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 ((u64
)hash
* size
) >> 32;
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 static void death_by_event(unsigned long ul_conntrack
)
357 struct nf_conn
*ct
= (void *)ul_conntrack
;
358 struct net
*net
= nf_ct_net(ct
);
359 struct nf_conntrack_ecache
*ecache
= nf_ct_ecache_find(ct
);
361 BUG_ON(ecache
== NULL
);
363 if (nf_conntrack_event(IPCT_DESTROY
, ct
) < 0) {
364 /* bad luck, let's retry again */
365 ecache
->timeout
.expires
= jiffies
+
366 (prandom_u32() % net
->ct
.sysctl_events_retry_timeout
);
367 add_timer(&ecache
->timeout
);
370 /* we've got the event delivered, now it's dying */
371 set_bit(IPS_DYING_BIT
, &ct
->status
);
375 static void nf_ct_dying_timeout(struct nf_conn
*ct
)
377 struct net
*net
= nf_ct_net(ct
);
378 struct nf_conntrack_ecache
*ecache
= nf_ct_ecache_find(ct
);
380 BUG_ON(ecache
== NULL
);
382 /* set a new timer to retry event delivery */
383 setup_timer(&ecache
->timeout
, death_by_event
, (unsigned long)ct
);
384 ecache
->timeout
.expires
= jiffies
+
385 (prandom_u32() % net
->ct
.sysctl_events_retry_timeout
);
386 add_timer(&ecache
->timeout
);
389 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
391 struct nf_conn_tstamp
*tstamp
;
393 tstamp
= nf_conn_tstamp_find(ct
);
394 if (tstamp
&& tstamp
->stop
== 0)
395 tstamp
->stop
= ktime_to_ns(ktime_get_real());
397 if (!nf_ct_is_dying(ct
) &&
398 unlikely(nf_conntrack_event_report(IPCT_DESTROY
, ct
,
399 portid
, report
) < 0)) {
400 /* destroy event was not delivered */
401 nf_ct_delete_from_lists(ct
);
402 nf_ct_dying_timeout(ct
);
405 set_bit(IPS_DYING_BIT
, &ct
->status
);
406 nf_ct_delete_from_lists(ct
);
410 EXPORT_SYMBOL_GPL(nf_ct_delete
);
412 static void death_by_timeout(unsigned long ul_conntrack
)
414 nf_ct_delete((struct nf_conn
*)ul_conntrack
, 0, 0);
418 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
419 const struct nf_conntrack_tuple
*tuple
,
422 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
424 /* A conntrack can be recreated with the equal tuple,
425 * so we need to check that the conntrack is confirmed
427 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
428 nf_ct_zone(ct
) == zone
&&
429 nf_ct_is_confirmed(ct
);
434 * - Caller must take a reference on returned object
435 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
437 static struct nf_conntrack_tuple_hash
*
438 ____nf_conntrack_find(struct net
*net
, u16 zone
,
439 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
441 struct nf_conntrack_tuple_hash
*h
;
442 struct hlist_nulls_node
*n
;
443 unsigned int bucket
= hash_bucket(hash
, net
);
445 /* Disable BHs the entire time since we normally need to disable them
446 * at least once for the stats anyway.
450 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[bucket
], hnnode
) {
451 if (nf_ct_key_equal(h
, tuple
, zone
)) {
452 NF_CT_STAT_INC(net
, found
);
456 NF_CT_STAT_INC(net
, searched
);
459 * if the nulls value we got at the end of this lookup is
460 * not the expected one, we must restart lookup.
461 * We probably met an item that was moved to another chain.
