 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
158 tuple
->dst
.protonum
));
161 static u32
__hash_bucket(u32 hash
, unsigned int size
)
163 return reciprocal_scale(hash
, size
);
166 static u32
hash_bucket(u32 hash
, const struct net
*net
)
168 return __hash_bucket(hash
, net
->ct
.htable_size
);
171 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
174 return __hash_bucket(hash_conntrack_raw(tuple
), size
);
177 static inline u_int32_t
hash_conntrack(const struct net
*net
,
178 const struct nf_conntrack_tuple
*tuple
)
180 return __hash_conntrack(tuple
, net
->ct
.htable_size
);
184 nf_ct_get_tuple(const struct sk_buff
*skb
,
186 unsigned int dataoff
,
190 struct nf_conntrack_tuple
*tuple
,
191 const struct nf_conntrack_l3proto
*l3proto
,
192 const struct nf_conntrack_l4proto
*l4proto
)
194 memset(tuple
, 0, sizeof(*tuple
));
196 tuple
->src
.l3num
= l3num
;
197 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
200 tuple
->dst
.protonum
= protonum
;
201 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
203 return l4proto
->pkt_to_tuple(skb
, dataoff
, net
, tuple
);
205 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
207 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
209 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
211 struct nf_conntrack_l3proto
*l3proto
;
212 struct nf_conntrack_l4proto
*l4proto
;
213 unsigned int protoff
;
219 l3proto
= __nf_ct_l3proto_find(l3num
);
220 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
221 if (ret
!= NF_ACCEPT
) {
226 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
228 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
,
234 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
237 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
238 const struct nf_conntrack_tuple
*orig
,
239 const struct nf_conntrack_l3proto
*l3proto
,
240 const struct nf_conntrack_l4proto
*l4proto
)
242 memset(inverse
, 0, sizeof(*inverse
));
244 inverse
->src
.l3num
= orig
->src
.l3num
;
245 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
248 inverse
->dst
.dir
= !orig
->dst
.dir
;
250 inverse
->dst
.protonum
= orig
->dst
.protonum
;
251 return l4proto
->invert_tuple(inverse
, orig
);
253 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
256 clean_from_lists(struct nf_conn
*ct
)
258 pr_debug("clean_from_lists(%p)\n", ct
);
259 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
260 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
262 /* Destroy all pending expectations */
263 nf_ct_remove_expectations(ct
);
266 /* must be called with local_bh_disable */
267 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
269 struct ct_pcpu
*pcpu
;
271 /* add this conntrack to the (per cpu) dying list */
272 ct
->cpu
= smp_processor_id();
273 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
275 spin_lock(&pcpu
->lock
);
276 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
278 spin_unlock(&pcpu
->lock
);
281 /* must be called with local_bh_disable */
282 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
284 struct ct_pcpu
*pcpu
;
286 /* add this conntrack to the (per cpu) unconfirmed list */
287 ct
->cpu
= smp_processor_id();
288 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
290 spin_lock(&pcpu
->lock
);
291 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
293 spin_unlock(&pcpu
->lock
);
296 /* must be called with local_bh_disable */
297 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
299 struct ct_pcpu
*pcpu
;
301 /* We overload first tuple to link into unconfirmed or dying list.*/
302 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
304 spin_lock(&pcpu
->lock
);
305 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
306 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
307 spin_unlock(&pcpu
->lock
);
310 /* Released via destroy_conntrack() */
311 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
312 const struct nf_conntrack_zone
*zone
,
315 struct nf_conn
*tmpl
;
317 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
321 tmpl
->status
= IPS_TEMPLATE
;
322 write_pnet(&tmpl
->ct_net
, net
);
324 if (nf_ct_zone_add(tmpl
, flags
, zone
) < 0)
327 atomic_set(&tmpl
->ct_general
.use
, 0);
334 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
336 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
338 nf_ct_ext_destroy(tmpl
);
339 nf_ct_ext_free(tmpl
);
342 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
345 destroy_conntrack(struct nf_conntrack
*nfct
)
347 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
348 struct net
*net
= nf_ct_net(ct
);
349 struct nf_conntrack_l4proto
*l4proto
;
351 pr_debug("destroy_conntrack(%p)\n", ct
);
352 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
353 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
355 if (unlikely(nf_ct_is_template(ct
))) {
360 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
361 if (l4proto
&& l4proto
->destroy
)
362 l4proto
->destroy(ct
);
367 /* Expectations will have been removed in clean_from_lists,
368 * except TFTP can create an expectation on the first packet,
369 * before connection is in the list, so we need to clean here,
372 nf_ct_remove_expectations(ct
);
374 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
376 NF_CT_STAT_INC(net
, delete);
380 nf_ct_put(ct
->master
);
382 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
383 nf_conntrack_free(ct
);
386 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
388 struct net
*net
= nf_ct_net(ct
);
389 unsigned int hash
, reply_hash
;
390 unsigned int sequence
;
392 nf_ct_helper_destroy(ct
);
396 sequence
= read_seqcount_begin(&net
->ct
.generation
);
397 hash
= hash_conntrack(net
,
398 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
399 reply_hash
= hash_conntrack(net
,
400 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
401 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
403 clean_from_lists(ct
);
404 nf_conntrack_double_unlock(hash
, reply_hash
);
406 nf_ct_add_to_dying_list(ct
);
408 NF_CT_STAT_INC(net
