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 void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
71 h1
%= CONNTRACK_LOCKS
;
72 h2
%= CONNTRACK_LOCKS
;
73 spin_unlock(&nf_conntrack_locks
[h1
]);
75 spin_unlock(&nf_conntrack_locks
[h2
]);
78 /* return true if we need to recompute hashes (in case hash table was resized) */
79 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
80 unsigned int h2
, unsigned int sequence
)
82 h1
%= CONNTRACK_LOCKS
;
83 h2
%= CONNTRACK_LOCKS
;
85 spin_lock(&nf_conntrack_locks
[h1
]);
87 spin_lock_nested(&nf_conntrack_locks
[h2
],
88 SINGLE_DEPTH_NESTING
);
90 spin_lock(&nf_conntrack_locks
[h2
]);
91 spin_lock_nested(&nf_conntrack_locks
[h1
],
92 SINGLE_DEPTH_NESTING
);
94 if (read_seqcount_retry(&net
->ct
.generation
, sequence
)) {
95 nf_conntrack_double_unlock(h1
, h2
);
101 static void nf_conntrack_all_lock(void)
105 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
106 spin_lock_nested(&nf_conntrack_locks
[i
], i
);
109 static void nf_conntrack_all_unlock(void)
113 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
114 spin_unlock(&nf_conntrack_locks
[i
]);
117 unsigned int nf_conntrack_htable_size __read_mostly
;
118 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
120 unsigned int nf_conntrack_max __read_mostly
;
121 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
123 DEFINE_PER_CPU(struct nf_conn
, nf_conntrack_untracked
);
124 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked
);
126 unsigned int nf_conntrack_hash_rnd __read_mostly
;
127 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd
);
129 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
, u16 zone
)
133 /* The direction must be ignored, so we hash everything up to the
134 * destination ports (which is a multiple of 4) and treat the last
135 * three bytes manually.
137 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
138 return jhash2((u32
*)tuple
, n
, zone
^ nf_conntrack_hash_rnd
^
139 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
140 tuple
->dst
.protonum
));
143 static u32
__hash_bucket(u32 hash
, unsigned int size
)
145 return reciprocal_scale(hash
, size
);
148 static u32
hash_bucket(u32 hash
, const struct net
*net
)
150 return __hash_bucket(hash
, net
->ct
.htable_size
);
153 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
154 u16 zone
, unsigned int size
)
156 return __hash_bucket(hash_conntrack_raw(tuple
, zone
), size
);
159 static inline u_int32_t
hash_conntrack(const struct net
*net
, u16 zone
,
160 const struct nf_conntrack_tuple
*tuple
)
162 return __hash_conntrack(tuple
, zone
, net
->ct
.htable_size
);
166 nf_ct_get_tuple(const struct sk_buff
*skb
,
168 unsigned int dataoff
,
171 struct nf_conntrack_tuple
*tuple
,
172 const struct nf_conntrack_l3proto
*l3proto
,
173 const struct nf_conntrack_l4proto
*l4proto
)
175 memset(tuple
, 0, sizeof(*tuple
));
177 tuple
->src
.l3num
= l3num
;
178 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
181 tuple
->dst
.protonum
= protonum
;
182 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
184 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
186 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
188 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
189 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
191 struct nf_conntrack_l3proto
*l3proto
;
192 struct nf_conntrack_l4proto
*l4proto
;
193 unsigned int protoff
;
199 l3proto
= __nf_ct_l3proto_find(l3num
);
200 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
201 if (ret
!= NF_ACCEPT
) {
206 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
208 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
214 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
217 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
218 const struct nf_conntrack_tuple
*orig
,
219 const struct nf_conntrack_l3proto
*l3proto
,
220 const struct nf_conntrack_l4proto
*l4proto
)
222 memset(inverse
, 0, sizeof(*inverse
));
224 inverse
->src
.l3num
= orig
->src
.l3num
;
225 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
228 inverse
->dst
.dir
= !orig
->dst
.dir
;
230 inverse
->dst
.protonum
= orig
->dst
.protonum
;
231 return l4proto
->invert_tuple(inverse
, orig
);
233 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
236 clean_from_lists(struct nf_conn
*ct
)
238 pr_debug("clean_from_lists(%p)\n", ct
);
239 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
240 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
242 /* Destroy all pending expectations */
243 nf_ct_remove_expectations(ct
);
246 /* must be called with local_bh_disable */
247 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
249 struct ct_pcpu
*pcpu
;
251 /* add this conntrack to the (per cpu) dying list */
