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
)
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
, 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
,
156 return __hash_bucket(hash_conntrack_raw(tuple
), size
);
159 static inline u_int32_t
hash_conntrack(const struct net
*net
,
160 const struct nf_conntrack_tuple
*tuple
)
162 return __hash_conntrack(tuple
, net
->ct
.htable_size
);
166 nf_ct_get_tuple(const struct sk_buff
*skb
,
168 unsigned int dataoff
,
172 struct nf_conntrack_tuple
*tuple
,
173 const struct nf_conntrack_l3proto
*l3proto
,
174 const struct nf_conntrack_l4proto
*l4proto
)
176 memset(tuple
, 0, sizeof(*tuple
));
178 tuple
->src
.l3num
= l3num
;
179 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
182 tuple
->dst
.protonum
= protonum
;
183 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
185 return l4proto
->pkt_to_tuple(skb
, dataoff
, net
, tuple
);
187 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
189 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
191 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
193 struct nf_conntrack_l3proto
*l3proto
;
194 struct nf_conntrack_l4proto
*l4proto
;
195 unsigned int protoff
;
201 l3proto
= __nf_ct_l3proto_find(l3num
);
202 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
203 if (ret
!= NF_ACCEPT
) {
208 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
210 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
,
216 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
219 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
220 const struct nf_conntrack_tuple
*orig
,
221 const struct nf_conntrack_l3proto
*l3proto
,
222 const struct nf_conntrack_l4proto
*l4proto
)
224 memset(inverse
, 0, sizeof(*inverse
));
226 inverse
->src
.l3num
= orig
->src
.l3num
;
227 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
230 inverse
->dst
.dir
= !orig
->dst
.dir
;
232 inverse
->dst
.protonum
= orig
->dst
.protonum
;
233 return l4proto
->invert_tuple(inverse
, orig
);
235 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
238 clean_from_lists(struct nf_conn
*ct
)
240 pr_debug("clean_from_lists(%p)\n", ct
);
241 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
242 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
244 /* Destroy all pending expectations */
245 nf_ct_remove_expectations(ct
);
248 /* must be called with local_bh_disable */
249 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
251 struct ct_pcpu
*pcpu
;
253 /* add this conntrack to the (per cpu) dying list */
254 ct
->cpu
= smp_processor_id();
255 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
257 spin_lock(&pcpu
->lock
);
258 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
260 spin_unlock(&pcpu
->lock
);
263 /* must be called with local_bh_disable */
264 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
266 struct ct_pcpu
*pcpu
;
268 /* add this conntrack to the (per cpu) unconfirmed list */
269 ct
->cpu
= smp_processor_id();
270 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
272 spin_lock(&pcpu
->lock
);
273 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
275 spin_unlock(&pcpu
->lock
);
278 /* must be called with local_bh_disable */
279 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
281 struct ct_pcpu
*pcpu
;
283 /* We overload first tuple to link into unconfirmed or dying list.*/
284 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
286 spin_lock(&pcpu
->lock
);
287 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
288 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
289 spin_unlock(&pcpu
->lock
);
292 /* Released via destroy_conntrack() */
293 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
294 const struct nf_conntrack_zone
*zone
,
297 struct nf_conn
*tmpl
;
299 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
303 tmpl
->status
= IPS_TEMPLATE
;
304 write_pnet(&tmpl
->ct_net
, net
);
306 if (nf_ct_zone_add(tmpl
, flags
, zone
) < 0)
309 atomic_set(&tmpl
->ct_general
.use
, 0);
316 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
318 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
320 nf_ct_ext_destroy(tmpl
);
321 nf_ct_ext_free(tmpl
);
324 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
327 destroy_conntrack(struct nf_conntrack
*nfct
)
329 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
330 struct net
*net
= nf_ct_net(ct
);
331 struct nf_conntrack_l4proto
*l4proto
;
333 pr_debug("destroy_conntrack(%p)\n", ct
);
334 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
335 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
337 if (unlikely(nf_ct_is_template(ct
))) {
342 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
343 if (l4proto
&& l4proto
->destroy
)
344 l4proto
->destroy(ct
);
349 /* Expectations will have been removed in clean_from_lists,
350 * except TFTP can create an expectation on the first packet,
351 * before connection is in the list, so we need to clean here,
354 nf_ct_remove_expectations(ct
);
356 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
358 NF_CT_STAT_INC(net
, delete);
362 nf_ct_put(ct
->master
);
364 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
365 nf_conntrack_free(ct
);
368 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
370 struct net
*net
= nf_ct_net(ct
);
371 unsigned int hash
, reply_hash
;
372 unsigned int sequence
;
374 nf_ct_helper_destroy(ct
);
378 sequence
= read_seqcount_begin(&net
->ct
.generation
);
379 hash
= hash_conntrack(net
,
380 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
381 reply_hash
= hash_conntrack(net
,
382 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
383 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
385 clean_from_lists(ct
);
386 nf_conntrack_double_unlock(hash
, reply_hash
);
388 nf_ct_add_to_dying_list(ct
);
390 NF_CT_STAT_INC(net
, delete_list
);
394 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
396 struct nf_conn_tstamp