463 if (get_nulls_value(n
) != bucket
) {
464 NF_CT_STAT_INC(net
, search_restart
);
472 /* Find a connection corresponding to a tuple. */
473 static struct nf_conntrack_tuple_hash
*
474 __nf_conntrack_find_get(struct net
*net
, u16 zone
,
475 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
477 struct nf_conntrack_tuple_hash
*h
;
482 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
484 ct
= nf_ct_tuplehash_to_ctrack(h
);
485 if (unlikely(nf_ct_is_dying(ct
) ||
486 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
489 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
))) {
500 struct nf_conntrack_tuple_hash
*
501 nf_conntrack_find_get(struct net
*net
, u16 zone
,
502 const struct nf_conntrack_tuple
*tuple
)
504 return __nf_conntrack_find_get(net
, zone
, tuple
,
505 hash_conntrack_raw(tuple
, zone
));
507 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
509 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
511 unsigned int reply_hash
)
513 struct net
*net
= nf_ct_net(ct
);
515 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
516 &net
->ct
.hash
[hash
]);
517 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
518 &net
->ct
.hash
[reply_hash
]);
522 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
524 struct net
*net
= nf_ct_net(ct
);
525 unsigned int hash
, reply_hash
;
526 struct nf_conntrack_tuple_hash
*h
;
527 struct hlist_nulls_node
*n
;
529 unsigned int sequence
;
531 zone
= nf_ct_zone(ct
);
535 sequence
= read_seqcount_begin(&net
->ct
.generation
);
536 hash
= hash_conntrack(net
, zone
,
537 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
538 reply_hash
= hash_conntrack(net
, zone
,
539 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
540 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
542 /* See if there's one in the list already, including reverse */
543 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
544 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
546 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
548 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
549 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
551 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
554 add_timer(&ct
->timeout
);
556 /* The caller holds a reference to this object */
557 atomic_set(&ct
->ct_general
.use
, 2);
558 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
559 nf_conntrack_double_unlock(hash
, reply_hash
);
560 NF_CT_STAT_INC(net
, insert
);
565 nf_conntrack_double_unlock(hash
, reply_hash
);
566 NF_CT_STAT_INC(net
, insert_failed
);
570 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
572 /* deletion from this larval template list happens via nf_ct_put() */
573 void nf_conntrack_tmpl_insert(struct net
*net
, struct nf_conn
*tmpl
)
575 struct ct_pcpu
*pcpu
;
577 __set_bit(IPS_TEMPLATE_BIT
, &tmpl
->status
);
578 __set_bit(IPS_CONFIRMED_BIT
, &tmpl
->status
);
579 nf_conntrack_get(&tmpl
->ct_general
);
581 /* add this conntrack to the (per cpu) tmpl list */
583 tmpl
->cpu
= smp_processor_id();
584 pcpu
= per_cpu_ptr(nf_ct_net(tmpl
)->ct
.pcpu_lists
, tmpl
->cpu
);
586 spin_lock(&pcpu
->lock
);
587 /* Overload tuple linked list to put us in template list. */
588 hlist_nulls_add_head_rcu(&tmpl
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
590 spin_unlock_bh(&pcpu
->lock
);
592 EXPORT_SYMBOL_GPL(nf_conntrack_tmpl_insert
);
594 /* Confirm a connection given skb; places it in hash table */
596 __nf_conntrack_confirm(struct sk_buff
*skb
)
598 unsigned int hash
, reply_hash
;
599 struct nf_conntrack_tuple_hash
*h
;
601 struct nf_conn_help
*help
;
602 struct nf_conn_tstamp
*tstamp
;
603 struct hlist_nulls_node
*n
;
604 enum ip_conntrack_info ctinfo
;
607 unsigned int sequence
;
609 ct
= nf_ct_get(skb
, &ctinfo
);
612 /* ipt_REJECT uses nf_conntrack_attach to attach related
613 ICMP/TCP RST packets in other direction. Actual packet
614 which created connection will be IP_CT_NEW or for an
615 expected connection, IP_CT_RELATED. */
616 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
619 zone
= nf_ct_zone(ct
);
623 sequence
= read_seqcount_begin(&net
->ct
.generation
);
624 /* reuse the hash saved before */
625 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
626 hash
= hash_bucket(hash
, net
);
627 reply_hash
= hash_conntrack(net
, zone
,
628 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
630 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
632 /* We're not in hash table, and we refuse to set up related
633 * connections for unconfirmed conns. But packet copies and
634 * REJECT will give spurious warnings here.
636 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
638 /* No external references means no one else could have
641 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
642 pr_debug("Confirming conntrack %p\n", ct
);
643 /* We have to check the DYING flag inside the lock to prevent
644 a race against nf_ct_get_next_corpse() possibly called from
645 user context, else we insert an already 'dead' hash, blocking
646 further use of that particular connection -JM */
648 if (unlikely(nf_ct_is_dying(ct
))) {
649 nf_conntrack_double_unlock(hash
, reply_hash
);
654 /* See if there's one in the list already, including reverse:
655 NAT could have grabbed it without realizing, since we're
656 not in the hash. If there is, we lost race. */
657 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
658 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
660 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
662 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
663 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
665 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
668 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
670 /* Timer relative to confirmation time, not original
671 setting time, otherwise we'd get timer wrap in
672 weird delay cases. */
673 ct
->timeout
.expires
+= jiffies
;
674 add_timer(&ct
->timeout
);
675 atomic_inc(&ct
->ct_general
.use
);
676 ct
->status
|= IPS_CONFIRMED
;
678 /* set conntrack timestamp, if enabled. */
679 tstamp
= nf_conn_tstamp_find(ct
);
681 if (skb
->tstamp
.tv64
== 0)
682 __net_timestamp(skb
);
684 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
686 /* Since the lookup is lockless, hash insertion must be done after
687 * starting the timer and setting the CONFIRMED bit. The RCU barriers
688 * guarantee that no other CPU can find the conntrack before the above
689 * stores are visible.