252 ct
->cpu
= smp_processor_id();
253 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
255 spin_lock(&pcpu
->lock
);
256 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
258 spin_unlock(&pcpu
->lock
);
261 /* must be called with local_bh_disable */
262 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
264 struct ct_pcpu
*pcpu
;
266 /* add this conntrack to the (per cpu) unconfirmed list */
267 ct
->cpu
= smp_processor_id();
268 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
270 spin_lock(&pcpu
->lock
);
271 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
273 spin_unlock(&pcpu
->lock
);
276 /* must be called with local_bh_disable */
277 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
279 struct ct_pcpu
*pcpu
;
281 /* We overload first tuple to link into unconfirmed or dying list.*/
282 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
284 spin_lock(&pcpu
->lock
);
285 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
286 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
287 spin_unlock(&pcpu
->lock
);
291 destroy_conntrack(struct nf_conntrack
*nfct
)
293 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
294 struct net
*net
= nf_ct_net(ct
);
295 struct nf_conntrack_l4proto
*l4proto
;
297 pr_debug("destroy_conntrack(%p)\n", ct
);
298 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
299 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
302 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
303 if (l4proto
&& l4proto
->destroy
)
304 l4proto
->destroy(ct
);
309 /* Expectations will have been removed in clean_from_lists,
310 * except TFTP can create an expectation on the first packet,
311 * before connection is in the list, so we need to clean here,
314 nf_ct_remove_expectations(ct
);
316 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
318 NF_CT_STAT_INC(net
, delete);
322 nf_ct_put(ct
->master
);
324 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
325 nf_conntrack_free(ct
);
328 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
330 struct net
*net
= nf_ct_net(ct
);
331 unsigned int hash
, reply_hash
;
332 u16 zone
= nf_ct_zone(ct
);
333 unsigned int sequence
;
335 nf_ct_helper_destroy(ct
);
339 sequence
= read_seqcount_begin(&net
->ct
.generation
);
340 hash
= hash_conntrack(net
, zone
,
341 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
342 reply_hash
= hash_conntrack(net
, zone
,
343 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
344 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
346 clean_from_lists(ct
);
347 nf_conntrack_double_unlock(hash
, reply_hash
);
349 nf_ct_add_to_dying_list(ct
);
351 NF_CT_STAT_INC(net
, delete_list
);
355 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
357 struct nf_conn_tstamp
*tstamp
;
359 tstamp
= nf_conn_tstamp_find(ct
);
360 if (tstamp
&& tstamp
->stop
== 0)
361 tstamp
->stop
= ktime_get_real_ns();
363 if (nf_ct_is_dying(ct
))
366 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
367 portid
, report
) < 0) {
368 /* destroy event was not delivered */
369 nf_ct_delete_from_lists(ct
);
370 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
374 nf_conntrack_ecache_work(nf_ct_net(ct
));
375 set_bit(IPS_DYING_BIT
, &ct
->status
);
377 nf_ct_delete_from_lists(ct
);
381 EXPORT_SYMBOL_GPL(nf_ct_delete
);
383 static void death_by_timeout(unsigned long ul_conntrack
)
385 nf_ct_delete((struct nf_conn
*)ul_conntrack
, 0, 0);
389 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
390 const struct nf_conntrack_tuple
*tuple
,
393 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
395 /* A conntrack can be recreated with the equal tuple,
396 * so we need to check that the conntrack is confirmed
398 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
399 nf_ct_zone(ct
) == zone
&&
400 nf_ct_is_confirmed(ct
);
405 * - Caller must take a reference on returned object
406 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
408 static struct nf_conntrack_tuple_hash
*
409 ____nf_conntrack_find(struct net
*net
, u16 zone
,
410 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
412 struct nf_conntrack_tuple_hash
*h
;
413 struct hlist_nulls_node
*n
;
414 unsigned int bucket
= hash_bucket(hash
, net
);
416 /* Disable BHs the entire time since we normally need to disable them
417 * at least once for the stats anyway.
421 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[bucket
], hnnode
) {
422 if (nf_ct_key_equal(h
, tuple
, zone
)) {
423 NF_CT_STAT_INC(net
, found
);
427 NF_CT_STAT_INC(net
, searched
);
430 * if the nulls value we got at the end of this lookup is
431 * not the expected one, we must restart lookup.
432 * We probably met an item that was moved to another chain.
