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>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/module.h>
17 #include <linux/sched.h>
18 #include <linux/skbuff.h>
19 #include <linux/proc_fs.h>
20 #include <linux/vmalloc.h>
21 #include <linux/stddef.h>
22 #include <linux/slab.h>
23 #include <linux/random.h>
24 #include <linux/jhash.h>
25 #include <linux/err.h>
26 #include <linux/percpu.h>
27 #include <linux/moduleparam.h>
28 #include <linux/notifier.h>
29 #include <linux/kernel.h>
30 #include <linux/netdevice.h>
31 #include <linux/socket.h>
33 #include <linux/nsproxy.h>
34 #include <linux/rculist_nulls.h>
36 #include <net/netfilter/nf_conntrack.h>
37 #include <net/netfilter/nf_conntrack_l3proto.h>
38 #include <net/netfilter/nf_conntrack_l4proto.h>
39 #include <net/netfilter/nf_conntrack_expect.h>
40 #include <net/netfilter/nf_conntrack_helper.h>
41 #include <net/netfilter/nf_conntrack_core.h>
42 #include <net/netfilter/nf_conntrack_extend.h>
43 #include <net/netfilter/nf_conntrack_acct.h>
44 #include <net/netfilter/nf_conntrack_ecache.h>
45 #include <net/netfilter/nf_conntrack_zones.h>
46 #include <net/netfilter/nf_conntrack_timestamp.h>
47 #include <net/netfilter/nf_nat.h>
48 #include <net/netfilter/nf_nat_core.h>
50 #define NF_CONNTRACK_VERSION "0.5.0"
52 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
53 enum nf_nat_manip_type manip
,
54 const struct nlattr
*attr
) __read_mostly
;
55 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
57 DEFINE_SPINLOCK(nf_conntrack_lock
);
58 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
60 unsigned int nf_conntrack_htable_size __read_mostly
;
61 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
63 unsigned int nf_conntrack_max __read_mostly
;
64 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
66 DEFINE_PER_CPU(struct nf_conn
, nf_conntrack_untracked
);
67 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked
);
69 unsigned int nf_conntrack_hash_rnd __read_mostly
;
70 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd
);
72 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
, u16 zone
)
76 /* The direction must be ignored, so we hash everything up to the
77 * destination ports (which is a multiple of 4) and treat the last
78 * three bytes manually.
80 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
81 return jhash2((u32
*)tuple
, n
, zone
^ nf_conntrack_hash_rnd
^
82 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
83 tuple
->dst
.protonum
));
86 static u32
__hash_bucket(u32 hash
, unsigned int size
)
88 return ((u64
)hash
* size
) >> 32;
91 static u32
hash_bucket(u32 hash
, const struct net
*net
)
93 return __hash_bucket(hash
, net
->ct
.htable_size
);
96 static u_int32_t
__hash_conntrack(const struct nf_conntrack_tuple
*tuple
,
97 u16 zone
, unsigned int size
)
99 return __hash_bucket(hash_conntrack_raw(tuple
, zone
), size
);
102 static inline u_int32_t
hash_conntrack(const struct net
*net
, u16 zone
,
103 const struct nf_conntrack_tuple
*tuple
)
105 return __hash_conntrack(tuple
, zone
, net
->ct
.htable_size
);
109 nf_ct_get_tuple(const struct sk_buff
*skb
,
111 unsigned int dataoff
,
114 struct nf_conntrack_tuple
*tuple
,
115 const struct nf_conntrack_l3proto
*l3proto
,
116 const struct nf_conntrack_l4proto
*l4proto
)
118 memset(tuple
, 0, sizeof(*tuple
));
120 tuple
->src
.l3num
= l3num
;
121 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
124 tuple
->dst
.protonum
= protonum
;
125 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
127 return l4proto
->pkt_to_tuple(skb
, dataoff
, tuple
);
129 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
131 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
132 u_int16_t l3num
, struct nf_conntrack_tuple
*tuple
)
134 struct nf_conntrack_l3proto
*l3proto
;
135 struct nf_conntrack_l4proto
*l4proto
;
136 unsigned int protoff
;
142 l3proto
= __nf_ct_l3proto_find(l3num
);
143 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
144 if (ret
!= NF_ACCEPT
) {
149 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
151 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, tuple
,
157 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
160 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
161 const struct nf_conntrack_tuple
*orig
,
162 const struct nf_conntrack_l3proto
*l3proto
,
163 const struct nf_conntrack_l4proto
*l4proto
)
165 memset(inverse
, 0, sizeof(*inverse
));
167 inverse
->src
.l3num
= orig
->src
.l3num
;
168 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
171 inverse
->dst
.dir
= !orig
->dst
.dir
;
173 inverse
->dst
.protonum
= orig
->dst
.protonum
;
174 return l4proto
->invert_tuple(inverse
, orig
);
176 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
179 clean_from_lists(struct nf_conn
*ct
)
181 pr_debug("clean_from_lists(%p)\n", ct