*tstamp
;
398 tstamp
= nf_conn_tstamp_find(ct
);
399 if (tstamp
&& tstamp
->stop
== 0)
400 tstamp
->stop
= ktime_get_real_ns();
402 if (nf_ct_is_dying(ct
))
405 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
406 portid
, report
) < 0) {
407 /* destroy event was not delivered */
408 nf_ct_delete_from_lists(ct
);
409 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
413 nf_conntrack_ecache_work(nf_ct_net(ct
));
414 set_bit(IPS_DYING_BIT
, &ct
->status
);
416 nf_ct_delete_from_lists(ct
);
420 EXPORT_SYMBOL_GPL(nf_ct_delete
);
422 static void death_by_timeout(unsigned long ul_conntrack
)
424 nf_ct_delete((struct nf_conn
*)ul_conntrack
, 0, 0);
428 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
429 const struct nf_conntrack_tuple
*tuple
,
430 const struct nf_conntrack_zone
*zone
)
432 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
434 /* A conntrack can be recreated with the equal tuple,
435 * so we need to check that the conntrack is confirmed
437 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
438 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
439 nf_ct_is_confirmed(ct
);
444 * - Caller must take a reference on returned object
445 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
447 static struct nf_conntrack_tuple_hash
*
448 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
449 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
451 struct nf_conntrack_tuple_hash
*h
;
452 struct hlist_nulls_node
*n
;
453 unsigned int bucket
= hash_bucket(hash
, net
);
455 /* Disable BHs the entire time since we normally need to disable them
456 * at least once for the stats anyway.
460 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[bucket
], hnnode
) {
461 if (nf_ct_key_equal(h
, tuple
, zone
)) {
462 NF_CT_STAT_INC(net
, found
);
466 NF_CT_STAT_INC(net
, searched
);
469 * if the nulls value we got at the end of this lookup is
470 * not the expected one, we must restart lookup.
471 * We probably met an item that was moved to another chain.
473 if (get_nulls_value(n
) != bucket
) {
474 NF_CT_STAT_INC(net
, search_restart
);
482 /* Find a connection corresponding to a tuple. */
483 static struct nf_conntrack_tuple_hash
*
484 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
485 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
487 struct nf_conntrack_tuple_hash
*h
;
492 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
494 ct
= nf_ct_tuplehash_to_ctrack(h
);
495 if (unlikely(nf_ct_is_dying(ct
) ||
496 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
499 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
))) {
510 struct nf_conntrack_tuple_hash
*
511 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
512 const struct nf_conntrack_tuple
*tuple
)
514 return __nf_conntrack_find_get(net
, zone
, tuple
,
515 hash_conntrack_raw(tuple
));
517 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
519 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
521 unsigned int reply_hash
)
523 struct net
*net
= nf_ct_net(ct
);
525 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
526 &net
->ct
.hash
[hash
]);
527 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
528 &net
->ct
.hash
[reply_hash
]);
532 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
534 const struct nf_conntrack_zone
*zone
;
535 struct net
*net
= nf_ct_net(ct
);
536 unsigned int hash
, reply_hash
;
537 struct nf_conntrack_tuple_hash
*h
;
538 struct hlist_nulls_node
*n
;
539 unsigned int sequence
;
541 zone
= nf_ct_zone(ct
);
545 sequence
= read_seqcount_begin(&net
->ct
.generation
);
546 hash
= hash_conntrack(net
,
547 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
548 reply_hash
= hash_conntrack(net
,
549 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
550 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
552 /* See if there's one in the list already, including reverse */
553 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
554 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
556 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
559 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
560 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
562 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
566 add_timer(&ct
->timeout
);
568 /* The caller holds a reference to this object */
569 atomic_set(&ct
->ct_general
.use
, 2);
570 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
571 nf_conntrack_double_unlock(hash
, reply_hash
);
572 NF_CT_STAT_INC(net
, insert
);
577 nf_conntrack_double_unlock(hash
, reply_hash
);
578 NF_CT_STAT_INC(net
, insert_failed
);
582 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
584 /* Confirm a connection given skb; places it in hash table */
586 __nf_conntrack_confirm(struct sk_buff
*skb
)
588 const struct nf_conntrack_zone
*zone
;
589 unsigned int hash
, reply_hash
;
590 struct nf_conntrack_tuple_hash
*h
;
592 struct nf_conn_help
*help
;
593 struct nf_conn_tstamp
*tstamp
;
594 struct hlist_nulls_node
*n
;
595 enum ip_conntrack_info ctinfo
;
597 unsigned int sequence
;
599 ct
= nf_ct_get(skb
, &ctinfo
);
602 /* ipt_REJECT uses nf_conntrack_attach to attach related
603 ICMP/TCP RST packets in other direction. Actual packet
604 which created connection will be IP_CT_NEW or for an
605 expected connection, IP_CT_RELATED. */
606 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
609 zone
= nf_ct_zone(ct
);
613 sequence
= read_seqcount_begin(&net
->ct
.generation
);
614 /* reuse the hash saved before */
615 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
616 hash
= hash_bucket(hash
, net
);
617 reply_hash
= hash_conntrack(net
,
618 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
620 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
622 /* We're not in hash table, and we refuse to set up related
623 * connections for unconfirmed conns. But packet copies and
624 * REJECT will give spurious warnings here.
