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
,
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
);
290 /* Released via destroy_conntrack() */
291 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
292 const struct nf_conntrack_zone
*zone
,
295 struct nf_conn
*tmpl
;
297 tmpl
= kzalloc(sizeof(struct nf_conn
), GFP_KERNEL
);
301 tmpl
->status
= IPS_TEMPLATE
;
302 write_pnet(&tmpl
->ct_net
, net
);
304 #ifdef CONFIG_NF_CONNTRACK_ZONES
306 struct nf_conntrack_zone
*nf_ct_zone
;
308 nf_ct_zone
= nf_ct_ext_add(tmpl
, NF_CT_EXT_ZONE
, GFP_ATOMIC
);
311 nf_ct_zone
->id
= zone
->id
;
312 nf_ct_zone
->dir
= zone
->dir
;
315 atomic_set(&tmpl
->ct_general
.use
, 0);
318 #ifdef CONFIG_NF_CONNTRACK_ZONES
324 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
326 static void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
328 nf_ct_ext_destroy(tmpl
);
329 nf_ct_ext_free(tmpl
);
334 destroy_conntrack(struct nf_conntrack
*nfct
)
336 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
337 struct net
*net
= nf_ct_net(ct
);
338 struct nf_conntrack_l4proto
*l4proto
;
340 pr_debug("destroy_conntrack(%p)\n", ct
);
341 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
342 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
344 if (unlikely(nf_ct_is_template(ct
))) {
349 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
350 if (l4proto
&& l4proto
->destroy
)
351 l4proto
->destroy(ct
);
356 /* Expectations will have been removed in clean_from_lists,
357 * except TFTP can create an expectation on the first packet,
358 * before connection is in the list, so we need to clean here,
361 nf_ct_remove_expectations(ct
);
363 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
365 NF_CT_STAT_INC(net
, delete);
369 nf_ct_put(ct
->master
);
371 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
372 nf_conntrack_free(ct
);
375 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
377 struct net
*net
= nf_ct_net(ct
);
378 unsigned int hash
, reply_hash
;
379 unsigned int sequence
;
381 nf_ct_helper_destroy(ct
);
385 sequence
= read_seqcount_begin(&net
->ct
.generation
);
386 hash
= hash_conntrack(net
,
387 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
388 reply_hash
= hash_conntrack(net
,
389 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
390 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
392 clean_from_lists(ct
);
393 nf_conntrack_double_unlock(hash
, reply_hash
);
395 nf_ct_add_to_dying_list(ct
);
397 NF_CT_STAT_INC(net
, delete_list
);
401 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
403 struct nf_conn_tstamp
*tstamp
;
405 tstamp
= nf_conn_tstamp_find(ct
);
406 if (tstamp
&& tstamp
->stop
== 0)
407 tstamp
->stop
= ktime_get_real_ns();
409 if (nf_ct_is_dying(ct
))
412 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
413 portid
, report
) < 0) {
414 /* destroy event was not delivered */
415 nf_ct_delete_from_lists(ct
);
416 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
420 nf_conntrack_ecache_work(nf_ct_net(ct
));
421 set_bit(IPS_DYING_BIT
, &ct
->status
);
423 nf_ct_delete_from_lists(ct
);
427 EXPORT_SYMBOL_GPL(nf_ct_delete
);
429 static void death_by_timeout(unsigned long ul_conntrack
)
431 nf_ct_delete((struct nf_conn
*)ul_conntrack
, 0, 0);
435 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
436 const struct nf_conntrack_tuple
*tuple
,
437 const struct nf_conntrack_zone
*zone
)
439 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
441 /* A conntrack can be recreated with the equal tuple,
442 * so we need to check that the conntrack is confirmed
444 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
445 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
446 nf_ct_is_confirmed(ct
);
451 * - Caller must take a reference on returned object
452 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
454 static struct nf_conntrack_tuple_hash
*
455 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
456 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
458 struct nf_conntrack_tuple_hash
*h
;
459 struct hlist_nulls_node
*n
;
460 unsigned int bucket
= hash_bucket(hash
, net
);
462 /* Disable BHs the entire time since we normally need to disable them
463 * at least once for the stats anyway.
467 hlist_nulls_for_each_entry_rcu(h
, n
, &net
->ct
.hash
[bucket
], hnnode
) {
468 if (nf_ct_key_equal(h
, tuple
, zone
)) {
469 NF_CT_STAT_INC(net
, found
);
473 NF_CT_STAT_INC(net
, searched
);
476 * if the nulls value we got at the end of this lookup is
477 * not the expected one, we must restart lookup.
478 * We probably met an item that was moved to another chain.