, delete_list
);
412 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
414 struct nf_conn_tstamp
*tstamp
;
416 tstamp
= nf_conn_tstamp_find(ct
);
417 if (tstamp
&& tstamp
->stop
== 0)
418 tstamp
->stop
= ktime_get_real_ns();
420 if (nf_ct_is_dying(ct
))
423 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
424 portid
, report
) < 0) {
425 /* destroy event was not delivered */
426 nf_ct_delete_from_lists(ct
);
427 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
431 nf_conntrack_ecache_work(nf_ct_net(ct
));
432 set_bit(IPS_DYING_BIT
, &ct
->status
);
434 nf_ct_delete_from_lists(ct
);
438 EXPORT_SYMBOL_GPL(nf_ct_delete
);
440 static void death_by_timeout(unsigned long ul_conntrack
)
442 nf_ct_delete((struct nf_conn
*)ul_conntrack
, 0, 0);
446 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
447 const struct nf_conntrack_tuple
*tuple
,
448 const struct nf_conntrack_zone
*zone
)
450 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
452 /* A conntrack can be recreated with the equal tuple,
453 * so we need to check that the conntrack is confirmed
455 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
456 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
457 nf_ct_is_confirmed(ct
);
462 * - Caller must take a reference on returned object
463 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
465 static struct nf_conntrack_tuple_hash
*
466 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
467 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
469 struct nf_conntrack_tuple_hash
*h
;
470 struct hlist_nulls_node
*n
;
471 unsigned int bucket
= hash_bucket(hash
, net
);
473 /* Disable BHs the entire time since we normally need to disable them
474 * at least once for the stats anyway.
478 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[bucket
], hnnode
) {
479 if (nf_ct_key_equal(h
, tuple
, zone
)) {
480 NF_CT_STAT_INC(net
, found
);
484 NF_CT_STAT_INC(net
, searched
);
487 * if the nulls value we got at the end of this lookup is
488 * not the expected one, we must restart lookup.
489 * We probably met an item that was moved to another chain.
491 if (get_nulls_value(n
) != bucket
) {
492 NF_CT_STAT_INC(net
, search_restart
);
500 /* Find a connection corresponding to a tuple. */
501 static struct nf_conntrack_tuple_hash
*
502 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
503 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
505 struct nf_conntrack_tuple_hash
*h
;
510 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
512 ct
= nf_ct_tuplehash_to_ctrack(h
);
513 if (unlikely(nf_ct_is_dying(ct
) ||
514 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
517 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
))) {
528 struct nf_conntrack_tuple_hash
*
529 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
530 const struct nf_conntrack_tuple
*tuple
)
532 return __nf_conntrack_find_get(net
, zone
, tuple
,
533 hash_conntrack_raw(tuple
));
535 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
537 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
539 unsigned int reply_hash
)
541 struct net
*net
= nf_ct_net(ct
);
543 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
544 &net
->ct
.hash
[hash
]);
545 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
546 &net
->ct
.hash
[reply_hash
]);
550 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
552 const struct nf_conntrack_zone
*zone
;
553 struct net
*net
= nf_ct_net(ct
);
554 unsigned int hash
, reply_hash
;
555 struct nf_conntrack_tuple_hash
*h
;
556 struct hlist_nulls_node
*n
;
557 unsigned int sequence
;
559 zone
= nf_ct_zone(ct
);
563 sequence
= read_seqcount_begin(&net
->ct
.generation
);
564 hash
= hash_conntrack(net
,
565 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
566 reply_hash
= hash_conntrack(net
,
567 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
568 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
570 /* See if there's one in the list already, including reverse */
571 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
572 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
574 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
577 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
578 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
580 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
584 add_timer(&ct
->timeout
);
586 /* The caller holds a reference to this object */
587 atomic_set(&ct
->ct_general
.use
, 2);
588 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
589 nf_conntrack_double_unlock(hash
, reply_hash
);
590 NF_CT_STAT_INC(net
, insert
);
595 nf_conntrack_double_unlock(hash
, reply_hash
);
596 NF_CT_STAT_INC(net
, insert_failed
);
600 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
602 /* Confirm a connection given skb; places it in hash table */
604 __nf_conntrack_confirm(struct sk_buff
*skb
)
606 const struct nf_conntrack_zone
*zone
;
607 unsigned int hash
, reply_hash
;
608 struct nf_conntrack_tuple_hash
*h
;
610 struct nf_conn_help
*help
;
611 struct nf_conn_tstamp
*tstamp
;
612 struct hlist_nulls_node
*n
;
613 enum ip_conntrack_info ctinfo
;
615 unsigned int sequence
;
617 ct
= nf_ct_get(skb
, &ctinfo
);
620 /* ipt_REJECT uses nf_conntrack_attach to attach related
621 ICMP/TCP RST packets in other direction. Actual packet
622 which created connection will be IP_CT_NEW or for an
623 expected connection, IP_CT_RELATED. */
624 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
627 zone
= nf_ct_zone(ct
);
631 sequence
= read_seqcount_begin(&net
->ct
.generation
);
632 /* reuse the hash saved before */
633 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
634 hash
= hash_bucket(hash
, net
);
635 reply_hash
= hash_conntrack(net
,
636 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
638 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
640 /* We're not in hash table, and we refuse to set up related
641 * connections for unconfirmed conns. But packet copies and
642 * REJECT will give spurious warnings here.