434 if (get_nulls_value(n
) != bucket
) {
435 NF_CT_STAT_INC(net
, search_restart
);
443 /* Find a connection corresponding to a tuple. */
444 static struct nf_conntrack_tuple_hash
*
445 __nf_conntrack_find_get(struct net
*net
, u16 zone
,
446 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
448 struct nf_conntrack_tuple_hash
*h
;
453 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
455 ct
= nf_ct_tuplehash_to_ctrack(h
);
456 if (unlikely(nf_ct_is_dying(ct
) ||
457 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
460 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
))) {
471 struct nf_conntrack_tuple_hash
*
472 nf_conntrack_find_get(struct net
*net
, u16 zone
,
473 const struct nf_conntrack_tuple
*tuple
)
475 return __nf_conntrack_find_get(net
, zone
, tuple
,
476 hash_conntrack_raw(tuple
, zone
));
478 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
480 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
482 unsigned int reply_hash
)
484 struct net
*net
= nf_ct_net(ct
);
486 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
487 &net
->ct
.hash
[hash
]);
488 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
489 &net
->ct
.hash
[reply_hash
]);
493 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
495 struct net
*net
= nf_ct_net(ct
);
496 unsigned int hash
, reply_hash
;
497 struct nf_conntrack_tuple_hash
*h
;
498 struct hlist_nulls_node
*n
;
500 unsigned int sequence
;
502 zone
= nf_ct_zone(ct
);
506 sequence
= read_seqcount_begin(&net
->ct
.generation
);
507 hash
= hash_conntrack(net
, zone
,
508 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
509 reply_hash
= hash_conntrack(net
, zone
,
510 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
511 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
513 /* See if there's one in the list already, including reverse */
514 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
515 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
517 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
519 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
520 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
522 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
525 add_timer(&ct
->timeout
);
527 /* The caller holds a reference to this object */
528 atomic_set(&ct
->ct_general
.use
, 2);
529 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
530 nf_conntrack_double_unlock(hash
, reply_hash
);
531 NF_CT_STAT_INC(net
, insert
);
536 nf_conntrack_double_unlock(hash
, reply_hash
);
537 NF_CT_STAT_INC(net
, insert_failed
);
541 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
543 /* deletion from this larval template list happens via nf_ct_put() */
544 void nf_conntrack_tmpl_insert(struct net
*net
, struct nf_conn
*tmpl
)
546 struct ct_pcpu
*pcpu
;
548 __set_bit(IPS_TEMPLATE_BIT
, &tmpl
->status
);
549 __set_bit(IPS_CONFIRMED_BIT
, &tmpl
->status
);
550 nf_conntrack_get(&tmpl
->ct_general
);
552 /* add this conntrack to the (per cpu) tmpl list */
554 tmpl
->cpu
= smp_processor_id();
555 pcpu
= per_cpu_ptr(nf_ct_net(tmpl
)->ct
.pcpu_lists
, tmpl
->cpu
);
557 spin_lock(&pcpu
->lock
);
558 /* Overload tuple linked list to put us in template list. */
559 hlist_nulls_add_head_rcu(&tmpl
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
561 spin_unlock_bh(&pcpu
->lock
);
563 EXPORT_SYMBOL_GPL(nf_conntrack_tmpl_insert
);
565 /* Confirm a connection given skb; places it in hash table */
567 __nf_conntrack_confirm(struct sk_buff
*skb
)
569 unsigned int hash
, reply_hash
;
570 struct nf_conntrack_tuple_hash
*h
;
572 struct nf_conn_help
*help
;
573 struct nf_conn_tstamp
*tstamp
;
574 struct hlist_nulls_node
*n
;
575 enum ip_conntrack_info ctinfo
;
578 unsigned int sequence
;
580 ct
= nf_ct_get(skb
, &ctinfo
);
583 /* ipt_REJECT uses nf_conntrack_attach to attach related
584 ICMP/TCP RST packets in other direction. Actual packet
585 which created connection will be IP_CT_NEW or for an
586 expected connection, IP_CT_RELATED. */
587 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
590 zone
= nf_ct_zone(ct
);
594 sequence
= read_seqcount_begin(&net
->ct
.generation
);
595 /* reuse the hash saved before */
596 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
597 hash
= hash_bucket(hash
, net
);
598 reply_hash
= hash_conntrack(net
, zone
,
599 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
601 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
603 /* We're not in hash table, and we refuse to set up related
604 * connections for unconfirmed conns. But packet copies and
605 * REJECT will give spurious warnings here.
607 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
609 /* No external references means no one else could have
612 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
613 pr_debug("Confirming conntrack %p\n", ct
);
614 /* We have to check the DYING flag after unlink to prevent
615 * a race against nf_ct_get_next_corpse() possibly called from
616 * user context, else we insert an already 'dead' hash, blocking
617 * further use of that particular connection -JM.
619 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
621 if (unlikely(nf_ct_is_dying(ct
)))
624 /* See if there's one in the list already, including reverse:
625 NAT could have grabbed it without realizing, since we're
626 not in the hash. If there is, we lost race. */
627 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
628 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
630 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
632 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
633 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
635 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
638 /* Timer relative to confirmation time, not original
639 setting time, otherwise we'd get timer wrap in
640 weird delay cases. */
641 ct
->timeout
.expires
+= jiffies
;
642 add_timer(&ct
->timeout
);
643 atomic_inc(&ct
->ct_general
.use
);
644 ct
->status
|= IPS_CONFIRMED
;
646 /* set conntrack timestamp, if enabled. */
647 tstamp
= nf_conn_tstamp_find(ct
);
649 if (skb
->tstamp
.tv64
== 0)
650 __net_timestamp(skb
);
652 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
654 /* Since the lookup is lockless, hash insertion must be done after
655 * starting the timer and setting the CONFIRMED bit. The RCU barriers
656 * guarantee that no other CPU can find the conntrack before the above
657 * stores are visible.
659 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
660 nf_conntrack_double_unlock(hash
, reply_hash
);
661 NF_CT_STAT_INC(net
, insert
);
664 help
= nfct_help(ct
);
665 if (help
&& help
->helper
)
666 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
668 nf_conntrack_event_cache(master_ct(ct
) ?