480 if (get_nulls_value(n
) != bucket
) {
481 NF_CT_STAT_INC(net
, search_restart
);
489 /* Find a connection corresponding to a tuple. */
490 static struct nf_conntrack_tuple_hash
*
491 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
492 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
494 struct nf_conntrack_tuple_hash
*h
;
499 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
501 ct
= nf_ct_tuplehash_to_ctrack(h
);
502 if (unlikely(nf_ct_is_dying(ct
) ||
503 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
506 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
))) {
517 struct nf_conntrack_tuple_hash
*
518 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
519 const struct nf_conntrack_tuple
*tuple
)
521 return __nf_conntrack_find_get(net
, zone
, tuple
,
522 hash_conntrack_raw(tuple
));
524 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
526 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
528 unsigned int reply_hash
)
530 struct net
*net
= nf_ct_net(ct
);
532 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
533 &net
->ct
.hash
[hash
]);
534 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
535 &net
->ct
.hash
[reply_hash
]);
539 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
541 const struct nf_conntrack_zone
*zone
;
542 struct net
*net
= nf_ct_net(ct
);
543 unsigned int hash
, reply_hash
;
544 struct nf_conntrack_tuple_hash
*h
;
545 struct hlist_nulls_node
*n
;
546 unsigned int sequence
;
548 zone
= nf_ct_zone(ct
);
552 sequence
= read_seqcount_begin(&net
->ct
.generation
);
553 hash
= hash_conntrack(net
,
554 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
555 reply_hash
= hash_conntrack(net
,
556 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
557 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
559 /* See if there's one in the list already, including reverse */
560 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
561 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
563 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
566 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
567 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
569 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
573 add_timer(&ct
->timeout
);
575 /* The caller holds a reference to this object */
576 atomic_set(&ct
->ct_general
.use
, 2);
577 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
578 nf_conntrack_double_unlock(hash
, reply_hash
);
579 NF_CT_STAT_INC(net
, insert
);
584 nf_conntrack_double_unlock(hash
, reply_hash
);
585 NF_CT_STAT_INC(net
, insert_failed
);
589 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
591 /* Confirm a connection given skb; places it in hash table */
593 __nf_conntrack_confirm(struct sk_buff
*skb
)
595 const struct nf_conntrack_zone
*zone
;
596 unsigned int hash
, reply_hash
;
597 struct nf_conntrack_tuple_hash
*h
;
599 struct nf_conn_help
*help
;
600 struct nf_conn_tstamp
*tstamp
;
601 struct hlist_nulls_node
*n
;
602 enum ip_conntrack_info ctinfo
;
604 unsigned int sequence
;
606 ct
= nf_ct_get(skb
, &ctinfo
);
609 /* ipt_REJECT uses nf_conntrack_attach to attach related
610 ICMP/TCP RST packets in other direction. Actual packet
611 which created connection will be IP_CT_NEW or for an
612 expected connection, IP_CT_RELATED. */
613 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
616 zone
= nf_ct_zone(ct
);
620 sequence
= read_seqcount_begin(&net
->ct
.generation
);
621 /* reuse the hash saved before */
622 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
623 hash
= hash_bucket(hash
, net
);
624 reply_hash
= hash_conntrack(net
,
625 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
627 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
629 /* We're not in hash table, and we refuse to set up related
630 * connections for unconfirmed conns. But packet copies and
631 * REJECT will give spurious warnings here.
633 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
635 /* No external references means no one else could have
638 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
639 pr_debug("Confirming conntrack %p\n", ct
);
640 /* We have to check the DYING flag after unlink to prevent
641 * a race against nf_ct_get_next_corpse() possibly called from
642 * user context, else we insert an already 'dead' hash, blocking
643 * further use of that particular connection -JM.
645 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
647 if (unlikely(nf_ct_is_dying(ct
)))
650 /* See if there's one in the list already, including reverse:
651 NAT could have grabbed it without realizing, since we're
652 not in the hash. If there is, we lost race. */
653 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[hash
], hnnode
)
654 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
656 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
659 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[reply_hash
], hnnode
)
660 if (nf_ct_tuple_equal(&ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
662 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h
), zone
,
666 /* Timer relative to confirmation time, not original
667 setting time, otherwise we'd get timer wrap in
668 weird delay cases. */
669 ct
->timeout
.expires
+= jiffies
;
670 add_timer(&ct
->timeout
);
671 atomic_inc(&ct
->ct_general
.use
);
672 ct
->status
|= IPS_CONFIRMED
;
674 /* set conntrack timestamp, if enabled. */
675 tstamp
= nf_conn_tstamp_find(ct
);
677 if (skb
->tstamp
.tv64
== 0)
678 __net_timestamp(skb
);
680 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
682 /* Since the lookup is lockless, hash insertion must be done after
683 * starting the timer and setting the CONFIRMED bit. The RCU barriers
684 * guarantee that no other CPU can find the conntrack before the above
685 * stores are visible.
687 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
688 nf_conntrack_double_unlock(hash
, reply_hash
);
689 NF_CT_STAT_INC(net
, insert
);
692 help
= nfct_help(ct
);
693 if (help
&& help
->helper
)
694 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
696 nf_conntrack_event_cache(master_ct(ct
) ?