644 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
646 /* No external references means no one else could have
649 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
650 pr_debug("Confirming conntrack %p\n", ct
);
651 /* We have to check the DYING flag after unlink to prevent
652 * a race against nf_ct_get_next_corpse() possibly called from
653 * user context, else we insert an already 'dead' hash, blocking
654 * further use of that particular connection -JM.
656 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
658 if (unlikely(nf_ct_is_dying(ct
)))
661 /* See if there's one in the list already, including reverse:
662 NAT could have grabbed it without realizing, since we're
663 not in the hash. If there is, we lost race. */
664 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
665 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
667 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
670 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
671 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
673 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
677 /* Timer relative to confirmation time, not original
678 setting time, otherwise we'd get timer wrap in
679 weird delay cases. */
680 ct
->timeout
.expires
+= jiffies
;
681 add_timer(&ct
->timeout
);
682 atomic_inc(&ct
->ct_general
.use
);
683 ct
->status
|= IPS_CONFIRMED
;
685 /* set conntrack timestamp, if enabled. */
686 tstamp
= nf_conn_tstamp_find(ct
);
688 if (skb
->tstamp
.tv64
== 0)
689 __net_timestamp(skb
);
691 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
693 /* Since the lookup is lockless, hash insertion must be done after
694 * starting the timer and setting the CONFIRMED bit. The RCU barriers
695 * guarantee that no other CPU can find the conntrack before the above
696 * stores are visible.
698 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
699 nf_conntrack_double_unlock(hash
, reply_hash
);
700 NF_CT_STAT_INC(net
, insert
);
703 help
= nfct_help(ct
);
704 if (help
&& help
->helper
)
705 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
707 nf_conntrack_event_cache(master_ct(ct
) ?
708 IPCT_RELATED
: IPCT_NEW
, ct
);
712 nf_ct_add_to_dying_list(ct
);
713 nf_conntrack_double_unlock(hash
, reply_hash
);
714 NF_CT_STAT_INC(net
, insert_failed
);
718 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
720 /* Returns true if a connection correspondings to the tuple (required
723 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
724 const struct nf_conn
*ignored_conntrack
)
726 struct net
*net
= nf_ct_net(ignored_conntrack
);
727 const struct nf_conntrack_zone
*zone
;
728 struct nf_conntrack_tuple_hash
*h
;
729 struct hlist_nulls_node
*n
;
733 zone
= nf_ct_zone(ignored_conntrack
);
734 hash
= hash_conntrack(net
, tuple
);
736 /* Disable BHs the entire time since we need to disable them at
737 * least once for the stats anyway.
740 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
741 ct
= nf_ct_tuplehash_to_ctrack(h
);
742 if (ct
!= ignored_conntrack
&&
743 nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
744 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
))) {
745 NF_CT_STAT_INC(net
, found
);
746 rcu_read_unlock_bh();
749 NF_CT_STAT_INC(net
, searched
);
751 rcu_read_unlock_bh();
755 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
757 #define NF_CT_EVICTION_RANGE 8
759 /* There's a small race here where we may free a just-assured
760 connection. Too bad: we're in trouble anyway. */
761 static noinline
int early_drop(struct net
*net
, unsigned int _hash
)
763 /* Use oldest entry, which is roughly LRU */
764 struct nf_conntrack_tuple_hash
*h
;
765 struct nf_conn
*ct
= NULL
, *tmp
;
766 struct hlist_nulls_node
*n
;
767 unsigned int i
= 0, cnt
= 0;
769 unsigned int hash
, sequence
;
774 sequence
= read_seqcount_begin(&net
->ct
.generation
);
775 hash
= hash_bucket(_hash
, net
);
776 for (; i
< net
->ct
.htable_size
; i
++) {
777 lockp
= &nf_conntrack_locks
[hash
% CONNTRACK_LOCKS
];
778 nf_conntrack_lock(lockp
);
779 if (read_seqcount_retry(&net
->ct
.generation
, sequence
)) {
783 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
785 tmp
= nf_ct_tuplehash_to_ctrack(h
);
786 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
) &&
787 !nf_ct_is_dying(tmp
) &&
788 atomic_inc_not_zero(&tmp
->ct_general
.use
)) {
795 hash
= (hash
+ 1) % net
->ct
.htable_size
;
798 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
807 if (del_timer(&ct
->timeout
)) {
808 if (nf_ct_delete(ct
, 0, 0)) {
810 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
817 void init_nf_conntrack_hash_rnd(void)
822 * Why not initialize nf_conntrack_rnd in a "init()" function ?