691 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
692 nf_conntrack_double_unlock(hash
, reply_hash
);
693 NF_CT_STAT_INC(net
, insert
);
696 help
= nfct_help(ct
);
697 if (help
&& help
->helper
)
698 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
700 nf_conntrack_event_cache(master_ct(ct
) ?
701 IPCT_RELATED
: IPCT_NEW
, ct
);
705 nf_conntrack_double_unlock(hash
, reply_hash
);
706 NF_CT_STAT_INC(net
, insert_failed
);
710 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
712 /* Returns true if a connection correspondings to the tuple (required
715 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
716 const struct nf_conn
*ignored_conntrack
)
718 struct net
*net
= nf_ct_net(ignored_conntrack
);
719 struct nf_conntrack_tuple_hash
*h
;
720 struct hlist_nulls_node
*n
;
722 u16 zone
= nf_ct_zone(ignored_conntrack
);
723 unsigned int hash
= hash_conntrack(net
, zone
, tuple
);
725 /* Disable BHs the entire time since we need to disable them at
726 * least once for the stats anyway.
729 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
730 ct
= nf_ct_tuplehash_to_ctrack(h
);
731 if (ct
!= ignored_conntrack
&&
732 nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
733 nf_ct_zone(ct
) == zone
) {
734 NF_CT_STAT_INC(net
, found
);
735 rcu_read_unlock_bh();
738 NF_CT_STAT_INC(net
, searched
);
740 rcu_read_unlock_bh();
744 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
746 #define NF_CT_EVICTION_RANGE 8
748 /* There's a small race here where we may free a just-assured
749 connection. Too bad: we're in trouble anyway. */
750 static noinline
int early_drop(struct net
*net
, unsigned int _hash
)
752 /* Use oldest entry, which is roughly LRU */
753 struct nf_conntrack_tuple_hash
*h
;
754 struct nf_conn
*ct
= NULL
, *tmp
;
755 struct hlist_nulls_node
*n
;
756 unsigned int i
= 0, cnt
= 0;
758 unsigned int hash
, sequence
;
763 sequence
= read_seqcount_begin(&net
->ct
.generation
);
764 hash
= hash_bucket(_hash
, net
);
765 for (; i
< net
->ct
.htable_size
; i
++) {
766 lockp
= &nf_conntrack_locks
[hash
% CONNTRACK_LOCKS
];
768 if (read_seqcount_retry(&net
->ct
.generation
, sequence
)) {
772 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
774 tmp
= nf_ct_tuplehash_to_ctrack(h
);
775 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
) &&
776 !nf_ct_is_dying(tmp
) &&
777 atomic_inc_not_zero(&tmp
->ct_general
.use
)) {
784 hash
= (hash
+ 1) % net
->ct
.htable_size
;
787 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
796 if (del_timer(&ct
->timeout
)) {
797 if (nf_ct_delete(ct
, 0, 0)) {
799 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
806 void init_nf_conntrack_hash_rnd(void)
811 * Why not initialize nf_conntrack_rnd in a "init()" function ?
812 * Because there isn't enough entropy when system initializing,
813 * and we initialize it as late as possible.
816 get_random_bytes(&rand
, sizeof(rand
));
818 cmpxchg(&nf_conntrack_hash_rnd
, 0, rand
);
821 static struct nf_conn
*
822 __nf_conntrack_alloc(struct net
*net
, u16 zone
,
823 const struct nf_conntrack_tuple
*orig
,
824 const struct nf_conntrack_tuple
*repl
,
829 if (unlikely(!nf_conntrack_hash_rnd
)) {
830 init_nf_conntrack_hash_rnd();
831 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
832 hash
= hash_conntrack_raw(orig
, zone
);
835 /* We don't want any race condition at early drop stage */
836 atomic_inc(&net
->ct
.count
);
838 if (nf_conntrack_max
&&
839 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
840 if (!early_drop(net
, hash
)) {
841 atomic_dec(&net
->ct
.count
);
842 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
843 return ERR_PTR(-ENOMEM
);
848 * Do not use kmem_cache_zalloc(), as this cache uses
849 * SLAB_DESTROY_BY_RCU.