);
182 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
183 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
185 /* Destroy all pending expectations */
186 nf_ct_remove_expectations(ct
);
190 destroy_conntrack(struct nf_conntrack
*nfct
)
192 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
193 struct net
*net
= nf_ct_net(ct
);
194 struct nf_conntrack_l4proto
*l4proto
;
196 pr_debug("destroy_conntrack(%p)\n", ct
);
197 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
198 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
200 /* To make sure we don't get any weird locking issues here:
201 * destroy_conntrack() MUST NOT be called with a write lock
202 * to nf_conntrack_lock!!! -HW */
204 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
205 if (l4proto
&& l4proto
->destroy
)
206 l4proto
->destroy(ct
);
210 spin_lock_bh(&nf_conntrack_lock
);
211 /* Expectations will have been removed in clean_from_lists,
212 * except TFTP can create an expectation on the first packet,
213 * before connection is in the list, so we need to clean here,
215 nf_ct_remove_expectations(ct
);
217 /* We overload first tuple to link into unconfirmed list. */
218 if (!nf_ct_is_confirmed(ct
)) {
219 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
220 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
223 NF_CT_STAT_INC(net
, delete);
224 spin_unlock_bh(&nf_conntrack_lock
);
227 nf_ct_put(ct
->master
);
229 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
230 nf_conntrack_free(ct
);
233 void nf_ct_delete_from_lists(struct nf_conn
*ct
)
235 struct net
*net
= nf_ct_net(ct
);
237 nf_ct_helper_destroy(ct
);
238 spin_lock_bh(&nf_conntrack_lock
);
239 /* Inside lock so preempt is disabled on module removal path.
240 * Otherwise we can get spurious warnings. */
241 NF_CT_STAT_INC(net
, delete_list
);
242 clean_from_lists(ct
);
243 spin_unlock_bh(&nf_conntrack_lock
);
245 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists
);
247 static void death_by_event(unsigned long ul_conntrack
)
249 struct nf_conn
*ct
= (void *)ul_conntrack
;
250 struct net
*net
= nf_ct_net(ct
);
252 if (nf_conntrack_event(IPCT_DESTROY
, ct
) < 0) {
253 /* bad luck, let's retry again */
254 ct
->timeout
.expires
= jiffies
+
255 (random32() % net
->ct
.sysctl_events_retry_timeout
);
256 add_timer(&ct
->timeout
);
259 /* we've got the event delivered, now it's dying */
260 set_bit(IPS_DYING_BIT
, &ct
->status
);
261 spin_lock(&nf_conntrack_lock
);
262 hlist_nulls_del(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
263 spin_unlock(&nf_conntrack_lock
);
267 void nf_ct_insert_dying_list(struct nf_conn
*ct
)
269 struct net
*net
= nf_ct_net(ct
);
271 /* add this conntrack to the dying list */
272 spin_lock_bh(&nf_conntrack_lock
);
273 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
275 spin_unlock_bh(&nf_conntrack_lock
);
276 /* set a new timer to retry event delivery */
277 setup_timer(&ct
->timeout
, death_by_event
, (unsigned long)ct
);
278 ct
->timeout
.expires
= jiffies
+
279 (random32() % net
->ct
.sysctl_events_retry_timeout
);
280 add_timer(&ct
->timeout
);
282 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list
);
284 static void death_by_timeout(unsigned long ul_conntrack
)
286 struct nf_conn
*ct
= (void *)ul_conntrack
;
287 struct nf_conn_tstamp
*tstamp
;
289 tstamp
= nf_conn_tstamp_find(ct
);
290 if (tstamp
&& tstamp
->stop
== 0)
291 tstamp
->stop
= ktime_to_ns(ktime_get_real());
293 if (!test_bit(IPS_DYING_BIT
, &ct
->status
) &&
294 unlikely(nf_conntrack_event(IPCT_DESTROY
, ct
) < 0)) {
295 /* destroy event was not delivered */
296 nf_ct_delete_from_lists(ct
);
297 nf_ct_insert_dying_list(ct
);
300 set_bit(IPS_DYING_BIT
, &ct
->status
);
301 nf_ct_delete_from_lists(ct
);
307 * - Caller must take a reference on returned object
308 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
310 * - Caller must lock nf_conntrack_lock before calling this function
312 static struct nf_conntrack_tuple_hash
*
313 ____nf_conntrack_find(struct net
*net
, u16 zone
,
314 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
316 struct nf_conntrack_tuple_hash
*h
;
317 struct hlist_nulls_node
*n
;
318 unsigned int bucket
= hash_bucket(hash
, net
);
320 /* Disable BHs the entire time since we normally need to disable them
321 * at least once for the stats anyway.
325 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[bucket
], hnnode
) {
326 if (nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
327 nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)) == zone
) {
328 NF_CT_STAT_INC(net
, found
);
332 NF_CT_STAT_INC(net
, searched
);
335 * if the nulls value we got at the end of this lookup is
336 * not the expected one, we must restart lookup.
337 * We probably met an item that was moved to another chain.