823 * Because there isn't enough entropy when system initializing,
824 * and we initialize it as late as possible.
827 get_random_bytes(&rand
, sizeof(rand
));
829 cmpxchg(&nf_conntrack_hash_rnd
, 0, rand
);
832 static struct nf_conn
*
833 __nf_conntrack_alloc(struct net
*net
,
834 const struct nf_conntrack_zone
*zone
,
835 const struct nf_conntrack_tuple
*orig
,
836 const struct nf_conntrack_tuple
*repl
,
841 if (unlikely(!nf_conntrack_hash_rnd
)) {
842 init_nf_conntrack_hash_rnd();
843 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
844 hash
= hash_conntrack_raw(orig
);
847 /* We don't want any race condition at early drop stage */
848 atomic_inc(&net
->ct
.count
);
850 if (nf_conntrack_max
&&
851 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
852 if (!early_drop(net
, hash
)) {
853 atomic_dec(&net
->ct
.count
);
854 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
855 return ERR_PTR(-ENOMEM
);
860 * Do not use kmem_cache_zalloc(), as this cache uses
861 * SLAB_DESTROY_BY_RCU.
863 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
867 spin_lock_init(&ct
->lock
);
868 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
869 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
870 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
871 /* save hash for reusing when confirming */
872 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
874 /* Don't set timer yet: wait for confirmation */
875 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
876 write_pnet(&ct
->ct_net
, net
);
877 memset(&ct
->__nfct_init_offset
[0], 0,
878 offsetof(struct nf_conn
, proto
) -
879 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
881 if (zone
&& nf_ct_zone_add(ct
, GFP_ATOMIC
, zone
) < 0)
884 /* Because we use RCU lookups, we set ct_general.use to zero before
885 * this is inserted in any list.
887 atomic_set(&ct
->ct_general
.use
, 0);
890 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
892 atomic_dec(&net
->ct
.count
);
893 return ERR_PTR(-ENOMEM
);
896 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
897 const struct nf_conntrack_zone
*zone
,
898 const struct nf_conntrack_tuple
*orig
,
899 const struct nf_conntrack_tuple
*repl
,
902 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
904 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
906 void nf_conntrack_free(struct nf_conn
*ct
)
908 struct net
*net
= nf_ct_net(ct
);
910 /* A freed object has refcnt == 0, that's
911 * the golden rule for SLAB_DESTROY_BY_RCU
913 NF_CT_ASSERT(atomic_read(&ct
->ct_general
.use
) == 0);
915 nf_ct_ext_destroy(ct
);
917 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
918 smp_mb__before_atomic();
919 atomic_dec(&net
->ct
.count
);
921 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
924 /* Allocate a new conntrack: we return -ENOMEM if classification
925 failed due to stress. Otherwise it really is unclassifiable. */
926 static struct nf_conntrack_tuple_hash
*
927 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
928 const struct nf_conntrack_tuple
*tuple
,
929 struct nf_conntrack_l3proto
*l3proto
,
930 struct nf_conntrack_l4proto
*l4proto
,
932 unsigned int dataoff
, u32 hash
)
935 struct nf_conn_help
*help
;
936 struct nf_conntrack_tuple repl_tuple
;
937 struct nf_conntrack_ecache
*ecache
;
938 struct nf_conntrack_expect
*exp
= NULL
;
939 const struct nf_conntrack_zone
*zone
;
940 struct nf_conn_timeout
*timeout_ext
;
941 struct nf_conntrack_zone tmp
;
942 unsigned int *timeouts
;
944 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
945 pr_debug("Can't invert tuple.\n");
949 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
950 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
953 return (struct nf_conntrack_tuple_hash
*)ct
;
955 if (tmpl
&& nfct_synproxy(tmpl
)) {
956 nfct_seqadj_ext_add(ct
);
957 nfct_synproxy_ext_add(ct
);
960 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
962 timeouts
= nf_ct_timeout_data(timeout_ext
);
963 if (unlikely(!timeouts
))
964 timeouts
= l4proto
->get_timeouts(net
);
966 timeouts
= l4proto
->get_timeouts(net
);
969 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
970 nf_conntrack_free(ct
);
971 pr_debug("init conntrack: can't track with proto module\n");