851 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
853 atomic_dec(&net
->ct
.count
);
854 return ERR_PTR(-ENOMEM
);
857 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
858 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
860 memset(&ct
->tuplehash
[IP_CT_DIR_MAX
], 0,
861 offsetof(struct nf_conn
, proto
) -
862 offsetof(struct nf_conn
, tuplehash
[IP_CT_DIR_MAX
]));
863 spin_lock_init(&ct
->lock
);
864 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
865 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
866 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
867 /* save hash for reusing when confirming */
868 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
869 /* Don't set timer yet: wait for confirmation */
870 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
871 write_pnet(&ct
->ct_net
, net
);
872 #ifdef CONFIG_NF_CONNTRACK_ZONES
874 struct nf_conntrack_zone
*nf_ct_zone
;
876 nf_ct_zone
= nf_ct_ext_add(ct
, NF_CT_EXT_ZONE
, GFP_ATOMIC
);
879 nf_ct_zone
->id
= zone
;
882 /* Because we use RCU lookups, we set ct_general.use to zero before
883 * this is inserted in any list.
885 atomic_set(&ct
->ct_general
.use
, 0);
888 #ifdef CONFIG_NF_CONNTRACK_ZONES
890 atomic_dec(&net
->ct
.count
);
891 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
892 return ERR_PTR(-ENOMEM
);
896 struct nf_conn
*nf_conntrack_alloc(struct net
*net
, u16 zone
,
897 const struct nf_conntrack_tuple
*orig
,
898 const struct nf_conntrack_tuple
*repl
,
901 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
903 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
905 void nf_conntrack_free(struct nf_conn
*ct
)
907 struct net
*net
= nf_ct_net(ct
);
909 /* A freed object has refcnt == 0, that's
910 * the golden rule for SLAB_DESTROY_BY_RCU
912 NF_CT_ASSERT(atomic_read(&ct
->ct_general
.use
) == 0);
914 nf_ct_ext_destroy(ct
);
916 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
917 smp_mb__before_atomic();
918 atomic_dec(&net
->ct
.count
);
920 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
923 /* Allocate a new conntrack: we return -ENOMEM if classification
924 failed due to stress. Otherwise it really is unclassifiable. */
925 static struct nf_conntrack_tuple_hash
*
926 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
927 const struct nf_conntrack_tuple
*tuple
,
928 struct nf_conntrack_l3proto
*l3proto
,
929 struct nf_conntrack_l4proto
*l4proto
,
931 unsigned int dataoff
, u32 hash
)
934 struct nf_conn_help
*help
;
935 struct nf_conntrack_tuple repl_tuple
;
936 struct nf_conntrack_ecache
*ecache
;
937 struct nf_conntrack_expect
*exp
= NULL
;
938 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
939 struct nf_conn_timeout
*timeout_ext
;
940 unsigned int *timeouts
;
942 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
943 pr_debug("Can't invert tuple.\n");
947 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
950 return (struct nf_conntrack_tuple_hash
*)ct
;
952 if (tmpl
&& nfct_synproxy(tmpl
)) {
953 nfct_seqadj_ext_add(ct
);
954 nfct_synproxy_ext_add(ct
);
957 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
959 timeouts
= NF_CT_TIMEOUT_EXT_DATA(timeout_ext
);
961 timeouts
= l4proto
->get_timeouts(net
);
963 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
964 nf_conntrack_free(ct
);
965 pr_debug("init conntrack: can't track with proto module\n");
970 nf_ct_timeout_ext_add(ct
, timeout_ext
->timeout
, GFP_ATOMIC
);
972 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
973 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
974 nf_ct_labels_ext_add(ct
);
976 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
977 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
978 ecache
? ecache
->expmask
: 0,
982 if (net
->ct
.expect_count
) {
983 spin_lock(&nf_conntrack_expect_lock
);
984 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
986 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
988 /* Welcome, Mr. Bond. We've been expecting you... */
989 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
990 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
991 ct
->master
= exp
->master
;
993 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
996 rcu_assign_pointer(help
->helper
, exp
->helper
);
999 #ifdef CONFIG_NF_CONNTRACK_MARK
1000 ct
->mark
= exp
->master
->mark
;
1002 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1003 ct
->secmark
= exp
->master
->secmark
;
1005 NF_CT_STAT_INC(net
, expect_new
);
1007 spin_unlock(&nf_conntrack_expect_lock
);
1010 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1011 NF_CT_STAT_INC(net
, new);
1014 /* Now it is inserted into the unconfirmed list, bump refcount */
1015 nf_conntrack_get(&ct
->ct_general
);
1016 nf_ct_add_to_unconfirmed_list(ct
);
1022 exp
->expectfn(ct
, exp
);
1023 nf_ct_expect_put(exp
);
1026 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1029 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1030 static inline struct nf_conn