339 if (get_nulls_value(n
) != bucket
) {
340 NF_CT_STAT_INC(net
, search_restart
);
348 struct nf_conntrack_tuple_hash
*
349 __nf_conntrack_find(struct net
*net
, u16 zone
,
350 const struct nf_conntrack_tuple
*tuple
)
352 return ____nf_conntrack_find(net
, zone
, tuple
,
353 hash_conntrack_raw(tuple
, zone
));
355 EXPORT_SYMBOL_GPL(__nf_conntrack_find
);
357 /* Find a connection corresponding to a tuple. */
358 static struct nf_conntrack_tuple_hash
*
359 __nf_conntrack_find_get(struct net
*net
, u16 zone
,
360 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
362 struct nf_conntrack_tuple_hash
*h
;
367 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
369 ct
= nf_ct_tuplehash_to_ctrack(h
);
370 if (unlikely(nf_ct_is_dying(ct
) ||
371 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
374 if (unlikely(!nf_ct_tuple_equal(tuple
, &h
->tuple
) ||
375 nf_ct_zone(ct
) != zone
)) {
386 struct nf_conntrack_tuple_hash
*
387 nf_conntrack_find_get(struct net
*net
, u16 zone
,
388 const struct nf_conntrack_tuple
*tuple
)
390 return __nf_conntrack_find_get(net
, zone
, tuple
,
391 hash_conntrack_raw(tuple
, zone
));
393 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
395 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
397 unsigned int repl_hash
)
399 struct net
*net
= nf_ct_net(ct
);
401 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
402 &net
->ct
.hash
[hash
]);
403 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
404 &net
->ct
.hash
[repl_hash
]);
407 void nf_conntrack_hash_insert(struct nf_conn
*ct
)
409 struct net
*net
= nf_ct_net(ct
);
410 unsigned int hash
, repl_hash
;
413 zone
= nf_ct_zone(ct
);
414 hash
= hash_conntrack(net
, zone
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
415 repl_hash
= hash_conntrack(net
, zone
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
417 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
419 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert
);
421 /* Confirm a connection given skb; places it in hash table */
423 __nf_conntrack_confirm(struct sk_buff
*skb
)
425 unsigned int hash
, repl_hash
;
426 struct nf_conntrack_tuple_hash
*h
;
428 struct nf_conn_help
*help
;
429 struct nf_conn_tstamp
*tstamp
;
430 struct hlist_nulls_node
*n
;
431 enum ip_conntrack_info ctinfo
;
435 ct
= nf_ct_get(skb
, &ctinfo
);
438 /* ipt_REJECT uses nf_conntrack_attach to attach related
439 ICMP/TCP RST packets in other direction. Actual packet
440 which created connection will be IP_CT_NEW or for an
441 expected connection, IP_CT_RELATED. */
442 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
445 zone
= nf_ct_zone(ct
);
446 /* reuse the hash saved before */
447 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
448 hash
= hash_bucket(hash
, net
);
449 repl_hash
= hash_conntrack(net
, zone
,
450 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
452 /* We're not in hash table, and we refuse to set up related
453 connections for unconfirmed conns. But packet copies and
454 REJECT will give spurious warnings here. */
455 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
457 /* No external references means no one else could have
459 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
460 pr_debug("Confirming conntrack %p\n", ct
);
462 spin_lock_bh(&nf_conntrack_lock
);
464 /* We have to check the DYING flag inside the lock to prevent
465 a race against nf_ct_get_next_corpse() possibly called from
466 user context, else we insert an already 'dead' hash, blocking
467 further use of that particular connection -JM */
469 if (unlikely(nf_ct_is_dying(ct
))) {
470 spin_unlock_bh(&nf_conntrack_lock
);
474 /* See if there's one in the list already, including reverse:
475 NAT could have grabbed it without realizing, since we're
476 not in the hash. If there is, we lost race. */
477 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
478 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
480 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
482 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[repl_hash
], hnnode
)
483 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
485 zone
== nf_ct_zone(nf_ct_tuplehash_to_ctrack(h
)))
488 /* Remove from unconfirmed list */
489 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
491 /* Timer relative to confirmation time, not original
492 setting time, otherwise we'd get timer wrap in
493 weird delay cases. */
494 ct
->timeout
.expires
+= jiffies
;
495 add_timer(&ct
->timeout
);
496 atomic_inc(&ct
->ct_general
.use
);
497 ct
->status
|= IPS_CONFIRMED
;
499 /* set conntrack timestamp, if enabled. */
500 tstamp
= nf_conn_tstamp_find(ct
);
502 if (skb
->tstamp
.tv64
== 0)
503 __net_timestamp((struct sk_buff
*)skb
);
505 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
507 /* Since the lookup is lockless, hash insertion must be done after
508 * starting the timer and setting the CONFIRMED bit. The RCU barriers
509 * guarantee that no other CPU can find the conntrack before the above
510 * stores are visible.
512 __nf_conntrack_hash_insert(ct
, hash
, repl_hash
);
513 NF_CT_STAT_INC(net
, insert
);
514 spin_unlock_bh(&nf_conntrack_lock
);
516 help
= nfct_help(ct
);
517 if (help
&& help
->helper
)
518 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
520 nf_conntrack_event_cache(master_ct(ct
) ?