976 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
979 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
980 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
981 nf_ct_labels_ext_add(ct
);
983 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
984 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
985 ecache
? ecache
->expmask
: 0,
989 if (net
->ct
.expect_count
) {
990 spin_lock(&nf_conntrack_expect_lock
);
991 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
993 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
995 /* Welcome, Mr. Bond. We've been expecting you... */
996 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
997 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
998 ct
->master
= exp
->master
;
1000 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
1003 rcu_assign_pointer(help
->helper
, exp
->helper
);
1006 #ifdef CONFIG_NF_CONNTRACK_MARK
1007 ct
->mark
= exp
->master
->mark
;
1009 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1010 ct
->secmark
= exp
->master
->secmark
;
1012 NF_CT_STAT_INC(net
, expect_new
);
1014 spin_unlock(&nf_conntrack_expect_lock
);
1017 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1018 NF_CT_STAT_INC(net
, new);
1021 /* Now it is inserted into the unconfirmed list, bump refcount */
1022 nf_conntrack_get(&ct
->ct_general
);
1023 nf_ct_add_to_unconfirmed_list(ct
);
1029 exp
->expectfn(ct
, exp
);
1030 nf_ct_expect_put(exp
);
1033 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1036 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1037 static inline struct nf_conn
*
1038 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1039 struct sk_buff
*skb
,
1040 unsigned int dataoff
,
1043 struct nf_conntrack_l3proto
*l3proto
,
1044 struct nf_conntrack_l4proto
*l4proto
,
1046 enum ip_conntrack_info
*ctinfo
)
1048 const struct nf_conntrack_zone
*zone
;
1049 struct nf_conntrack_tuple tuple
;
1050 struct nf_conntrack_tuple_hash
*h
;
1051 struct nf_conntrack_zone tmp
;
1055 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1056 dataoff
, l3num
, protonum
, net
, &tuple
, l3proto
,
1058 pr_debug("resolve_normal_ct: Can't get tuple\n");
1062 /* look for tuple match */
1063 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1064 hash
= hash_conntrack_raw(&tuple
);
1065 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1067 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1068 skb
, dataoff
, hash
);
1074 ct
= nf_ct_tuplehash_to_ctrack(h
);
1076 /* It exists; we have (non-exclusive) reference. */
1077 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1078 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1079 /* Please set reply bit if this packet OK */
1082 /* Once we've had two way comms, always ESTABLISHED. */
1083 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1084 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
1085 *ctinfo
= IP_CT_ESTABLISHED
;
1086 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1087 pr_debug("nf_conntrack_in: related packet for %p\n",
1089 *ctinfo
= IP_CT_RELATED
;
1091 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
1092 *ctinfo
= IP_CT_NEW
;
1096 skb
->nfct
= &ct
->ct_general
;
1097 skb
->nfctinfo
= *ctinfo
;
1102 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1103 struct sk_buff
*skb
)
1105 struct nf_conn
*ct
, *tmpl
= NULL
;
1106 enum ip_conntrack_info ctinfo
;
1107 struct nf_conntrack_l3proto
*l3proto
;
1108 struct nf_conntrack_l4proto
*l4proto
;
1109 unsigned int *timeouts
;
1110 unsigned int dataoff
;
1116 /* Previously seen (loopback or untracked)? Ignore. */
1117 tmpl
= (struct nf_conn
*)skb
->nfct
;
1118 if (!nf_ct_is_template(tmpl
)) {
1119 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1125 /* rcu_read_lock()ed by nf_hook_slow */
1126 l3proto
= __nf_ct_l3proto_find(pf
);
1127 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1128 &dataoff
, &protonum
);
1130 pr_debug("not prepared to track yet or error occurred\n");
1131 NF_CT_STAT_INC_ATOMIC(net
, error
);
1132 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1137 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1139 /* It may be an special packet, error, unclean...