669 IPCT_RELATED
: IPCT_NEW
, ct
);
673 nf_ct_add_to_dying_list(ct
);
674 nf_conntrack_double_unlock(hash
, reply_hash
);
675 NF_CT_STAT_INC(net
, insert_failed
);
679 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
681 /* Returns true if a connection correspondings to the tuple (required
684 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
685 const struct nf_conn
*ignored_conntrack
)
687 struct net
*net
= nf_ct_net(ignored_conntrack
);
688 struct nf_conntrack_tuple_hash
*h
;
689 struct hlist_nulls_node
*n
;
691 u16 zone
= nf_ct_zone(ignored_conntrack
);
692 unsigned int hash
= hash_conntrack(net
, zone
, tuple
);
694 /* Disable BHs the entire time since we need to disable them at
695 * least once for the stats anyway.
698 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
699 ct
= nf_ct_tuplehash_to_ctrack(h
);
700 if (ct
!= ignored_conntrack
&&
701 nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
702 nf_ct_zone(ct
) == zone
) {
703 NF_CT_STAT_INC(net
, found
);
704 rcu_read_unlock_bh();
707 NF_CT_STAT_INC(net
, searched
);
709 rcu_read_unlock_bh();
713 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
715 #define NF_CT_EVICTION_RANGE 8
717 /* There's a small race here where we may free a just-assured
718 connection. Too bad: we're in trouble anyway. */
719 static noinline
int early_drop(struct net
*net
, unsigned int _hash
)
721 /* Use oldest entry, which is roughly LRU */
722 struct nf_conntrack_tuple_hash
*h
;
723 struct nf_conn
*ct
= NULL
, *tmp
;
724 struct hlist_nulls_node
*n
;
725 unsigned int i
= 0, cnt
= 0;
727 unsigned int hash
, sequence
;
732 sequence
= read_seqcount_begin(&net
->ct
.generation
);
733 hash
= hash_bucket(_hash
, net
);
734 for (; i
< net
->ct
.htable_size
; i
++) {
735 lockp
= &nf_conntrack_locks
[hash
% CONNTRACK_LOCKS
];
737 if (read_seqcount_retry(&net
->ct
.generation
, sequence
)) {
741 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
743 tmp
= nf_ct_tuplehash_to_ctrack(h
);
744 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
) &&
745 !nf_ct_is_dying(tmp
) &&
746 atomic_inc_not_zero(&tmp
->ct_general
.use
)) {
753 hash
= (hash
+ 1) % net
->ct
.htable_size
;
756 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
765 if (del_timer(&ct
->timeout
)) {
766 if (nf_ct_delete(ct
, 0, 0)) {
768 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
775 void init_nf_conntrack_hash_rnd(void)
780 * Why not initialize nf_conntrack_rnd in a "init()" function ?
781 * Because there isn't enough entropy when system initializing,
782 * and we initialize it as late as possible.
785 get_random_bytes(&rand
, sizeof(rand
));
787 cmpxchg(&nf_conntrack_hash_rnd
, 0, rand
);
790 static struct nf_conn
*
791 __nf_conntrack_alloc(struct net
*net
, u16 zone
,
792 const struct nf_conntrack_tuple
*orig
,
793 const struct nf_conntrack_tuple
*repl
,
798 if (unlikely(!nf_conntrack_hash_rnd
)) {
799 init_nf_conntrack_hash_rnd();
800 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
801 hash
= hash_conntrack_raw(orig
, zone
);
804 /* We don't want any race condition at early drop stage */
805 atomic_inc(&net
->ct
.count
);
807 if (nf_conntrack_max
&&
808 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
809 if (!early_drop(net
, hash
)) {
810 atomic_dec(&net
->ct
.count
);
811 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
812 return ERR_PTR(-ENOMEM
);
817 * Do not use kmem_cache_zalloc(), as this cache uses
818 * SLAB_DESTROY_BY_RCU.
820 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
822 atomic_dec(&net
->ct
.count
);
823 return ERR_PTR(-ENOMEM
);
825 spin_lock_init(&ct
->lock
);
826 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
827 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
828 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
829 /* save hash for reusing when confirming */
830 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
832 /* Don't set timer yet: wait for confirmation */
833 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
834 write_pnet(&ct
->ct_net
, net
);
835 memset(&ct
->__nfct_init_offset
[0], 0,
836 offsetof(struct nf_conn
, proto
) -
837 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
838 #ifdef CONFIG_NF_CONNTRACK_ZONES
840 struct nf_conntrack_zone
*nf_ct_zone
;
842 nf_ct_zone
= nf_ct_ext_add(ct
, NF_CT_EXT_ZONE
, GFP_ATOMIC
);
845 nf_ct_zone
->id
= zone
;
848 /* Because we use RCU lookups, we set ct_general.use to zero before
849 * this is inserted in any list.