*
1031 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1032 struct sk_buff
*skb
,
1033 unsigned int dataoff
,
1036 struct nf_conntrack_l3proto
*l3proto
,
1037 struct nf_conntrack_l4proto
*l4proto
,
1039 enum ip_conntrack_info
*ctinfo
)
1041 struct nf_conntrack_tuple tuple
;
1042 struct nf_conntrack_tuple_hash
*h
;
1044 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
1047 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1048 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
1050 pr_debug("resolve_normal_ct: Can't get tuple\n");
1054 /* look for tuple match */
1055 hash
= hash_conntrack_raw(&tuple
, zone
);
1056 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1058 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1059 skb
, dataoff
, hash
);
1065 ct
= nf_ct_tuplehash_to_ctrack(h
);
1067 /* It exists; we have (non-exclusive) reference. */
1068 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1069 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1070 /* Please set reply bit if this packet OK */
1073 /* Once we've had two way comms, always ESTABLISHED. */
1074 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1075 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
1076 *ctinfo
= IP_CT_ESTABLISHED
;
1077 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1078 pr_debug("nf_conntrack_in: related packet for %p\n",
1080 *ctinfo
= IP_CT_RELATED
;
1082 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
1083 *ctinfo
= IP_CT_NEW
;
1087 skb
->nfct
= &ct
->ct_general
;
1088 skb
->nfctinfo
= *ctinfo
;
1093 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1094 struct sk_buff
*skb
)
1096 struct nf_conn
*ct
, *tmpl
= NULL
;
1097 enum ip_conntrack_info ctinfo
;
1098 struct nf_conntrack_l3proto
*l3proto
;
1099 struct nf_conntrack_l4proto
*l4proto
;
1100 unsigned int *timeouts
;
1101 unsigned int dataoff
;
1107 /* Previously seen (loopback or untracked)? Ignore. */
1108 tmpl
= (struct nf_conn
*)skb
->nfct
;
1109 if (!nf_ct_is_template(tmpl
)) {
1110 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1116 /* rcu_read_lock()ed by nf_hook_slow */
1117 l3proto
= __nf_ct_l3proto_find(pf
);
1118 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1119 &dataoff
, &protonum
);
1121 pr_debug("not prepared to track yet or error occurred\n");
1122 NF_CT_STAT_INC_ATOMIC(net
, error
);
1123 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1128 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1130 /* It may be an special packet, error, unclean...
1131 * inverse of the return code tells to the netfilter
1132 * core what to do with the packet. */
1133 if (l4proto
->error
!= NULL
) {
1134 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
1137 NF_CT_STAT_INC_ATOMIC(net
, error
);
1138 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1142 /* ICMP[v6] protocol trackers may assign one conntrack. */
1147 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1148 l3proto
, l4proto
, &set_reply
, &ctinfo
);
1150 /* Not valid part of a connection */
1151 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1157 /* Too stressed to deal. */
1158 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1163 NF_CT_ASSERT(skb
->nfct
);
1165 /* Decide what timeout policy we want to apply to this flow. */
1166 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1168 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1170 /* Invalid: inverse of the return code tells
1171 * the netfilter core what to do */
1172 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1173 nf_conntrack_put(skb
->nfct
);
1175 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1176 if (ret
== -NF_DROP
)
1177 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1182 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1183 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1186 /* Special case: we have to repeat this hook, assign the
1187 * template again to this packet. We assume that this packet
1188 * has no conntrack assigned. This is used by nf_ct_tcp. */
1189 if (ret
== NF_REPEAT
)
1190 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
1197 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1199 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1200 const struct nf_conntrack_tuple
*orig
)
1205 ret
= nf_ct_invert_tuple(inverse
, orig
,
1206 __nf_ct_l3proto_find(orig
->src
.l3num
),
1207 __nf_ct_l4proto_find(orig
->src
.l3num
,
1208 orig
->dst
.protonum
));
1212 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1214 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1215 implicitly racy: see __nf_conntrack_confirm */
1216 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1217 const struct nf_conntrack_tuple
*newreply
)
1219 struct nf_conn_help
*help
= nfct_help(ct
);
1221 /* Should be unconfirmed, so not in hash table yet */
1222 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1224 pr_debug("Altering reply tuple of %p to ", ct
);