626 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
628 /* No external references means no one else could have
631 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
632 pr_debug("Confirming conntrack %p\n", ct
);
633 /* We have to check the DYING flag after unlink to prevent
634 * a race against nf_ct_get_next_corpse() possibly called from
635 * user context, else we insert an already 'dead' hash, blocking
636 * further use of that particular connection -JM.
638 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
640 if (unlikely(nf_ct_is_dying(ct
)))
643 /* See if there's one in the list already, including reverse:
644 NAT could have grabbed it without realizing, since we're
645 not in the hash. If there is, we lost race. */
646 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
647 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
649 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
652 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
653 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
655 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
659 /* Timer relative to confirmation time, not original
660 setting time, otherwise we'd get timer wrap in
661 weird delay cases. */
662 ct
->timeout
.expires
+= jiffies
;
663 add_timer(&ct
->timeout
);
664 atomic_inc(&ct
->ct_general
.use
);
665 ct
->status
|= IPS_CONFIRMED
;
667 /* set conntrack timestamp, if enabled. */
668 tstamp
= nf_conn_tstamp_find(ct
);
670 if (skb
->tstamp
.tv64
== 0)
671 __net_timestamp(skb
);
673 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
675 /* Since the lookup is lockless, hash insertion must be done after
676 * starting the timer and setting the CONFIRMED bit. The RCU barriers
677 * guarantee that no other CPU can find the conntrack before the above
678 * stores are visible.
680 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
681 nf_conntrack_double_unlock(hash
, reply_hash
);
682 NF_CT_STAT_INC(net
, insert
);
685 help
= nfct_help(ct
);
686 if (help
&& help
->helper
)
687 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
689 nf_conntrack_event_cache(master_ct(ct
) ?
690 IPCT_RELATED
: IPCT_NEW
, ct
);
694 nf_ct_add_to_dying_list(ct
);
695 nf_conntrack_double_unlock(hash
, reply_hash
);
696 NF_CT_STAT_INC(net
, insert_failed
);
700 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
702 /* Returns true if a connection correspondings to the tuple (required
705 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
706 const struct nf_conn
*ignored_conntrack
)
708 struct net
*net
= nf_ct_net(ignored_conntrack
);
709 const struct nf_conntrack_zone
*zone
;
710 struct nf_conntrack_tuple_hash
*h
;
711 struct hlist_nulls_node
*n
;
715 zone
= nf_ct_zone(ignored_conntrack
);
716 hash
= hash_conntrack(net
, tuple
);
718 /* Disable BHs the entire time since we need to disable them at
719 * least once for the stats anyway.
722 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
723 ct
= nf_ct_tuplehash_to_ctrack(h
);
724 if (ct
!= ignored_conntrack
&&
725 nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
726 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
))) {
727 NF_CT_STAT_INC(net
, found
);
728 rcu_read_unlock_bh();
731 NF_CT_STAT_INC(net
, searched
);
733 rcu_read_unlock_bh();
737 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
739 #define NF_CT_EVICTION_RANGE 8
741 /* There's a small race here where we may free a just-assured
742 connection. Too bad: we're in trouble anyway. */
743 static noinline
int early_drop(struct net
*net
, unsigned int _hash
)
745 /* Use oldest entry, which is roughly LRU */
746 struct nf_conntrack_tuple_hash
*h
;
747 struct nf_conn
*ct
= NULL
, *tmp
;
748 struct hlist_nulls_node
*n
;
749 unsigned int i
= 0, cnt
= 0;
751 unsigned int hash
, sequence
;
756 sequence
= read_seqcount_begin(&net
->ct
.generation
);
757 hash
= hash_bucket(_hash
, net
);
758 for (; i
< net
->ct
.htable_size
; i
++) {
759 lockp
= &nf_conntrack_locks
[hash
% CONNTRACK_LOCKS
];
761 if (read_seqcount_retry(&net
->ct
.generation
, sequence
)) {
765 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
767 tmp
= nf_ct_tuplehash_to_ctrack(h
);
768 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
) &&
769 !nf_ct_is_dying(tmp
) &&
770 atomic_inc_not_zero(&tmp
->ct_general
.use
)) {
777 hash
= (hash
+ 1) % net
->ct
.htable_size
;
780 if (ct
|| cnt
>= NF_CT_EVICTION_RANGE
)
789 if (del_timer(&ct
->timeout
)) {
790 if (nf_ct_delete(ct
, 0, 0)) {
792 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
799 void init_nf_conntrack_hash_rnd(void)
804 * Why not initialize nf_conntrack_rnd in a "init()" function ?