851 atomic_set(&ct
->ct_general
.use
, 0);
854 #ifdef CONFIG_NF_CONNTRACK_ZONES
856 atomic_dec(&net
->ct
.count
);
857 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
858 return ERR_PTR(-ENOMEM
);
862 struct nf_conn
*nf_conntrack_alloc(struct net
*net
, u16 zone
,
863 const struct nf_conntrack_tuple
*orig
,
864 const struct nf_conntrack_tuple
*repl
,
867 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
869 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
871 void nf_conntrack_free(struct nf_conn
*ct
)
873 struct net
*net
= nf_ct_net(ct
);
875 /* A freed object has refcnt == 0, that's
876 * the golden rule for SLAB_DESTROY_BY_RCU
878 NF_CT_ASSERT(atomic_read(&ct
->ct_general
.use
) == 0);
880 nf_ct_ext_destroy(ct
);
882 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
883 smp_mb__before_atomic();
884 atomic_dec(&net
->ct
.count
);
886 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
889 /* Allocate a new conntrack: we return -ENOMEM if classification
890 failed due to stress. Otherwise it really is unclassifiable. */
891 static struct nf_conntrack_tuple_hash
*
892 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
893 const struct nf_conntrack_tuple
*tuple
,
894 struct nf_conntrack_l3proto
*l3proto
,
895 struct nf_conntrack_l4proto
*l4proto
,
897 unsigned int dataoff
, u32 hash
)
900 struct nf_conn_help
*help
;
901 struct nf_conntrack_tuple repl_tuple
;
902 struct nf_conntrack_ecache
*ecache
;
903 struct nf_conntrack_expect
*exp
= NULL
;
904 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
905 struct nf_conn_timeout
*timeout_ext
;
906 unsigned int *timeouts
;
908 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
909 pr_debug("Can't invert tuple.\n");
913 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
916 return (struct nf_conntrack_tuple_hash
*)ct
;
918 if (tmpl
&& nfct_synproxy(tmpl
)) {
919 nfct_seqadj_ext_add(ct
);
920 nfct_synproxy_ext_add(ct
);
923 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
925 timeouts
= NF_CT_TIMEOUT_EXT_DATA(timeout_ext
);
927 timeouts
= l4proto
->get_timeouts(net
);
929 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
930 nf_conntrack_free(ct
);
931 pr_debug("init conntrack: can't track with proto module\n");
936 nf_ct_timeout_ext_add(ct
, timeout_ext
->timeout
, GFP_ATOMIC
);
938 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
939 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
940 nf_ct_labels_ext_add(ct
);
942 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
943 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
944 ecache
? ecache
->expmask
: 0,
948 if (net
->ct
.expect_count
) {
949 spin_lock(&nf_conntrack_expect_lock
);
950 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
952 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
954 /* Welcome, Mr. Bond. We've been expecting you... */
955 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
956 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
957 ct
->master
= exp
->master
;
959 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
962 rcu_assign_pointer(help
->helper
, exp
->helper
);
965 #ifdef CONFIG_NF_CONNTRACK_MARK
966 ct
->mark
= exp
->master
->mark
;
968 #ifdef CONFIG_NF_CONNTRACK_SECMARK
969 ct
->secmark
= exp
->master
->secmark
;
971 NF_CT_STAT_INC(net
, expect_new
);
973 spin_unlock(&nf_conntrack_expect_lock
);
976 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
977 NF_CT_STAT_INC(net
, new);
980 /* Now it is inserted into the unconfirmed list, bump refcount */
981 nf_conntrack_get(&ct
->ct_general
);
982 nf_ct_add_to_unconfirmed_list(ct
);
988 exp
->expectfn(ct
, exp
);
989 nf_ct_expect_put(exp
);
992 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
995 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
996 static inline struct nf_conn
*
997 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
999 unsigned int dataoff
,
1002 struct nf_conntrack_l3proto
*l3proto
,
1003 struct nf_conntrack_l4proto
*l4proto
,
1005 enum ip_conntrack_info
*ctinfo
)
1007 struct nf_conntrack_tuple tuple
;
1008 struct nf_conntrack_tuple_hash
*h
;
1010 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
1013 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1014 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
1016 pr_debug("resolve_normal_ct: Can't get tuple\n");
1020 /* look for tuple match */
1021 hash
= hash_conntrack_raw(&tuple
, zone
);
1022 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1024 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1025 skb
, dataoff
, hash
);
1031 ct
= nf_ct_tuplehash_to_ctrack(h
);
1033 /* It exists; we have (non-exclusive) reference. */
1034 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1035 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1036 /* Please set reply bit if this packet OK */
1039 /* Once we've had two way comms, always ESTABLISHED. */
1040 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1041 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
1042 *ctinfo
= IP_CT_ESTABLISHED
;
1043 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1044 pr_debug("nf_conntrack_in: related packet for %p\n",
1046 *ctinfo
= IP_CT_RELATED
;
1048 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
1049 *ctinfo
= IP_CT_NEW
;
1053 skb
->nfct
= &ct
->ct_general
;
1054 skb
->nfctinfo
= *ctinfo
;
1059 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1060 struct sk_buff
*skb
)
1062 struct nf_conn
*ct
, *tmpl
= NULL
;
1063 enum ip_conntrack_info ctinfo
;
1064 struct nf_conntrack_l3proto
*l3proto
;
1065 struct nf_conntrack_l4proto
*l4proto
;
1066 unsigned int *timeouts
;
1067 unsigned int dataoff
;
1073 /* Previously seen (loopback or untracked)? Ignore. */
1074 tmpl
= (struct nf_conn
*)skb
->nfct
;
1075 if (!nf_ct_is_template(tmpl
)) {
1076 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1082 /* rcu_read_lock()ed by nf_hook_slow */
1083 l3proto
= __nf_ct_l3proto_find(pf
);
1084 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1085 &dataoff
, &protonum
);
1087 pr_debug("not prepared to track yet or error occurred\n");
1088 NF_CT_STAT_INC_ATOMIC(net
, error
);
1089 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1094 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1096 /* It may be an special packet, error, unclean...