1225 nf_ct_dump_tuple(newreply
);
1227 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1228 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1232 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1235 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1237 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1238 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1239 enum ip_conntrack_info ctinfo
,
1240 const struct sk_buff
*skb
,
1241 unsigned long extra_jiffies
,
1244 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1247 /* Only update if this is not a fixed timeout */
1248 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1251 /* If not in hash table, timer will not be active yet */
1252 if (!nf_ct_is_confirmed(ct
)) {
1253 ct
->timeout
.expires
= extra_jiffies
;
1255 unsigned long newtime
= jiffies
+ extra_jiffies
;
1257 /* Only update the timeout if the new timeout is at least
1258 HZ jiffies from the old timeout. Need del_timer for race
1259 avoidance (may already be dying). */
1260 if (newtime
- ct
->timeout
.expires
>= HZ
)
1261 mod_timer_pending(&ct
->timeout
, newtime
);
1266 struct nf_conn_acct
*acct
;
1268 acct
= nf_conn_acct_find(ct
);
1270 struct nf_conn_counter
*counter
= acct
->counter
;
1272 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1273 atomic64_add(skb
->len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1277 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1279 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1280 enum ip_conntrack_info ctinfo
,
1281 const struct sk_buff
*skb
,
1285 struct nf_conn_acct
*acct
;
1287 acct
= nf_conn_acct_find(ct
);
1289 struct nf_conn_counter
*counter
= acct
->counter
;
1291 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1292 atomic64_add(skb
->len
- skb_network_offset(skb
),
1293 &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1297 if (del_timer(&ct
->timeout
)) {
1298 ct
->timeout
.function((unsigned long)ct
);
1303 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1305 #ifdef CONFIG_NF_CONNTRACK_ZONES
1306 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1307 .len
= sizeof(struct nf_conntrack_zone
),
1308 .align
= __alignof__(struct nf_conntrack_zone
),
1309 .id
= NF_CT_EXT_ZONE
,
1313 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1315 #include <linux/netfilter/nfnetlink.h>
1316 #include <linux/netfilter/nfnetlink_conntrack.h>
1317 #include <linux/mutex.h>
1319 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1320 * in ip_conntrack_core, since we don't want the protocols to autoload
1321 * or depend on ctnetlink */
1322 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1323 const struct nf_conntrack_tuple
*tuple
)
1325 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1326 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1327 goto nla_put_failure
;
1333 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1335 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1336 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1337 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1339 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1341 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1342 struct nf_conntrack_tuple
*t
)
1344 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1347 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1348 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1352 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1354 int nf_ct_port_nlattr_tuple_size(void)
1356 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1358 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1361 /* Used by ipt_REJECT and ip6t_REJECT. */
1362 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1365 enum ip_conntrack_info ctinfo
;
1367 /* This ICMP is in reverse direction to the packet which caused it */
1368 ct
= nf_ct_get(skb
, &ctinfo
);
1369 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1370 ctinfo
= IP_CT_RELATED_REPLY
;
1372 ctinfo
= IP_CT_RELATED
;
1374 /* Attach to new skbuff, and increment count */
1375 nskb
->nfct
= &ct
->ct_general
;
1376 nskb
->nfctinfo
= ctinfo
;
1377 nf_conntrack_get(nskb
->nfct
);
1380 /* Bring out ya dead! */
1381 static struct nf_conn
*
1382 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1383 void *data
, unsigned int *bucket
)
1385 struct nf_conntrack_tuple_hash
*h
;
1387 struct hlist_nulls_node
*n
;
1391 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1392 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1395 if (*bucket
< net
->ct
.htable_size
) {
1396 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1397 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1399 ct
= nf_ct_tuplehash_to_ctrack(h
);
1408 for_each_possible_cpu(cpu
) {
1409 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1411 spin_lock_bh(&pcpu
->lock
);
1412 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1413 ct
= nf_ct_tuplehash_to_ctrack(h
);
1415 set_bit(IPS_DYING_BIT
, &ct