1140 * inverse of the return code tells to the netfilter
1141 * core what to do with the packet. */
1142 if (l4proto
->error
!= NULL
) {
1143 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
1146 NF_CT_STAT_INC_ATOMIC(net
, error
);
1147 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1151 /* ICMP[v6] protocol trackers may assign one conntrack. */
1156 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1157 l3proto
, l4proto
, &set_reply
, &ctinfo
);
1159 /* Not valid part of a connection */
1160 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1166 /* Too stressed to deal. */
1167 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1172 NF_CT_ASSERT(skb
->nfct
);
1174 /* Decide what timeout policy we want to apply to this flow. */
1175 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1177 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1179 /* Invalid: inverse of the return code tells
1180 * the netfilter core what to do */
1181 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1182 nf_conntrack_put(skb
->nfct
);
1184 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1185 if (ret
== -NF_DROP
)
1186 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1191 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1192 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1195 /* Special case: we have to repeat this hook, assign the
1196 * template again to this packet. We assume that this packet
1197 * has no conntrack assigned. This is used by nf_ct_tcp. */
1198 if (ret
== NF_REPEAT
)
1199 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
1206 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1208 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1209 const struct nf_conntrack_tuple
*orig
)
1214 ret
= nf_ct_invert_tuple(inverse
, orig
,
1215 __nf_ct_l3proto_find(orig
->src
.l3num
),
1216 __nf_ct_l4proto_find(orig
->src
.l3num
,
1217 orig
->dst
.protonum
));
1221 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1223 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1224 implicitly racy: see __nf_conntrack_confirm */
1225 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1226 const struct nf_conntrack_tuple
*newreply
)
1228 struct nf_conn_help
*help
= nfct_help(ct
);
1230 /* Should be unconfirmed, so not in hash table yet */
1231 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1233 pr_debug("Altering reply tuple of %p to ", ct
);
1234 nf_ct_dump_tuple(newreply
);
1236 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1237 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1241 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1244 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1246 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1247 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1248 enum ip_conntrack_info ctinfo
,
1249 const struct sk_buff
*skb
,
1250 unsigned long extra_jiffies
,
1253 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1256 /* Only update if this is not a fixed timeout */
1257 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1260 /* If not in hash table, timer will not be active yet */
1261 if (!nf_ct_is_confirmed(ct
)) {
1262 ct
->timeout
.expires
= extra_jiffies
;
1264 unsigned long newtime
= jiffies
+ extra_jiffies
;
1266 /* Only update the timeout if the new timeout is at least
1267 HZ jiffies from the old timeout. Need del_timer for race
1268 avoidance (may already be dying). */
1269 if (newtime
- ct
->timeout
.expires
>= HZ
)
1270 mod_timer_pending(&ct
->timeout
, newtime
);
1275 struct nf_conn_acct
*acct
;
1277 acct
= nf_conn_acct_find(ct
);
1279 struct nf_conn_counter
*counter
= acct
->counter
;
1281 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1282 atomic64_add(skb
->len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1286 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1288 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1289 enum ip_conntrack_info ctinfo
,
1290 const struct sk_buff
*skb
,
1294 struct nf_conn_acct
*acct
;
1296 acct
= nf_conn_acct_find(ct
);
1298 struct nf_conn_counter
*counter
= acct
->counter
;
1300 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1301 atomic64_add(skb
->len
- skb_network_offset(skb
),
1302 &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1306 if (del_timer(&ct
->timeout
)) {
1307 ct
->timeout
.function((unsigned long)ct
);
1312 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1314 #ifdef CONFIG_NF_CONNTRACK_ZONES
1315 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1316 .len
= sizeof(struct nf_conntrack_zone
),
1317 .align
= __alignof__(struct nf_conntrack_zone
),
1318 .id
= NF_CT_EXT_ZONE
,
1322 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1324 #include <linux/netfilter/nfnetlink.h>
1325 #include <linux/netfilter/nfnetlink_conntrack.h>
1326 #include <linux/mutex.h>
1328 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1329 * in ip_conntrack_core, since we don't want the protocols to autoload
1330 * or depend on ctnetlink */
1331 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1332 const struct nf_conntrack_tuple
*tuple
)
1334 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1335 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1336 goto nla_put_failure
;
1342 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1344 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1345 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1346 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1348 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1350 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1351 struct nf_conntrack_tuple
*t
)
1353 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1356 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1357 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1361 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1363 int nf_ct_port_nlattr_tuple_size(void)