805 * Because there isn't enough entropy when system initializing,
806 * and we initialize it as late as possible.
809 get_random_bytes(&rand
, sizeof(rand
));
811 cmpxchg(&nf_conntrack_hash_rnd
, 0, rand
);
814 static struct nf_conn
*
815 __nf_conntrack_alloc(struct net
*net
,
816 const struct nf_conntrack_zone
*zone
,
817 const struct nf_conntrack_tuple
*orig
,
818 const struct nf_conntrack_tuple
*repl
,
823 if (unlikely(!nf_conntrack_hash_rnd
)) {
824 init_nf_conntrack_hash_rnd();
825 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
826 hash
= hash_conntrack_raw(orig
);
829 /* We don't want any race condition at early drop stage */
830 atomic_inc(&net
->ct
.count
);
832 if (nf_conntrack_max
&&
833 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
834 if (!early_drop(net
, hash
)) {
835 atomic_dec(&net
->ct
.count
);
836 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
837 return ERR_PTR(-ENOMEM
);
842 * Do not use kmem_cache_zalloc(), as this cache uses
843 * SLAB_DESTROY_BY_RCU.
845 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
849 spin_lock_init(&ct
->lock
);
850 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
851 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
852 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
853 /* save hash for reusing when confirming */
854 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
856 /* Don't set timer yet: wait for confirmation */
857 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
858 write_pnet(&ct
->ct_net
, net
);
859 memset(&ct
->__nfct_init_offset
[0], 0,
860 offsetof(struct nf_conn
, proto
) -
861 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
863 if (zone
&& nf_ct_zone_add(ct
, GFP_ATOMIC
, zone
) < 0)
866 /* Because we use RCU lookups, we set ct_general.use to zero before
867 * this is inserted in any list.
869 atomic_set(&ct
->ct_general
.use
, 0);
872 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
874 atomic_dec(&net
->ct
.count
);
875 return ERR_PTR(-ENOMEM
);
878 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
879 const struct nf_conntrack_zone
*zone
,
880 const struct nf_conntrack_tuple
*orig
,
881 const struct nf_conntrack_tuple
*repl
,
884 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
886 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
888 void nf_conntrack_free(struct nf_conn
*ct
)
890 struct net
*net
= nf_ct_net(ct
);
892 /* A freed object has refcnt == 0, that's
893 * the golden rule for SLAB_DESTROY_BY_RCU
895 NF_CT_ASSERT(atomic_read(&ct
->ct_general
.use
) == 0);
897 nf_ct_ext_destroy(ct
);
899 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
900 smp_mb__before_atomic();
901 atomic_dec(&net
->ct
.count
);
903 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
906 /* Allocate a new conntrack: we return -ENOMEM if classification
907 failed due to stress. Otherwise it really is unclassifiable. */
908 static struct nf_conntrack_tuple_hash
*
909 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
910 const struct nf_conntrack_tuple
*tuple
,
911 struct nf_conntrack_l3proto
*l3proto
,
912 struct nf_conntrack_l4proto
*l4proto
,
914 unsigned int dataoff
, u32 hash
)
917 struct nf_conn_help
*help
;
918 struct nf_conntrack_tuple repl_tuple
;
919 struct nf_conntrack_ecache
*ecache
;
920 struct nf_conntrack_expect
*exp
= NULL
;
921 const struct nf_conntrack_zone
*zone
;
922 struct nf_conn_timeout
*timeout_ext
;
923 struct nf_conntrack_zone tmp
;
924 unsigned int *timeouts
;
926 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
927 pr_debug("Can't invert tuple.\n");
931 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
932 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
935 return (struct nf_conntrack_tuple_hash
*)ct
;
937 if (tmpl
&& nfct_synproxy(tmpl
)) {
938 nfct_seqadj_ext_add(ct
);
939 nfct_synproxy_ext_add(ct
);
942 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
944 timeouts
= nf_ct_timeout_data(timeout_ext
);
945 if (unlikely(!timeouts
))
946 timeouts
= l4proto
->get_timeouts(net
);
948 timeouts
= l4proto
->get_timeouts(net
);
951 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
952 nf_conntrack_free(ct
);