->status
);
1417 spin_unlock_bh(&pcpu
->lock
);
1421 atomic_inc(&ct
->ct_general
.use
);
1427 void nf_ct_iterate_cleanup(struct net
*net
,
1428 int (*iter
)(struct nf_conn
*i
, void *data
),
1429 void *data
, u32 portid
, int report
)
1432 unsigned int bucket
= 0;
1434 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1435 /* Time to push up daises... */
1436 if (del_timer(&ct
->timeout
))
1437 nf_ct_delete(ct
, portid
, report
);
1439 /* ... else the timer will get him soon. */
1444 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1446 static int kill_all(struct nf_conn
*i
, void *data
)
1451 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1453 if (is_vmalloc_addr(hash
))
1456 free_pages((unsigned long)hash
,
1457 get_order(sizeof(struct hlist_head
) * size
));
1459 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1461 void nf_conntrack_flush_report(struct net
*net
, u32 portid
, int report
)
1463 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, portid
, report
);
1465 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report
);
1467 static void nf_ct_release_dying_list(struct net
*net
)
1469 struct nf_conntrack_tuple_hash
*h
;
1471 struct hlist_nulls_node
*n
;
1474 for_each_possible_cpu(cpu
) {
1475 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1477 spin_lock_bh(&pcpu
->lock
);
1478 hlist_nulls_for_each_entry(h
, n
, &pcpu
->dying
, hnnode
) {
1479 ct
= nf_ct_tuplehash_to_ctrack(h
);
1480 /* never fails to remove them, no listeners at this point */
1483 spin_unlock_bh(&pcpu
->lock
);
1487 static int untrack_refs(void)
1491 for_each_possible_cpu(cpu
) {
1492 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1494 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1499 void nf_conntrack_cleanup_start(void)
1501 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1504 void nf_conntrack_cleanup_end(void)
1506 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1507 while (untrack_refs() > 0)
1510 #ifdef CONFIG_NF_CONNTRACK_ZONES
1511 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1513 nf_conntrack_proto_fini();
1514 nf_conntrack_seqadj_fini();
1515 nf_conntrack_labels_fini();
1516 nf_conntrack_helper_fini();
1517 nf_conntrack_timeout_fini();
1518 nf_conntrack_ecache_fini();
1519 nf_conntrack_tstamp_fini();
1520 nf_conntrack_acct_fini();
1521 nf_conntrack_expect_fini();
1525 * Mishearing the voices in his head, our hero wonders how he's
1526 * supposed to kill the mall.
1528 void nf_conntrack_cleanup_net(struct net
*net
)
1532 list_add(&net
->exit_list
, &single
);
1533 nf_conntrack_cleanup_net_list(&single
);
1536 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1542 * This makes sure all current packets have passed through
1543 * netfilter framework. Roll on, two-stage module
1549 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1550 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1551 nf_ct_release_dying_list(net
);
1552 if (atomic_read(&net
->ct
.count
) != 0)
1557 goto i_see_dead_people
;
1560 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1561 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1562 nf_conntrack_proto_pernet_fini(net
);
1563 nf_conntrack_helper_pernet_fini(net
);
1564 nf_conntrack_ecache_pernet_fini(net
);
1565 nf_conntrack_tstamp_pernet_fini(net
);
1566 nf_conntrack_acct_pernet_fini(net
);
1567 nf_conntrack_expect_pernet_fini(net
);
1568 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1569 kfree(net
->ct
.slabname
);
1570 free_percpu(net
->ct
.stat
);
1571 free_percpu(net
->ct
.pcpu_lists
);
1575 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1577 struct hlist_nulls_head
*hash
;
1578 unsigned int nr_slots
, i
;
1581 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1582 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1583 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1584 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1587 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1592 for (i
= 0; i
< nr_slots
; i
++)
1593 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1597 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1599 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1602 unsigned int hashsize
, old_size
;
1603 struct hlist_nulls_head
*hash
, *old_hash
;
1604 struct nf_conntrack_tuple_hash
*h
;
1607 if (current
->nsproxy
->net_ns
!= &init_net
)
1610 /* On boot, we can set this without any fancy locking. */
1611 if (!nf_conntrack_htable_size
)
1612 return param_set_uint(val
, kp
);
1614 rc
= kstrtouint(val
, 0, &hashsize
);
1620 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1625 nf_conntrack_all_lock();
1626 write_seqcount_begin(&init_net
.ct
.generation
);
1628 /* Lookups in the old hash might happen in parallel, which means we
1629 * might get false negatives during connection lookup. New connections
1630 * created because of a false negative won't make it into the hash
1631 * though since that required taking the locks.