1097 * inverse of the return code tells to the netfilter
1098 * core what to do with the packet. */
1099 if (l4proto
->error
!= NULL
) {
1100 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
1103 NF_CT_STAT_INC_ATOMIC(net
, error
);
1104 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1108 /* ICMP[v6] protocol trackers may assign one conntrack. */
1113 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1114 l3proto
, l4proto
, &set_reply
, &ctinfo
);
1116 /* Not valid part of a connection */
1117 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1123 /* Too stressed to deal. */
1124 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1129 NF_CT_ASSERT(skb
->nfct
);
1131 /* Decide what timeout policy we want to apply to this flow. */
1132 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1134 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1136 /* Invalid: inverse of the return code tells
1137 * the netfilter core what to do */
1138 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1139 nf_conntrack_put(skb
->nfct
);
1141 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1142 if (ret
== -NF_DROP
)
1143 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1148 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1149 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1152 /* Special case: we have to repeat this hook, assign the
1153 * template again to this packet. We assume that this packet
1154 * has no conntrack assigned. This is used by nf_ct_tcp. */
1155 if (ret
== NF_REPEAT
)
1156 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
1163 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1165 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1166 const struct nf_conntrack_tuple
*orig
)
1171 ret
= nf_ct_invert_tuple(inverse
, orig
,
1172 __nf_ct_l3proto_find(orig
->src
.l3num
),
1173 __nf_ct_l4proto_find(orig
->src
.l3num
,
1174 orig
->dst
.protonum
));
1178 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1180 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1181 implicitly racy: see __nf_conntrack_confirm */
1182 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1183 const struct nf_conntrack_tuple
*newreply
)
1185 struct nf_conn_help
*help
= nfct_help(ct
);
1187 /* Should be unconfirmed, so not in hash table yet */
1188 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1190 pr_debug("Altering reply tuple of %p to ", ct
);
1191 nf_ct_dump_tuple(newreply
);
1193 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1194 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1198 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1201 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1203 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1204 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1205 enum ip_conntrack_info ctinfo
,
1206 const struct sk_buff
*skb
,
1207 unsigned long extra_jiffies
,
1210 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1213 /* Only update if this is not a fixed timeout */
1214 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1217 /* If not in hash table, timer will not be active yet */
1218 if (!nf_ct_is_confirmed(ct
)) {
1219 ct
->timeout
.expires
= extra_jiffies
;
1221 unsigned long newtime
= jiffies
+ extra_jiffies
;
1223 /* Only update the timeout if the new timeout is at least
1224 HZ jiffies from the old timeout. Need del_timer for race
1225 avoidance (may already be dying). */
1226 if (newtime
- ct
->timeout
.expires
>= HZ
)
1227 mod_timer_pending(&ct
->timeout
, newtime
);
1232 struct nf_conn_acct
*acct
;
1234 acct
= nf_conn_acct_find(ct
);
1236 struct nf_conn_counter
*counter
= acct
->counter
;
1238 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1239 atomic64_add(skb
->len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1243 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1245 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1246 enum ip_conntrack_info ctinfo
,
1247 const struct sk_buff
*skb
,
1251 struct nf_conn_acct
*acct
;
1253 acct
= nf_conn_acct_find(ct
);
1255 struct nf_conn_counter
*counter
= acct
->counter
;
1257 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1258 atomic64_add(skb
->len
- skb_network_offset(skb
),
1259 &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1263 if (del_timer(&ct
->timeout
)) {
1264 ct
->timeout
.function((unsigned long)ct
);
1269 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1271 #ifdef CONFIG_NF_CONNTRACK_ZONES
1272 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1273 .len
= sizeof(struct nf_conntrack_zone
),
1274 .align
= __alignof__(struct nf_conntrack_zone
),
1275 .id
= NF_CT_EXT_ZONE
,
1279 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1281 #include <linux/netfilter/nfnetlink.h>
1282 #include <linux/netfilter/nfnetlink_conntrack.h>
1283 #include <linux/mutex.h>
1285 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1286 * in ip_conntrack_core, since we don't want the protocols to autoload
1287 * or depend on ctnetlink */
1288 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1289 const struct nf_conntrack_tuple
*tuple
)
1291 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1292 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1293 goto nla_put_failure
;
1299 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1301 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1302 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1303 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1305 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1307 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1308 struct nf_conntrack_tuple
*t
)
1310 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1313 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1314 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1318 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1320 int nf_ct_port_nlattr_tuple_size(void)