953 pr_debug("init conntrack: can't track with proto module\n");
958 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
961 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
962 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
963 nf_ct_labels_ext_add(ct
);
965 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
966 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
967 ecache
? ecache
->expmask
: 0,
971 if (net
->ct
.expect_count
) {
972 spin_lock(&nf_conntrack_expect_lock
);
973 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
975 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
977 /* Welcome, Mr. Bond. We've been expecting you... */
978 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
979 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
980 ct
->master
= exp
->master
;
982 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
985 rcu_assign_pointer(help
->helper
, exp
->helper
);
988 #ifdef CONFIG_NF_CONNTRACK_MARK
989 ct
->mark
= exp
->master
->mark
;
991 #ifdef CONFIG_NF_CONNTRACK_SECMARK
992 ct
->secmark
= exp
->master
->secmark
;
994 NF_CT_STAT_INC(net
, expect_new
);
996 spin_unlock(&nf_conntrack_expect_lock
);
999 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1000 NF_CT_STAT_INC(net
, new);
1003 /* Now it is inserted into the unconfirmed list, bump refcount */
1004 nf_conntrack_get(&ct
->ct_general
);
1005 nf_ct_add_to_unconfirmed_list(ct
);
1011 exp
->expectfn(ct
, exp
);
1012 nf_ct_expect_put(exp
);
1015 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1018 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1019 static inline struct nf_conn
*
1020 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1021 struct sk_buff
*skb
,
1022 unsigned int dataoff
,
1025 struct nf_conntrack_l3proto
*l3proto
,
1026 struct nf_conntrack_l4proto
*l4proto
,
1028 enum ip_conntrack_info
*ctinfo
)
1030 const struct nf_conntrack_zone
*zone
;
1031 struct nf_conntrack_tuple tuple
;
1032 struct nf_conntrack_tuple_hash
*h
;
1033 struct nf_conntrack_zone tmp
;
1037 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1038 dataoff
, l3num
, protonum
, net
, &tuple
, l3proto
,
1040 pr_debug("resolve_normal_ct: Can't get tuple\n");
1044 /* look for tuple match */
1045 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1046 hash
= hash_conntrack_raw(&tuple
);
1047 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1049 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1050 skb
, dataoff
, hash
);
1056 ct
= nf_ct_tuplehash_to_ctrack(h
);
1058 /* It exists; we have (non-exclusive) reference. */
1059 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1060 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1061 /* Please set reply bit if this packet OK */
1064 /* Once we've had two way comms, always ESTABLISHED. */
1065 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1066 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
1067 *ctinfo
= IP_CT_ESTABLISHED
;
1068 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1069 pr_debug("nf_conntrack_in: related packet for %p\n",
1071 *ctinfo
= IP_CT_RELATED
;
1073 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
1074 *ctinfo
= IP_CT_NEW
;
1078 skb
->nfct
= &ct
->ct_general
;
1079 skb
->nfctinfo
= *ctinfo
;
1084 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1085 struct sk_buff
*skb
)
1087 struct nf_conn
*ct
, *tmpl
= NULL
;
1088 enum ip_conntrack_info ctinfo
;
1089 struct nf_conntrack_l3proto
*l3proto
;
1090 struct nf_conntrack_l4proto
*l4proto
;
1091 unsigned int *timeouts
;
1092 unsigned int dataoff
;
1098 /* Previously seen (loopback or untracked)? Ignore. */
1099 tmpl
= (struct nf_conn
*)skb
->nfct
;
1100 if (!nf_ct_is_template(tmpl
)) {
1101 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1107 /* rcu_read_lock()ed by nf_hook_slow */
1108 l3proto
= __nf_ct_l3proto_find(pf
);
1109 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1110 &dataoff
, &protonum
);
1112 pr_debug("not prepared to track yet or error occurred\n");
1113 NF_CT_STAT_INC_ATOMIC(net
, error
);
1114 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1119 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1121 /* It may be an special packet, error, unclean...