1634 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1635 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1636 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1637 struct nf_conntrack_tuple_hash
, hnnode
);
1638 ct
= nf_ct_tuplehash_to_ctrack(h
);
1639 hlist_nulls_del_rcu(&h
->hnnode
);
1640 bucket
= __hash_conntrack(&h
->tuple
, nf_ct_zone(ct
),
1642 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1645 old_size
= init_net
.ct
.htable_size
;
1646 old_hash
= init_net
.ct
.hash
;
1648 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1649 init_net
.ct
.hash
= hash
;
1651 write_seqcount_end(&init_net
.ct
.generation
);
1652 nf_conntrack_all_unlock();
1655 nf_ct_free_hashtable(old_hash
, old_size
);
1658 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1660 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1661 &nf_conntrack_htable_size
, 0600);
1663 void nf_ct_untracked_status_or(unsigned long bits
)
1667 for_each_possible_cpu(cpu
)
1668 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1670 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1672 int nf_conntrack_init_start(void)
1677 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
1678 spin_lock_init(&nf_conntrack_locks
[i
]);
1680 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1681 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1682 if (!nf_conntrack_htable_size
) {
1683 nf_conntrack_htable_size
1684 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1685 / sizeof(struct hlist_head
));
1686 if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1687 nf_conntrack_htable_size
= 16384;
1688 if (nf_conntrack_htable_size
< 32)
1689 nf_conntrack_htable_size
= 32;
1691 /* Use a max. factor of four by default to get the same max as
1692 * with the old struct list_heads. When a table size is given
1693 * we use the old value of 8 to avoid reducing the max.
1697 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1699 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1700 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1703 ret
= nf_conntrack_expect_init();
1707 ret
= nf_conntrack_acct_init();
1711 ret
= nf_conntrack_tstamp_init();
1715 ret
= nf_conntrack_ecache_init();
1719 ret
= nf_conntrack_timeout_init();
1723 ret
= nf_conntrack_helper_init();
1727 ret
= nf_conntrack_labels_init();
1731 ret
= nf_conntrack_seqadj_init();
1735 #ifdef CONFIG_NF_CONNTRACK_ZONES
1736 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1740 ret
= nf_conntrack_proto_init();
1744 /* Set up fake conntrack: to never be deleted, not in any hashes */
1745 for_each_possible_cpu(cpu
) {
1746 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1747 write_pnet(&ct
->ct_net
, &init_net
);
1748 atomic_set(&ct
->ct_general
.use
, 1);
1750 /* - and look it like as a confirmed connection */
1751 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1755 #ifdef CONFIG_NF_CONNTRACK_ZONES
1756 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1759 nf_conntrack_seqadj_fini();
1761 nf_conntrack_labels_fini();
1763 nf_conntrack_helper_fini();
1765 nf_conntrack_timeout_fini();
1767 nf_conntrack_ecache_fini();
1769 nf_conntrack_tstamp_fini();
1771 nf_conntrack_acct_fini();
1773 nf_conntrack_expect_fini();
1778 void nf_conntrack_init_end(void)
1780 /* For use by REJECT target */
1781 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1782 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1786 * We need to use special "null" values, not used in hash table
1788 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1789 #define DYING_NULLS_VAL ((1<<30)+1)
1790 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1792 int nf_conntrack_init_net(struct net
*net
)
1797 atomic_set(&net
->ct
.count
, 0);
1798 seqcount_init(&net
->ct
.generation
);
1800 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
1801 if (!net
->ct
.pcpu_lists
)
1804 for_each_possible_cpu(cpu
) {
1805 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1807 spin_lock_init(&pcpu
->lock
);
1808 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1809 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
1810 INIT_HLIST_NULLS_HEAD(&pcpu
->tmpl
, TEMPLATE_NULLS_VAL
);
1813 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1815 goto err_pcpu_lists
;
1817 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1818 if (!net
->ct
.slabname
)
1821 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1822 sizeof(struct nf_conn
), 0,
1823 SLAB_DESTROY_BY_RCU
, NULL
);
1824 if (!net
->ct
.nf_conntrack_cachep
) {
1825 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1829 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1830 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
, 1);
1831 if (!net
->ct
.hash
) {
1832 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1835 ret
= nf_conntrack_expect_pernet_init(net
);
1838 ret
= nf_conntrack_acct_pernet_init(net
);
1841 ret
= nf_conntrack_tstamp_pernet_init(net
);
1844 ret
= nf_conntrack_ecache_pernet_init(net
);
1847 ret
= nf_conntrack_helper_pernet_init(net
);
1850 ret
= nf_conntrack_proto_pernet_init(net
);
1856 nf_conntrack_helper_pernet_fini(net
);
1858 nf_conntrack_ecache_pernet_fini(net
);
1860 nf_conntrack_tstamp_pernet_fini(net
);
1862 nf_conntrack_acct_pernet_fini(net
);
1864 nf_conntrack_expect_pernet_fini(net
);
1866 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1868 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1870 kfree(net
->ct
.slabname
);
1872 free_percpu(net
->ct
.stat
);
1874 free_percpu(net
->ct
.pcpu_lists
);