521 IPCT_RELATED
: IPCT_NEW
, ct
);
525 NF_CT_STAT_INC(net
, insert_failed
);
526 spin_unlock_bh(&nf_conntrack_lock
);
529 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
531 /* Returns true if a connection correspondings to the tuple (required
534 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
535 const struct nf_conn
*ignored_conntrack
)
537 struct net
*net
= nf_ct_net(ignored_conntrack
);
538 struct nf_conntrack_tuple_hash
*h
;
539 struct hlist_nulls_node
*n
;
541 u16 zone
= nf_ct_zone(ignored_conntrack
);
542 unsigned int hash
= hash_conntrack(net
, zone
, tuple
);
544 /* Disable BHs the entire time since we need to disable them at
545 * least once for the stats anyway.
548 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
], hnnode
) {
549 ct
= nf_ct_tuplehash_to_ctrack(h
);
550 if (ct
!= ignored_conntrack
&&
551 nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
552 nf_ct_zone(ct
) == zone
) {
553 NF_CT_STAT_INC(net
, found
);
554 rcu_read_unlock_bh();
557 NF_CT_STAT_INC(net
, searched
);
559 rcu_read_unlock_bh();
563 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
565 #define NF_CT_EVICTION_RANGE 8
567 /* There's a small race here where we may free a just-assured
568 connection. Too bad: we're in trouble anyway. */
569 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
571 /* Use oldest entry, which is roughly LRU */
572 struct nf_conntrack_tuple_hash
*h
;
573 struct nf_conn
*ct
= NULL
, *tmp
;
574 struct hlist_nulls_node
*n
;
575 unsigned int i
, cnt
= 0;
579 for (i
= 0; i
< net
->ct
.htable_size
; i
++) {
580 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[hash
],
582 tmp
= nf_ct_tuplehash_to_ctrack(h
);
583 if (!test_bit(IPS_ASSURED_BIT
, &tmp
->status
))
589 if (likely(!nf_ct_is_dying(ct
) &&
590 atomic_inc_not_zero(&ct
->ct_general
.use
)))
596 if (cnt
>= NF_CT_EVICTION_RANGE
)
599 hash
= (hash
+ 1) % net
->ct
.htable_size
;
606 if (del_timer(&ct
->timeout
)) {
607 death_by_timeout((unsigned long)ct
);
609 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
615 void init_nf_conntrack_hash_rnd(void)
620 * Why not initialize nf_conntrack_rnd in a "init()" function ?
621 * Because there isn't enough entropy when system initializing,
622 * and we initialize it as late as possible.
625 get_random_bytes(&rand
, sizeof(rand
));
627 cmpxchg(&nf_conntrack_hash_rnd
, 0, rand
);
630 static struct nf_conn
*
631 __nf_conntrack_alloc(struct net
*net
, u16 zone
,
632 const struct nf_conntrack_tuple
*orig
,
633 const struct nf_conntrack_tuple
*repl
,
638 if (unlikely(!nf_conntrack_hash_rnd
)) {
639 init_nf_conntrack_hash_rnd();
640 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
641 hash
= hash_conntrack_raw(orig
, zone
);
644 /* We don't want any race condition at early drop stage */
645 atomic_inc(&net
->ct
.count
);
647 if (nf_conntrack_max
&&
648 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
649 if (!early_drop(net
, hash_bucket(hash
, net
))) {
650 atomic_dec(&net
->ct
.count
);
653 "nf_conntrack: table full, dropping"
655 return ERR_PTR(-ENOMEM
);
660 * Do not use kmem_cache_zalloc(), as this cache uses
661 * SLAB_DESTROY_BY_RCU.