1122 * inverse of the return code tells to the netfilter
1123 * core what to do with the packet. */
1124 if (l4proto
->error
!= NULL
) {
1125 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
1128 NF_CT_STAT_INC_ATOMIC(net
, error
);
1129 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1133 /* ICMP[v6] protocol trackers may assign one conntrack. */
1138 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1139 l3proto
, l4proto
, &set_reply
, &ctinfo
);
1141 /* Not valid part of a connection */
1142 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1148 /* Too stressed to deal. */
1149 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1154 NF_CT_ASSERT(skb
->nfct
);
1156 /* Decide what timeout policy we want to apply to this flow. */
1157 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1159 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1161 /* Invalid: inverse of the return code tells
1162 * the netfilter core what to do */
1163 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1164 nf_conntrack_put(skb
->nfct
);
1166 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1167 if (ret
== -NF_DROP
)
1168 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1173 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1174 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1177 /* Special case: we have to repeat this hook, assign the
1178 * template again to this packet. We assume that this packet
1179 * has no conntrack assigned. This is used by nf_ct_tcp. */
1180 if (ret
== NF_REPEAT
)
1181 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
1188 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1190 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1191 const struct nf_conntrack_tuple
*orig
)
1196 ret
= nf_ct_invert_tuple(inverse
, orig
,
1197 __nf_ct_l3proto_find(orig
->src
.l3num
),
1198 __nf_ct_l4proto_find(orig
->src
.l3num
,
1199 orig
->dst
.protonum
));
1203 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1205 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1206 implicitly racy: see __nf_conntrack_confirm */
1207 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1208 const struct nf_conntrack_tuple
*newreply
)
1210 struct nf_conn_help
*help
= nfct_help(ct
);
1212 /* Should be unconfirmed, so not in hash table yet */
1213 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1215 pr_debug("Altering reply tuple of %p to ", ct
);
1216 nf_ct_dump_tuple(newreply
);
1218 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1219 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1223 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1226 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1228 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1229 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1230 enum ip_conntrack_info ctinfo
,
1231 const struct sk_buff
*skb
,
1232 unsigned long extra_jiffies
,
1235 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1238 /* Only update if this is not a fixed timeout */
1239 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1242 /* If not in hash table, timer will not be active yet */
1243 if (!nf_ct_is_confirmed(ct
)) {
1244 ct
->timeout
.expires
= extra_jiffies
;
1246 unsigned long newtime
= jiffies
+ extra_jiffies
;
1248 /* Only update the timeout if the new timeout is at least
1249 HZ jiffies from the old timeout. Need del_timer for race
1250 avoidance (may already be dying). */
1251 if (newtime
- ct
->timeout
.expires
>= HZ
)
1252 mod_timer_pending(&ct
->timeout
, newtime
);
1257 struct nf_conn_acct
*acct
;
1259 acct
= nf_conn_acct_find(ct
);
1261 struct nf_conn_counter
*counter
= acct
->counter
;
1263 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1264 atomic64_add(skb
->len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1268 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1270 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1271 enum ip_conntrack_info ctinfo
,
1272 const struct sk_buff
*skb
,
1276 struct nf_conn_acct
*acct
;
1278 acct
= nf_conn_acct_find(ct
);
1280 struct nf_conn_counter
*counter
= acct
->counter
;
1282 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1283 atomic64_add(skb
->len
- skb_network_offset(skb
),
1284 &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1288 if (del_timer(&ct
->timeout
)) {
1289 ct
->timeout
.function((unsigned long)ct
);
1294 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1296 #ifdef CONFIG_NF_CONNTRACK_ZONES
1297 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1298 .len
= sizeof(struct nf_conntrack_zone
),
1299 .align
= __alignof__(struct nf_conntrack_zone
),
1300 .id
= NF_CT_EXT_ZONE
,
1304 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1306 #include <linux/netfilter/nfnetlink.h>
1307 #include <linux/netfilter/nfnetlink_conntrack.h>
1308 #include <linux/mutex.h>
1310 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1311 * in ip_conntrack_core, since we don't want the protocols to autoload
1312 * or depend on ctnetlink */
1313 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1314 const struct nf_conntrack_tuple
*tuple
)
1316 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1317 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1318 goto nla_put_failure
;
1324 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1326 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1327 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1328 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1330 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1332 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1333 struct nf_conntrack_tuple
*t
)