1365 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1367 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1370 /* Used by ipt_REJECT and ip6t_REJECT. */
1371 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1374 enum ip_conntrack_info ctinfo
;
1376 /* This ICMP is in reverse direction to the packet which caused it */
1377 ct
= nf_ct_get(skb
, &ctinfo
);
1378 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1379 ctinfo
= IP_CT_RELATED_REPLY
;
1381 ctinfo
= IP_CT_RELATED
;
1383 /* Attach to new skbuff, and increment count */
1384 nskb
->nfct
= &ct
->ct_general
;
1385 nskb
->nfctinfo
= ctinfo
;
1386 nf_conntrack_get(nskb
->nfct
);
1389 /* Bring out ya dead! */
1390 static struct nf_conn
*
1391 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1392 void *data
, unsigned int *bucket
)
1394 struct nf_conntrack_tuple_hash
*h
;
1396 struct hlist_nulls_node
*n
;
1400 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1401 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1403 nf_conntrack_lock(lockp
);
1404 if (*bucket
< net
->ct
.htable_size
) {
1405 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1406 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1408 ct
= nf_ct_tuplehash_to_ctrack(h
);
1418 for_each_possible_cpu(cpu
) {
1419 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1421 spin_lock_bh(&pcpu
->lock
);
1422 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1423 ct
= nf_ct_tuplehash_to_ctrack(h
);
1425 set_bit(IPS_DYING_BIT
, &ct
->status
);
1427 spin_unlock_bh(&pcpu
->lock
);
1432 atomic_inc(&ct
->ct_general
.use
);
1438 void nf_ct_iterate_cleanup(struct net
*net
,
1439 int (*iter
)(struct nf_conn
*i
, void *data
),
1440 void *data
, u32 portid
, int report
)
1443 unsigned int bucket
= 0;
1447 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1448 /* Time to push up daises... */
1449 if (del_timer(&ct
->timeout
))
1450 nf_ct_delete(ct
, portid
, report
);
1452 /* ... else the timer will get him soon. */
1458 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1460 static int kill_all(struct nf_conn
*i
, void *data
)
1465 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1467 if (is_vmalloc_addr(hash
))
1470 free_pages((unsigned long)hash
,
1471 get_order(sizeof(struct hlist_head
) * size
));
1473 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1475 static int untrack_refs(void)
1479 for_each_possible_cpu(cpu
) {
1480 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1482 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1487 void nf_conntrack_cleanup_start(void)
1489 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1492 void nf_conntrack_cleanup_end(void)
1494 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1495 while (untrack_refs() > 0)
1498 #ifdef CONFIG_NF_CONNTRACK_ZONES
1499 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1501 nf_conntrack_proto_fini();
1502 nf_conntrack_seqadj_fini();
1503 nf_conntrack_labels_fini();
1504 nf_conntrack_helper_fini();
1505 nf_conntrack_timeout_fini();
1506 nf_conntrack_ecache_fini();
1507 nf_conntrack_tstamp_fini();
1508 nf_conntrack_acct_fini();
1509 nf_conntrack_expect_fini();
1513 * Mishearing the voices in his head, our hero wonders how he's
1514 * supposed to kill the mall.
1516 void nf_conntrack_cleanup_net(struct net
*net
)
1520 list_add(&net
->exit_list
, &single
);
1521 nf_conntrack_cleanup_net_list(&single
);
1524 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1530 * This makes sure all current packets have passed through
1531 * netfilter framework. Roll on, two-stage module
1537 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1538 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1539 if (atomic_read(&net
->ct
.count
) != 0)
1544 goto i_see_dead_people
;
1547 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1548 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1549 nf_conntrack_proto_pernet_fini(net
);
1550 nf_conntrack_helper_pernet_fini(net
);
1551 nf_conntrack_ecache_pernet_fini(net
);
1552 nf_conntrack_tstamp_pernet_fini(net
);
1553 nf_conntrack_acct_pernet_fini(net
);
1554 nf_conntrack_expect_pernet_fini(net
);
1555 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1556 kfree(net
->ct
.slabname
);
1557 free_percpu(net
->ct
.stat
);
1558 free_percpu(net
->ct
.pcpu_lists
);
1562 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1564 struct hlist_nulls_head
*hash
;
1565 unsigned int nr_slots
, i
;
1568 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1569 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1570 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1571 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1577 for (i
= 0; i
< nr_slots
; i
++)
1578 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1582 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1584 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1587 unsigned int hashsize
, old_size
;
1588 struct hlist_nulls_head
*hash
, *old_hash
;
1589 struct nf_conntrack_tuple_hash
*h
;
1592 if (current
->nsproxy
->net_ns
!= &init_net
)
1595 /* On boot, we can set this without any fancy locking. */
1596 if (!nf_conntrack_htable_size
)
1597 return param_set_uint(val
, kp
);
1599 rc
= kstrtouint(val
, 0, &hashsize
);
1605 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1610 nf_conntrack_all_lock();
1611 write_seqcount_begin(&init_net
.ct
.generation
);
1613 /* Lookups in the old hash might happen in parallel, which means we
1614 * might get false negatives during connection lookup. New connections
1615 * created because of a false negative won't make it into the hash
1616 * though since that required taking the locks.