1322 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1324 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1327 /* Used by ipt_REJECT and ip6t_REJECT. */
1328 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1331 enum ip_conntrack_info ctinfo
;
1333 /* This ICMP is in reverse direction to the packet which caused it */
1334 ct
= nf_ct_get(skb
, &ctinfo
);
1335 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1336 ctinfo
= IP_CT_RELATED_REPLY
;
1338 ctinfo
= IP_CT_RELATED
;
1340 /* Attach to new skbuff, and increment count */
1341 nskb
->nfct
= &ct
->ct_general
;
1342 nskb
->nfctinfo
= ctinfo
;
1343 nf_conntrack_get(nskb
->nfct
);
1346 /* Bring out ya dead! */
1347 static struct nf_conn
*
1348 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1349 void *data
, unsigned int *bucket
)
1351 struct nf_conntrack_tuple_hash
*h
;
1353 struct hlist_nulls_node
*n
;
1357 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1358 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1361 if (*bucket
< net
->ct
.htable_size
) {
1362 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1363 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1365 ct
= nf_ct_tuplehash_to_ctrack(h
);
1374 for_each_possible_cpu(cpu
) {
1375 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1377 spin_lock_bh(&pcpu
->lock
);
1378 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1379 ct
= nf_ct_tuplehash_to_ctrack(h
);
1381 set_bit(IPS_DYING_BIT
, &ct
->status
);
1383 spin_unlock_bh(&pcpu
->lock
);
1387 atomic_inc(&ct
->ct_general
.use
);
1393 void nf_ct_iterate_cleanup(struct net
*net
,
1394 int (*iter
)(struct nf_conn
*i
, void *data
),
1395 void *data
, u32 portid
, int report
)
1398 unsigned int bucket
= 0;
1400 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1401 /* Time to push up daises... */
1402 if (del_timer(&ct
->timeout
))
1403 nf_ct_delete(ct
, portid
, report
);
1405 /* ... else the timer will get him soon. */
1410 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1412 static int kill_all(struct nf_conn
*i
, void *data
)
1417 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1419 if (is_vmalloc_addr(hash
))
1422 free_pages((unsigned long)hash
,
1423 get_order(sizeof(struct hlist_head
) * size
));
1425 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1427 static int untrack_refs(void)
1431 for_each_possible_cpu(cpu
) {
1432 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1434 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1439 void nf_conntrack_cleanup_start(void)
1441 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1444 void nf_conntrack_cleanup_end(void)
1446 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1447 while (untrack_refs() > 0)
1450 #ifdef CONFIG_NF_CONNTRACK_ZONES
1451 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1453 nf_conntrack_proto_fini();
1454 nf_conntrack_seqadj_fini();
1455 nf_conntrack_labels_fini();
1456 nf_conntrack_helper_fini();
1457 nf_conntrack_timeout_fini();
1458 nf_conntrack_ecache_fini();
1459 nf_conntrack_tstamp_fini();
1460 nf_conntrack_acct_fini();
1461 nf_conntrack_expect_fini();
1465 * Mishearing the voices in his head, our hero wonders how he's
1466 * supposed to kill the mall.
1468 void nf_conntrack_cleanup_net(struct net
*net
)
1472 list_add(&net
->exit_list
, &single
);
1473 nf_conntrack_cleanup_net_list(&single
);
1476 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1482 * This makes sure all current packets have passed through
1483 * netfilter framework. Roll on, two-stage module
1489 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1490 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1491 if (atomic_read(&net
->ct
.count
) != 0)
1496 goto i_see_dead_people
;
1499 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1500 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1501 nf_conntrack_proto_pernet_fini(net
);
1502 nf_conntrack_helper_pernet_fini(net
);
1503 nf_conntrack_ecache_pernet_fini(net
);
1504 nf_conntrack_tstamp_pernet_fini(net
);
1505 nf_conntrack_acct_pernet_fini(net
);
1506 nf_conntrack_expect_pernet_fini(net
);
1507 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1508 kfree(net
->ct
.slabname
);
1509 free_percpu(net
->ct
.stat
);
1510 free_percpu(net
->ct
.pcpu_lists
);
1514 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1516 struct hlist_nulls_head
*hash
;
1517 unsigned int nr_slots
, i
;
1520 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1521 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1522 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1523 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1526 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1531 for (i
= 0; i
< nr_slots
; i
++)
1532 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1536 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1538 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1541 unsigned int hashsize
, old_size
;
1542 struct hlist_nulls_head
*hash
, *old_hash
;
1543 struct nf_conntrack_tuple_hash
*h
;
1546 if (current
->nsproxy
->net_ns
!= &init_net
)
1549 /* On boot, we can set this without any fancy locking. */
1550 if (!nf_conntrack_htable_size
)
1551 return param_set_uint(val
, kp
);
1553 rc
= kstrtouint(val
, 0, &hashsize
);
1559 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1564 nf_conntrack_all_lock();
1565 write_seqcount_begin(&init_net
.ct
.generation
);
1567 /* Lookups in the old hash might happen in parallel, which means we
1568 * might get false negatives during connection lookup. New connections
1569 * created because of a false negative won't make it into the hash
1570 * though since that required taking the locks.