1335 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1338 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1339 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1343 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1345 int nf_ct_port_nlattr_tuple_size(void)
1347 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1349 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1352 /* Used by ipt_REJECT and ip6t_REJECT. */
1353 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1356 enum ip_conntrack_info ctinfo
;
1358 /* This ICMP is in reverse direction to the packet which caused it */
1359 ct
= nf_ct_get(skb
, &ctinfo
);
1360 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1361 ctinfo
= IP_CT_RELATED_REPLY
;
1363 ctinfo
= IP_CT_RELATED
;
1365 /* Attach to new skbuff, and increment count */
1366 nskb
->nfct
= &ct
->ct_general
;
1367 nskb
->nfctinfo
= ctinfo
;
1368 nf_conntrack_get(nskb
->nfct
);
1371 /* Bring out ya dead! */
1372 static struct nf_conn
*
1373 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1374 void *data
, unsigned int *bucket
)
1376 struct nf_conntrack_tuple_hash
*h
;
1378 struct hlist_nulls_node
*n
;
1382 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1383 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1386 if (*bucket
< net
->ct
.htable_size
) {
1387 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1388 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1390 ct
= nf_ct_tuplehash_to_ctrack(h
);
1399 for_each_possible_cpu(cpu
) {
1400 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1402 spin_lock_bh(&pcpu
->lock
);
1403 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1404 ct
= nf_ct_tuplehash_to_ctrack(h
);
1406 set_bit(IPS_DYING_BIT
, &ct
->status
);
1408 spin_unlock_bh(&pcpu
->lock
);
1412 atomic_inc(&ct
->ct_general
.use
);
1418 void nf_ct_iterate_cleanup(struct net
*net
,
1419 int (*iter
)(struct nf_conn
*i
, void *data
),
1420 void *data
, u32 portid
, int report
)
1423 unsigned int bucket
= 0;
1425 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1426 /* Time to push up daises... */
1427 if (del_timer(&ct
->timeout
))
1428 nf_ct_delete(ct
, portid
, report
);
1430 /* ... else the timer will get him soon. */
1435 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1437 static int kill_all(struct nf_conn
*i
, void *data
)
1442 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1444 if (is_vmalloc_addr(hash
))
1447 free_pages((unsigned long)hash
,
1448 get_order(sizeof(struct hlist_head
) * size
));
1450 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1452 static int untrack_refs(void)
1456 for_each_possible_cpu(cpu
) {
1457 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1459 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1464 void nf_conntrack_cleanup_start(void)
1466 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1469 void nf_conntrack_cleanup_end(void)
1471 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1472 while (untrack_refs() > 0)
1475 #ifdef CONFIG_NF_CONNTRACK_ZONES
1476 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1478 nf_conntrack_proto_fini();
1479 nf_conntrack_seqadj_fini();
1480 nf_conntrack_labels_fini();
1481 nf_conntrack_helper_fini();
1482 nf_conntrack_timeout_fini();
1483 nf_conntrack_ecache_fini();
1484 nf_conntrack_tstamp_fini();
1485 nf_conntrack_acct_fini();
1486 nf_conntrack_expect_fini();
1490 * Mishearing the voices in his head, our hero wonders how he's
1491 * supposed to kill the mall.
1493 void nf_conntrack_cleanup_net(struct net
*net
)
1497 list_add(&net
->exit_list
, &single
);
1498 nf_conntrack_cleanup_net_list(&single
);
1501 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1507 * This makes sure all current packets have passed through
1508 * netfilter framework. Roll on, two-stage module
1514 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1515 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1516 if (atomic_read(&net
->ct
.count
) != 0)
1521 goto i_see_dead_people
;
1524 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1525 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1526 nf_conntrack_proto_pernet_fini(net
);
1527 nf_conntrack_helper_pernet_fini(net
);
1528 nf_conntrack_ecache_pernet_fini(net
);
1529 nf_conntrack_tstamp_pernet_fini(net
);
1530 nf_conntrack_acct_pernet_fini(net
);
1531 nf_conntrack_expect_pernet_fini(net
);
1532 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1533 kfree(net
->ct
.slabname
);
1534 free_percpu(net
->ct
.stat
);
1535 free_percpu(net
->ct
.pcpu_lists
);
1539 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1541 struct hlist_nulls_head
*hash
;
1542 unsigned int nr_slots
, i
;
1545 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1546 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1547 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1548 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1554 for (i
= 0; i
< nr_slots
; i
++)
1555 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1559 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1561 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1564 unsigned int hashsize
, old_size
;
1565 struct hlist_nulls_head
*hash
, *old_hash
;
1566 struct nf_conntrack_tuple_hash
*h
;
1569 if (current
->nsproxy
->net_ns
!= &init_net
)
1572 /* On boot, we can set this without any fancy locking. */
1573 if (!nf_conntrack_htable_size
)
1574 return param_set_uint(val
, kp
);
1576 rc
= kstrtouint(val
, 0, &hashsize
);
1582 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1587 nf_conntrack_all_lock();
1588 write_seqcount_begin(&init_net
.ct
.generation
);
1590 /* Lookups in the old hash might happen in parallel, which means we
1591 * might get false negatives during connection lookup. New connections
1592 * created because of a false negative won't make it into the hash
1593 * though since that required taking the locks.