1619 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1620 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1621 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1622 struct nf_conntrack_tuple_hash
, hnnode
);
1623 ct
= nf_ct_tuplehash_to_ctrack(h
);
1624 hlist_nulls_del_rcu(&h
->hnnode
);
1625 bucket
= __hash_conntrack(&h
->tuple
, hashsize
);
1626 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1629 old_size
= init_net
.ct
.htable_size
;
1630 old_hash
= init_net
.ct
.hash
;
1632 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1633 init_net
.ct
.hash
= hash
;
1635 write_seqcount_end(&init_net
.ct
.generation
);
1636 nf_conntrack_all_unlock();
1639 nf_ct_free_hashtable(old_hash
, old_size
);
1642 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1644 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1645 &nf_conntrack_htable_size
, 0600);
1647 void nf_ct_untracked_status_or(unsigned long bits
)
1651 for_each_possible_cpu(cpu
)
1652 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1654 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1656 int nf_conntrack_init_start(void)
1661 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
1662 spin_lock_init(&nf_conntrack_locks
[i
]);
1664 if (!nf_conntrack_htable_size
) {
1665 /* Idea from tcp.c: use 1/16384 of memory.
1666 * On i386: 32MB machine has 512 buckets.
1667 * >= 1GB machines have 16384 buckets.
1668 * >= 4GB machines have 65536 buckets.
1670 nf_conntrack_htable_size
1671 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1672 / sizeof(struct hlist_head
));
1673 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
1674 nf_conntrack_htable_size
= 65536;
1675 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1676 nf_conntrack_htable_size
= 16384;
1677 if (nf_conntrack_htable_size
< 32)
1678 nf_conntrack_htable_size
= 32;
1680 /* Use a max. factor of four by default to get the same max as
1681 * with the old struct list_heads. When a table size is given
1682 * we use the old value of 8 to avoid reducing the max.
1686 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1688 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1689 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1692 ret
= nf_conntrack_expect_init();
1696 ret
= nf_conntrack_acct_init();
1700 ret
= nf_conntrack_tstamp_init();
1704 ret
= nf_conntrack_ecache_init();
1708 ret
= nf_conntrack_timeout_init();
1712 ret
= nf_conntrack_helper_init();
1716 ret
= nf_conntrack_labels_init();
1720 ret
= nf_conntrack_seqadj_init();
1724 #ifdef CONFIG_NF_CONNTRACK_ZONES
1725 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1729 ret
= nf_conntrack_proto_init();
1733 /* Set up fake conntrack: to never be deleted, not in any hashes */
1734 for_each_possible_cpu(cpu
) {
1735 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1736 write_pnet(&ct
->ct_net
, &init_net
);
1737 atomic_set(&ct
->ct_general
.use
, 1);
1739 /* - and look it like as a confirmed connection */
1740 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1744 #ifdef CONFIG_NF_CONNTRACK_ZONES
1745 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1748 nf_conntrack_seqadj_fini();
1750 nf_conntrack_labels_fini();
1752 nf_conntrack_helper_fini();
1754 nf_conntrack_timeout_fini();
1756 nf_conntrack_ecache_fini();
1758 nf_conntrack_tstamp_fini();
1760 nf_conntrack_acct_fini();
1762 nf_conntrack_expect_fini();
1767 void nf_conntrack_init_end(void)
1769 /* For use by REJECT target */
1770 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1771 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1775 * We need to use special "null" values, not used in hash table
1777 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1778 #define DYING_NULLS_VAL ((1<<30)+1)
1779 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1781 int nf_conntrack_init_net(struct net
*net
)
1786 atomic_set(&net
->ct
.count
, 0);
1787 seqcount_init(&net
->ct
.generation
);
1789 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
1790 if (!net
->ct
.pcpu_lists
)
1793 for_each_possible_cpu(cpu
) {
1794 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1796 spin_lock_init(&pcpu
->lock
);
1797 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1798 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
1801 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1803 goto err_pcpu_lists
;
1805 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1806 if (!net
->ct
.slabname
)
1809 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1810 sizeof(struct nf_conn
), 0,
1811 SLAB_DESTROY_BY_RCU
, NULL
);
1812 if (!net
->ct
.nf_conntrack_cachep
) {
1813 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1817 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1818 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
, 1);
1819 if (!net
->ct
.hash
) {
1820 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1823 ret
= nf_conntrack_expect_pernet_init(net
);
1826 ret
= nf_conntrack_acct_pernet_init(net
);
1829 ret
= nf_conntrack_tstamp_pernet_init(net
);
1832 ret
= nf_conntrack_ecache_pernet_init(net
);
1835 ret
= nf_conntrack_helper_pernet_init(net
);
1838 ret
= nf_conntrack_proto_pernet_init(net
);
1844 nf_conntrack_helper_pernet_fini(net
);
1846 nf_conntrack_ecache_pernet_fini(net
);
1848 nf_conntrack_tstamp_pernet_fini(net
);
1850 nf_conntrack_acct_pernet_fini(net
);
1852 nf_conntrack_expect_pernet_fini(net
);
1854 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1856 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1858 kfree(net
->ct
.slabname
);
1860 free_percpu(net
->ct
.stat
);
1862 free_percpu(net
->ct
.pcpu_lists
);