663 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
665 atomic_dec(&net
->ct
.count
);
666 return ERR_PTR(-ENOMEM
);
669 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
670 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
672 memset(&ct
->tuplehash
[IP_CT_DIR_MAX
], 0,
673 offsetof(struct nf_conn
, proto
) -
674 offsetof(struct nf_conn
, tuplehash
[IP_CT_DIR_MAX
]));
675 spin_lock_init(&ct
->lock
);
676 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
677 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
678 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
679 /* save hash for reusing when confirming */
680 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
681 /* Don't set timer yet: wait for confirmation */
682 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
683 write_pnet(&ct
->ct_net
, net
);
684 #ifdef CONFIG_NF_CONNTRACK_ZONES
686 struct nf_conntrack_zone
*nf_ct_zone
;
688 nf_ct_zone
= nf_ct_ext_add(ct
, NF_CT_EXT_ZONE
, GFP_ATOMIC
);
691 nf_ct_zone
->id
= zone
;
695 * changes to lookup keys must be done before setting refcnt to 1
698 atomic_set(&ct
->ct_general
.use
, 1);
701 #ifdef CONFIG_NF_CONNTRACK_ZONES
703 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
704 return ERR_PTR(-ENOMEM
);
708 struct nf_conn
*nf_conntrack_alloc(struct net
*net
, u16 zone
,
709 const struct nf_conntrack_tuple
*orig
,
710 const struct nf_conntrack_tuple
*repl
,
713 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
715 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
717 void nf_conntrack_free(struct nf_conn
*ct
)
719 struct net
*net
= nf_ct_net(ct
);
721 nf_ct_ext_destroy(ct
);
722 atomic_dec(&net
->ct
.count
);
724 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
726 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
728 /* Allocate a new conntrack: we return -ENOMEM if classification
729 failed due to stress. Otherwise it really is unclassifiable. */
730 static struct nf_conntrack_tuple_hash
*
731 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
732 const struct nf_conntrack_tuple
*tuple
,
733 struct nf_conntrack_l3proto
*l3proto
,
734 struct nf_conntrack_l4proto
*l4proto
,
736 unsigned int dataoff
, u32 hash
)
739 struct nf_conn_help
*help
;
740 struct nf_conntrack_tuple repl_tuple
;
741 struct nf_conntrack_ecache
*ecache
;
742 struct nf_conntrack_expect
*exp
;
743 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
745 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
746 pr_debug("Can't invert tuple.\n");
750 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
753 return (struct nf_conntrack_tuple_hash
*)ct
;
755 if (!l4proto
->new(ct
, skb
, dataoff
)) {
756 nf_conntrack_free(ct
);
757 pr_debug("init conntrack: can't track with proto module\n");
761 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
762 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
764 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
765 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
766 ecache
? ecache
->expmask
: 0,
769 spin_lock_bh(&nf_conntrack_lock
);
770 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
772 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
774 /* Welcome, Mr. Bond. We've been expecting you... */
775 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
776 ct
->master
= exp
->master
;
778 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
780 rcu_assign_pointer(help
->helper
, exp
->helper
);
783 #ifdef CONFIG_NF_CONNTRACK_MARK
784 ct
->mark
= exp
->master
->mark
;
786 #ifdef CONFIG_NF_CONNTRACK_SECMARK
787 ct
->secmark
= exp
->master
->secmark
;
789 nf_conntrack_get(&ct
->master
->ct_general
);
790 NF_CT_STAT_INC(net
, expect_new
);
792 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
793 NF_CT_STAT_INC(net
, new);
796 /* Overload tuple linked list to put us in unconfirmed list. */
797 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
798 &net
->ct
.unconfirmed
);
800 spin_unlock_bh(&nf_conntrack_lock
);
804 exp
->expectfn(ct
, exp
);
805 nf_ct_expect_put(exp
);
808 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
811 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
812 static inline struct nf_conn
*
813 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
815 unsigned int dataoff
,
818 struct nf_conntrack_l3proto
*l3proto
,
819 struct nf_conntrack_l4proto
*l4proto
,
821 enum ip_conntrack_info
*ctinfo
)
823 struct nf_conntrack_tuple tuple
;
824 struct nf_conntrack_tuple_hash
*h
;
826 u16 zone
= tmpl
? nf_ct_zone(tmpl
) : NF_CT_DEFAULT_ZONE
;
829 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
830 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
832 pr_debug("resolve_normal_ct: Can't get tuple\n");
836 /* look for tuple match */
837 hash
= hash_conntrack_raw(&tuple
, zone
);
838 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
840 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
847 ct
= nf_ct_tuplehash_to_ctrack(h
);
849 /* It exists; we have (non-exclusive) reference. */
850 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
851 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
852 /* Please set reply bit if this packet OK */
855 /* Once we've had two way comms, always ESTABLISHED. */
856 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
857 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
858 *ctinfo
= IP_CT_ESTABLISHED
;
859 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
860 pr_debug("nf_conntrack_in: related packet for %p\n",
862 *ctinfo
= IP_CT_RELATED
;
864 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
869 skb
->nfct
= &ct
->ct_general
;
870 skb
->nfctinfo
= *ctinfo
;
875 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
878 struct nf_conn
*ct
, *tmpl
= NULL
;
879 enum ip_conntrack_info ctinfo
;
880 struct nf_conntrack_l3proto
*l3proto
;
881 struct nf_conntrack_l4proto
*l4proto
;
882 unsigned int dataoff
;
888 /* Previously seen (loopback or untracked)? Ignore. */
889 tmpl
= (struct nf_conn
*)skb
->nfct
;
890 if (!nf_ct_is_template(tmpl
)) {
891 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
897 /* rcu_read_lock()ed by nf_hook_slow */
898 l3proto
= __nf_ct_l3proto_find(pf
);
899 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
900 &dataoff
, &protonum
);
902 pr_debug("not prepared to track yet or error occurred\n");
903 NF_CT_STAT_INC_ATOMIC(net
, error
);
904 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
909 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
911 /* It may be an special packet, error, unclean...