1596 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1597 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1598 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1599 struct nf_conntrack_tuple_hash
, hnnode
);
1600 ct
= nf_ct_tuplehash_to_ctrack(h
);
1601 hlist_nulls_del_rcu(&h
->hnnode
);
1602 bucket
= __hash_conntrack(&h
->tuple
, hashsize
);
1603 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1606 old_size
= init_net
.ct
.htable_size
;
1607 old_hash
= init_net
.ct
.hash
;
1609 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1610 init_net
.ct
.hash
= hash
;
1612 write_seqcount_end(&init_net
.ct
.generation
);
1613 nf_conntrack_all_unlock();
1616 nf_ct_free_hashtable(old_hash
, old_size
);
1619 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1621 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1622 &nf_conntrack_htable_size
, 0600);
1624 void nf_ct_untracked_status_or(unsigned long bits
)
1628 for_each_possible_cpu(cpu
)
1629 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1631 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1633 int nf_conntrack_init_start(void)
1638 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
1639 spin_lock_init(&nf_conntrack_locks
[i
]);
1641 if (!nf_conntrack_htable_size
) {
1642 /* Idea from tcp.c: use 1/16384 of memory.
1643 * On i386: 32MB machine has 512 buckets.
1644 * >= 1GB machines have 16384 buckets.
1645 * >= 4GB machines have 65536 buckets.
1647 nf_conntrack_htable_size
1648 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1649 / sizeof(struct hlist_head
));
1650 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
1651 nf_conntrack_htable_size
= 65536;
1652 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1653 nf_conntrack_htable_size
= 16384;
1654 if (nf_conntrack_htable_size
< 32)
1655 nf_conntrack_htable_size
= 32;
1657 /* Use a max. factor of four by default to get the same max as
1658 * with the old struct list_heads. When a table size is given
1659 * we use the old value of 8 to avoid reducing the max.
1663 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1665 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1666 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1669 ret
= nf_conntrack_expect_init();
1673 ret
= nf_conntrack_acct_init();
1677 ret
= nf_conntrack_tstamp_init();
1681 ret
= nf_conntrack_ecache_init();
1685 ret
= nf_conntrack_timeout_init();
1689 ret
= nf_conntrack_helper_init();
1693 ret
= nf_conntrack_labels_init();
1697 ret
= nf_conntrack_seqadj_init();
1701 #ifdef CONFIG_NF_CONNTRACK_ZONES
1702 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1706 ret
= nf_conntrack_proto_init();
1710 /* Set up fake conntrack: to never be deleted, not in any hashes */
1711 for_each_possible_cpu(cpu
) {
1712 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1713 write_pnet(&ct
->ct_net
, &init_net
);
1714 atomic_set(&ct
->ct_general
.use
, 1);
1716 /* - and look it like as a confirmed connection */
1717 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1721 #ifdef CONFIG_NF_CONNTRACK_ZONES
1722 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1725 nf_conntrack_seqadj_fini();
1727 nf_conntrack_labels_fini();
1729 nf_conntrack_helper_fini();
1731 nf_conntrack_timeout_fini();
1733 nf_conntrack_ecache_fini();
1735 nf_conntrack_tstamp_fini();
1737 nf_conntrack_acct_fini();
1739 nf_conntrack_expect_fini();
1744 void nf_conntrack_init_end(void)
1746 /* For use by REJECT target */
1747 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1748 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1752 * We need to use special "null" values, not used in hash table
1754 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1755 #define DYING_NULLS_VAL ((1<<30)+1)
1756 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1758 int nf_conntrack_init_net(struct net
*net
)
1763 atomic_set(&net
->ct
.count
, 0);
1764 seqcount_init(&net
->ct
.generation
);
1766 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
1767 if (!net
->ct
.pcpu_lists
)
1770 for_each_possible_cpu(cpu
) {
1771 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1773 spin_lock_init(&pcpu
->lock
);
1774 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1775 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
1778 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1780 goto err_pcpu_lists
;
1782 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1783 if (!net
->ct
.slabname
)
1786 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1787 sizeof(struct nf_conn
), 0,
1788 SLAB_DESTROY_BY_RCU
, NULL
);
1789 if (!net
->ct
.nf_conntrack_cachep
) {
1790 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1794 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1795 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
, 1);
1796 if (!net
->ct
.hash
) {
1797 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1800 ret
= nf_conntrack_expect_pernet_init(net
);
1803 ret
= nf_conntrack_acct_pernet_init(net
);
1806 ret
= nf_conntrack_tstamp_pernet_init(net
);
1809 ret
= nf_conntrack_ecache_pernet_init(net
);
1812 ret
= nf_conntrack_helper_pernet_init(net
);
1815 ret
= nf_conntrack_proto_pernet_init(net
);
1821 nf_conntrack_helper_pernet_fini(net
);
1823 nf_conntrack_ecache_pernet_fini(net
);
1825 nf_conntrack_tstamp_pernet_fini(net
);
1827 nf_conntrack_acct_pernet_fini(net
);
1829 nf_conntrack_expect_pernet_fini(net
);
1831 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1833 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1835 kfree(net
->ct
.slabname
);
1837 free_percpu(net
->ct
.stat
);
1839 free_percpu(net
->ct
.pcpu_lists
);