1573 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1574 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1575 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1576 struct nf_conntrack_tuple_hash
, hnnode
);
1577 ct
= nf_ct_tuplehash_to_ctrack(h
);
1578 hlist_nulls_del_rcu(&h
->hnnode
);
1579 bucket
= __hash_conntrack(&h
->tuple
, nf_ct_zone(ct
),
1581 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1584 old_size
= init_net
.ct
.htable_size
;
1585 old_hash
= init_net
.ct
.hash
;
1587 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1588 init_net
.ct
.hash
= hash
;
1590 write_seqcount_end(&init_net
.ct
.generation
);
1591 nf_conntrack_all_unlock();
1594 nf_ct_free_hashtable(old_hash
, old_size
);
1597 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1599 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1600 &nf_conntrack_htable_size
, 0600);
1602 void nf_ct_untracked_status_or(unsigned long bits
)
1606 for_each_possible_cpu(cpu
)
1607 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1609 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1611 int nf_conntrack_init_start(void)
1616 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
1617 spin_lock_init(&nf_conntrack_locks
[i
]);
1619 if (!nf_conntrack_htable_size
) {
1620 /* Idea from tcp.c: use 1/16384 of memory.
1621 * On i386: 32MB machine has 512 buckets.
1622 * >= 1GB machines have 16384 buckets.
1623 * >= 4GB machines have 65536 buckets.
1625 nf_conntrack_htable_size
1626 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1627 / sizeof(struct hlist_head
));
1628 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
1629 nf_conntrack_htable_size
= 65536;
1630 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1631 nf_conntrack_htable_size
= 16384;
1632 if (nf_conntrack_htable_size
< 32)
1633 nf_conntrack_htable_size
= 32;
1635 /* Use a max. factor of four by default to get the same max as
1636 * with the old struct list_heads. When a table size is given
1637 * we use the old value of 8 to avoid reducing the max.
1641 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1643 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1644 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1647 ret
= nf_conntrack_expect_init();
1651 ret
= nf_conntrack_acct_init();
1655 ret
= nf_conntrack_tstamp_init();
1659 ret
= nf_conntrack_ecache_init();
1663 ret
= nf_conntrack_timeout_init();
1667 ret
= nf_conntrack_helper_init();
1671 ret
= nf_conntrack_labels_init();
1675 ret
= nf_conntrack_seqadj_init();
1679 #ifdef CONFIG_NF_CONNTRACK_ZONES
1680 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1684 ret
= nf_conntrack_proto_init();
1688 /* Set up fake conntrack: to never be deleted, not in any hashes */
1689 for_each_possible_cpu(cpu
) {
1690 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1691 write_pnet(&ct
->ct_net
, &init_net
);
1692 atomic_set(&ct
->ct_general
.use
, 1);
1694 /* - and look it like as a confirmed connection */
1695 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1699 #ifdef CONFIG_NF_CONNTRACK_ZONES
1700 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1703 nf_conntrack_seqadj_fini();
1705 nf_conntrack_labels_fini();
1707 nf_conntrack_helper_fini();
1709 nf_conntrack_timeout_fini();
1711 nf_conntrack_ecache_fini();
1713 nf_conntrack_tstamp_fini();
1715 nf_conntrack_acct_fini();
1717 nf_conntrack_expect_fini();
1722 void nf_conntrack_init_end(void)
1724 /* For use by REJECT target */
1725 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1726 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1730 * We need to use special "null" values, not used in hash table
1732 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1733 #define DYING_NULLS_VAL ((1<<30)+1)
1734 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1736 int nf_conntrack_init_net(struct net
*net
)
1741 atomic_set(&net
->ct
.count
, 0);
1742 seqcount_init(&net
->ct
.generation
);
1744 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
1745 if (!net
->ct
.pcpu_lists
)
1748 for_each_possible_cpu(cpu
) {
1749 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1751 spin_lock_init(&pcpu
->lock
);
1752 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1753 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
1754 INIT_HLIST_NULLS_HEAD(&pcpu
->tmpl
, TEMPLATE_NULLS_VAL
);
1757 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1759 goto err_pcpu_lists
;
1761 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1762 if (!net
->ct
.slabname
)
1765 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1766 sizeof(struct nf_conn
), 0,
1767 SLAB_DESTROY_BY_RCU
, NULL
);
1768 if (!net
->ct
.nf_conntrack_cachep
) {
1769 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1773 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1774 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
, 1);
1775 if (!net
->ct
.hash
) {
1776 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1779 ret
= nf_conntrack_expect_pernet_init(net
);
1782 ret
= nf_conntrack_acct_pernet_init(net
);
1785 ret
= nf_conntrack_tstamp_pernet_init(net
);
1788 ret
= nf_conntrack_ecache_pernet_init(net
);
1791 ret
= nf_conntrack_helper_pernet_init(net
);
1794 ret
= nf_conntrack_proto_pernet_init(net
);
1800 nf_conntrack_helper_pernet_fini(net
);
1802 nf_conntrack_ecache_pernet_fini(net
);
1804 nf_conntrack_tstamp_pernet_fini(net
);
1806 nf_conntrack_acct_pernet_fini(net
);
1808 nf_conntrack_expect_pernet_fini(net
);
1810 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1812 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1814 kfree(net
->ct
.slabname
);
1816 free_percpu(net
->ct
.stat
);
1818 free_percpu(net
->ct
.pcpu_lists
);