912 * inverse of the return code tells to the netfilter
913 * core what to do with the packet. */
914 if (l4proto
->error
!= NULL
) {
915 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
918 NF_CT_STAT_INC_ATOMIC(net
, error
);
919 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
923 /* ICMP[v6] protocol trackers may assign one conntrack. */
928 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
929 l3proto
, l4proto
, &set_reply
, &ctinfo
);
931 /* Not valid part of a connection */
932 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
938 /* Too stressed to deal. */
939 NF_CT_STAT_INC_ATOMIC(net
, drop
);
944 NF_CT_ASSERT(skb
->nfct
);
946 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
);
948 /* Invalid: inverse of the return code tells
949 * the netfilter core what to do */
950 pr_debug("nf_conntrack_in: Can't track with proto module\n");
951 nf_conntrack_put(skb
->nfct
);
953 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
955 NF_CT_STAT_INC_ATOMIC(net
, drop
);
960 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
961 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
964 /* Special case: we have to repeat this hook, assign the
965 * template again to this packet. We assume that this packet
966 * has no conntrack assigned. This is used by nf_ct_tcp. */
967 if (ret
== NF_REPEAT
)
968 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
975 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
977 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
978 const struct nf_conntrack_tuple
*orig
)
983 ret
= nf_ct_invert_tuple(inverse
, orig
,
984 __nf_ct_l3proto_find(orig
->src
.l3num
),
985 __nf_ct_l4proto_find(orig
->src
.l3num
,
986 orig
->dst
.protonum
));
990 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
992 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
993 implicitly racy: see __nf_conntrack_confirm */
994 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
995 const struct nf_conntrack_tuple
*newreply
)
997 struct nf_conn_help
*help
= nfct_help(ct
);
999 /* Should be unconfirmed, so not in hash table yet */
1000 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1002 pr_debug("Altering reply tuple of %p to ", ct
);
1003 nf_ct_dump_tuple(newreply
);
1005 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1006 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1010 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1013 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1015 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1016 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1017 enum ip_conntrack_info ctinfo
,
1018 const struct sk_buff
*skb
,
1019 unsigned long extra_jiffies
,
1022 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1025 /* Only update if this is not a fixed timeout */
1026 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1029 /* If not in hash table, timer will not be active yet */
1030 if (!nf_ct_is_confirmed(ct
)) {
1031 ct
->timeout
.expires
= extra_jiffies
;
1033 unsigned long newtime
= jiffies
+ extra_jiffies
;
1035 /* Only update the timeout if the new timeout is at least
1036 HZ jiffies from the old timeout. Need del_timer for race
1037 avoidance (may already be dying). */
1038 if (newtime
- ct
->timeout
.expires
>= HZ
)
1039 mod_timer_pending(&ct
->timeout
, newtime
);
1044 struct nf_conn_counter
*acct
;
1046 acct
= nf_conn_acct_find(ct
);
1048 atomic64_inc(&acct
[CTINFO2DIR(ctinfo
)].packets
);
1049 atomic64_add(skb
->len
, &acct
[CTINFO2DIR(ctinfo
)].bytes
);
1053 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1055 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1056 enum ip_conntrack_info ctinfo
,
1057 const struct sk_buff
*skb
,
1061 struct nf_conn_counter
*acct
;
1063 acct
= nf_conn_acct_find(ct
);
1065 atomic64_inc(&acct
[CTINFO2DIR(ctinfo
)].packets
);
1066 atomic64_add(skb
->len
- skb_network_offset(skb
),
1067 &acct
[CTINFO2DIR(ctinfo
)].bytes
);
1071 if (del_timer(&ct
->timeout
)) {
1072 ct
->timeout
.function((unsigned long)ct
);
1077 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1079 #ifdef CONFIG_NF_CONNTRACK_ZONES
1080 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1081 .len
= sizeof(struct nf_conntrack_zone
),
1082 .align
= __alignof__(struct nf_conntrack_zone
),
1083 .id
= NF_CT_EXT_ZONE
,
1087 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1089 #include <linux/netfilter/nfnetlink.h>
1090 #include <linux/netfilter/nfnetlink_conntrack.h>
1091 #include <linux/mutex.h>
1093 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1094 * in ip_conntrack_core, since we don't want the protocols to autoload
1095 * or depend on ctnetlink */
1096 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1097 const struct nf_conntrack_tuple
*tuple
)
1099 NLA_PUT_BE16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
);
1100 NLA_PUT_BE16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
);
1106 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1108 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1109 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1110 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1112 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1114 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1115 struct nf_conntrack_tuple
*t
)
1117 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1120 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1121 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1125 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1127 int nf_ct_port_nlattr_tuple_size(void)
1129 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1131 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1134 /* Used by ipt_REJECT and ip6t_REJECT. */
1135 static void nf_conntrack_attach(struct sk_buff
*nskb
, struct sk_buff
*skb
)
1138 enum ip_conntrack_info ctinfo
;
1140 /* This ICMP is in reverse direction to the packet which caused it */
1141 ct
= nf_ct_get(skb
, &ctinfo
);
1142 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1143 ctinfo
= IP_CT_RELATED_REPLY
;
1145 ctinfo
= IP_CT_RELATED
;
1147 /* Attach to new skbuff, and increment count */
1148 nskb
->nfct
= &ct
->ct_general
;
1149 nskb
->nfctinfo
= ctinfo
;
1150 nf_conntrack_get(nskb
->nfct
);
1153 /* Bring out ya dead! */
1154 static struct nf_conn
*
1155 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1156 void *data
, unsigned int *bucket
)
1158 struct nf_conntrack_tuple_hash
*h
;
1160 struct hlist_nulls_node
*n
;
1162 spin_lock_bh(&nf_conntrack_lock
);
1163 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1164 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1165 ct
= nf_ct_tuplehash_to_ctrack(h
);
1170 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.unconfirmed
, hnnode
) {
1171 ct
= nf_ct_tuplehash_to_ctrack(h
);
1173 set_bit(IPS_DYING_BIT
, &ct
->status
);
1175 spin_unlock_bh(&nf_conntrack_lock
);
1178 atomic_inc(&ct
->ct_general
.use
);
1179 spin_unlock_bh(&nf_conntrack_lock
);
1183 void nf_ct_iterate_cleanup(struct net
*net
,
1184 int (*iter
)(struct nf_conn
*i
, void *data
),
1188 unsigned int bucket
= 0;
1190 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1191 /* Time to push up daises... */
1192 if (del_timer(&ct
->timeout
))
1193 death_by_timeout((unsigned long)ct
);
1194 /* ... else the timer will get him soon. */
1199 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1201 struct __nf_ct_flush_report
{
1206 static int kill_report(struct nf_conn
*i
, void *data
)
1208 struct __nf_ct_flush_report
*fr
= (struct __nf_ct_flush_report
*)data
;
1209 struct nf_conn_tstamp
*tstamp
;
1211 tstamp
= nf_conn_tstamp_find(i
);
1212 if (tstamp
&& tstamp
->stop
== 0)
1213 tstamp
->stop
= ktime_to_ns(ktime_get_real());
1215 /* If we fail to deliver the event, death_by_timeout() will retry */
1216 if (nf_conntrack_event_report(IPCT_DESTROY
, i
,
1217 fr
->pid
, fr
->report
) < 0)
1220 /* Avoid the delivery of the destroy event in death_by_timeout(). */
1221 set_bit(IPS_DYING_BIT
, &i
->status
);
1225 static int kill_all(struct nf_conn
*i
, void *data
)
1230 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1232 if (is_vmalloc_addr(hash
))
1235 free_pages((unsigned long)hash
,
1236 get_order(sizeof(struct hlist_head
) * size
));
1238 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1240 void nf_conntrack_flush_report(struct net
*net
, u32 pid
, int report
)
1242 struct __nf_ct_flush_report fr
= {
1246 nf_ct_iterate_cleanup(net
, kill_report
, &fr
);
1248 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report
);
1250 static void nf_ct_release_dying_list(struct net
*net
)
1252 struct nf_conntrack_tuple_hash
*h
;
1254 struct hlist_nulls_node
*n
;
1256 spin_lock_bh(&nf_conntrack_lock
);
1257 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.dying
, hnnode
) {
1258 ct
= nf_ct_tuplehash_to_ctrack(h
);
1259 /* never fails to remove them, no listeners at this point */
1262 spin_unlock_bh(&nf_conntrack_lock
);
1265 static int untrack_refs(void)
1269 for_each_possible_cpu(cpu
) {
1270 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1272 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1277 static void nf_conntrack_cleanup_init_net(void)
1279 while (untrack_refs() > 0)
1282 nf_conntrack_helper_fini();
1283 nf_conntrack_proto_fini();
1284 #ifdef CONFIG_NF_CONNTRACK_ZONES
1285 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1289 static void nf_conntrack_cleanup_net(struct net
*net
)
1292 nf_ct_iterate_cleanup(net
, kill_all
, NULL
);
1293 nf_ct_release_dying_list(net
);
1294 if (atomic_read(&net
->ct
.count
) != 0) {
1296 goto i_see_dead_people
;
1299 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1300 nf_conntrack_ecache_fini(net
);
1301 nf_conntrack_tstamp_fini(net
);
1302 nf_conntrack_acct_fini(net
);
1303 nf_conntrack_expect_fini(net
);
1304 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1305 kfree(net
->ct
.slabname
);
1306 free_percpu(net
->ct
.stat
);
1309 /* Mishearing the voices in his head, our hero wonders how he's
1310 supposed to kill the mall. */
1311 void nf_conntrack_cleanup(struct net
*net
)
1313 if (net_eq(net
, &init_net
))
1314 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1316 /* This makes sure all current packets have passed through
1317 netfilter framework. Roll on, two-stage module
1321 nf_conntrack_cleanup_net(net
);
1323 if (net_eq(net
, &init_net
)) {
1324 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1325 nf_conntrack_cleanup_init_net();
1329 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1331 struct hlist_nulls_head
*hash
;
1332 unsigned int nr_slots
, i
;
1335 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1336 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1337 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1338 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1341 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1346 for (i
= 0; i
< nr_slots
; i
++)
1347 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1351 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1353 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1356 unsigned int hashsize
, old_size
;
1357 struct hlist_nulls_head
*hash
, *old_hash
;
1358 struct nf_conntrack_tuple_hash
*h
;
1361 if (current
->nsproxy
->net_ns
!= &init_net
)
1364 /* On boot, we can set this without any fancy locking. */
1365 if (!nf_conntrack_htable_size
)
1366 return param_set_uint(val
, kp
);
1368 hashsize
= simple_strtoul(val
, NULL
, 0);
1372 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1376 /* Lookups in the old hash might happen in parallel, which means we
1377 * might get false negatives during connection lookup. New connections
1378 * created because of a false negative won't make it into the hash
1379 * though since that required taking the lock.
1381 spin_lock_bh(&nf_conntrack_lock
);
1382 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1383 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1384 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1385 struct nf_conntrack_tuple_hash
, hnnode
);
1386 ct
= nf_ct_tuplehash_to_ctrack(h
);
1387 hlist_nulls_del_rcu(&h
->hnnode
);
1388 bucket
= __hash_conntrack(&h
->tuple
, nf_ct_zone(ct
),
1390 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1393 old_size
= init_net
.ct
.htable_size
;
1394 old_hash
= init_net
.ct
.hash
;
1396 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1397 init_net
.ct
.hash
= hash
;
1398 spin_unlock_bh(&nf_conntrack_lock
);
1400 nf_ct_free_hashtable(old_hash
, old_size
);
1403 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1405 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1406 &nf_conntrack_htable_size
, 0600);
1408 void nf_ct_untracked_status_or(unsigned long bits
)
1412 for_each_possible_cpu(cpu
)
1413 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1415 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1417 static int nf_conntrack_init_init_net(void)
1422 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1423 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1424 if (!nf_conntrack_htable_size
) {
1425 nf_conntrack_htable_size
1426 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1427 / sizeof(struct hlist_head
));
1428 if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1429 nf_conntrack_htable_size
= 16384;
1430 if (nf_conntrack_htable_size
< 32)
1431 nf_conntrack_htable_size
= 32;
1433 /* Use a max. factor of four by default to get the same max as
1434 * with the old struct list_heads. When a table size is given
1435 * we use the old value of 8 to avoid reducing the max.
1439 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1441 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1442 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1445 ret
= nf_conntrack_proto_init();
1449 ret
= nf_conntrack_helper_init();
1453 #ifdef CONFIG_NF_CONNTRACK_ZONES
1454 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1458 /* Set up fake conntrack: to never be deleted, not in any hashes */
1459 for_each_possible_cpu(cpu
) {
1460 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1461 write_pnet(&ct
->ct_net
, &init_net
);
1462 atomic_set(&ct
->ct_general
.use
, 1);
1464 /* - and look it like as a confirmed connection */
1465 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1468 #ifdef CONFIG_NF_CONNTRACK_ZONES
1470 nf_conntrack_helper_fini();
1473 nf_conntrack_proto_fini();
1479 * We need to use special "null" values, not used in hash table
1481 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1482 #define DYING_NULLS_VAL ((1<<30)+1)
1484 static int nf_conntrack_init_net(struct net
*net
)
1488 atomic_set(&net
->ct
.count
, 0);
1489 INIT_HLIST_NULLS_HEAD(&net
->ct
.unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1490 INIT_HLIST_NULLS_HEAD(&net
->ct
.dying
, DYING_NULLS_VAL
);
1491 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1492 if (!net
->ct
.stat
) {
1497 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1498 if (!net
->ct
.slabname
) {
1503 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1504 sizeof(struct nf_conn
), 0,
1505 SLAB_DESTROY_BY_RCU
, NULL
);
1506 if (!net
->ct
.nf_conntrack_cachep
) {
1507 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1512 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1513 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
, 1);
1514 if (!net
->ct
.hash
) {
1516 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1519 ret
= nf_conntrack_expect_init(net
);
1522 ret
= nf_conntrack_acct_init(net
);
1525 ret
= nf_conntrack_tstamp_init(net
);
1528 ret
= nf_conntrack_ecache_init(net
);
1535 nf_conntrack_tstamp_fini(net
);
1537 nf_conntrack_acct_fini(net
);
1539 nf_conntrack_expect_fini(net
);
1541 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1543 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1545 kfree(net
->ct
.slabname
);
1547 free_percpu(net
->ct
.stat
);
1552 s16 (*nf_ct_nat_offset
)(const struct nf_conn
*ct
,
1553 enum ip_conntrack_dir dir
,
1555 EXPORT_SYMBOL_GPL(nf_ct_nat_offset
);
1557 int nf_conntrack_init(struct net
*net
)
1561 if (net_eq(net
, &init_net
)) {
1562 ret
= nf_conntrack_init_init_net();
1566 ret
= nf_conntrack_init_net(net
);
1570 if (net_eq(net
, &init_net
)) {
1571 /* For use by REJECT target */
1572 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1573 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1575 /* Howto get NAT offsets */
1576 RCU_INIT_POINTER(nf_ct_nat_offset
, NULL
);
1581 if (net_eq(net
, &init_net
))
1582 nf_conntrack_cleanup_init_net();