697 IPCT_RELATED
: IPCT_NEW
, ct
);
701 nf_ct_add_to_dying_list(ct
);
702 nf_conntrack_double_unlock(hash
, reply_hash
);
703 NF_CT_STAT_INC(net
, insert_failed
);
707 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
709 /* Returns true if a connection correspondings to the tuple (required
712 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
713 const struct nf_conn
*ignored_conntrack
)
715 struct net
*net
= nf_ct_net(ignored_conntrack
);
716 const struct nf_conntrack_zone
*zone
;
717 struct nf_conntrack_tuple_hash
*h
;
718 struct hlist_nulls_node
*n
;
722 zone
= nf_ct_zone(ignored_conntrack
);
723 hash
= hash_conntrack(net
, 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_equal(ct
, zone
, NF_CT_DIRECTION(h
))) {
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
,
823 const struct nf_conntrack_zone
*zone
,
824 const struct nf_conntrack_tuple
*orig
,
825 const struct nf_conntrack_tuple
*repl
,
830 if (unlikely(!nf_conntrack_hash_rnd
)) {
831 init_nf_conntrack_hash_rnd();
832 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
833 hash
= hash_conntrack_raw(orig
);
836 /* We don't want any race condition at early drop stage */
837 atomic_inc(&net
->ct
.count
);
839 if (nf_conntrack_max
&&
840 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
841 if (!early_drop(net
, hash
)) {
842 atomic_dec(&net
->ct
.count
);
843 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
844 return ERR_PTR(-ENOMEM
);
849 * Do not use kmem_cache_zalloc(), as this cache uses
850 * SLAB_DESTROY_BY_RCU.
852 ct
= kmem_cache_alloc(net
->ct
.nf_conntrack_cachep
, gfp
);
854 atomic_dec(&net
->ct
.count
);
855 return ERR_PTR(-ENOMEM
);
857 spin_lock_init(&ct
->lock
);
858 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
859 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
860 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
861 /* save hash for reusing when confirming */
862 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
864 /* Don't set timer yet: wait for confirmation */
865 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
866 write_pnet(&ct
->ct_net
, net
);
867 memset(&ct
->__nfct_init_offset
[0], 0,
868 offsetof(struct nf_conn
, proto
) -
869 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
870 #ifdef CONFIG_NF_CONNTRACK_ZONES
872 struct nf_conntrack_zone
*nf_ct_zone
;
874 nf_ct_zone
= nf_ct_ext_add(ct
, NF_CT_EXT_ZONE
, GFP_ATOMIC
);
877 nf_ct_zone
->id
= zone
->id
;
878 nf_ct_zone
->dir
= zone
->dir
;
881 /* Because we use RCU lookups, we set ct_general.use to zero before
882 * this is inserted in any list.
884 atomic_set(&ct
->ct_general
.use
, 0);
887 #ifdef CONFIG_NF_CONNTRACK_ZONES
889 atomic_dec(&net
->ct
.count
);
890 kmem_cache_free(net
->ct
.nf_conntrack_cachep
, ct
);
891 return ERR_PTR(-ENOMEM
);
895 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
896 const struct nf_conntrack_zone
*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 const struct nf_conntrack_zone
*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 zone
= nf_ct_zone_tmpl(tmpl
);
948 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
951 return (struct nf_conntrack_tuple_hash
*)ct
;
953 if (tmpl
&& nfct_synproxy(tmpl
)) {
954 nfct_seqadj_ext_add(ct
);
955 nfct_synproxy_ext_add(ct
);
958 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
960 timeouts
= NF_CT_TIMEOUT_EXT_DATA(timeout_ext
);
962 timeouts
= l4proto
->get_timeouts(net
);
964 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
965 nf_conntrack_free(ct
);
966 pr_debug("init conntrack: can't track with proto module\n");
971 nf_ct_timeout_ext_add(ct
, timeout_ext
->timeout
, GFP_ATOMIC
);
973 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
974 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
975 nf_ct_labels_ext_add(ct
);
977 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
978 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
979 ecache
? ecache
->expmask
: 0,
983 if (net
->ct
.expect_count
) {
984 spin_lock(&nf_conntrack_expect_lock
);
985 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
987 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
989 /* Welcome, Mr. Bond. We've been expecting you... */
990 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
991 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
992 ct
->master
= exp
->master
;
994 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
997 rcu_assign_pointer(help
->helper
, exp
->helper
);
1000 #ifdef CONFIG_NF_CONNTRACK_MARK
1001 ct
->mark
= exp
->master
->mark
;
1003 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1004 ct
->secmark
= exp
->master
->secmark
;
1006 NF_CT_STAT_INC(net
, expect_new
);
1008 spin_unlock(&nf_conntrack_expect_lock
);
1011 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1012 NF_CT_STAT_INC(net
, new);
1015 /* Now it is inserted into the unconfirmed list, bump refcount */
1016 nf_conntrack_get(&ct
->ct_general
);
1017 nf_ct_add_to_unconfirmed_list(ct
);
1023 exp
->expectfn(ct
, exp
);
1024 nf_ct_expect_put(exp
);
1027 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1030 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1031 static inline struct nf_conn
*
1032 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1033 struct sk_buff
*skb
,
1034 unsigned int dataoff
,
1037 struct nf_conntrack_l3proto
*l3proto
,
1038 struct nf_conntrack_l4proto
*l4proto
,
1040 enum ip_conntrack_info
*ctinfo
)
1042 const struct nf_conntrack_zone
*zone
;
1043 struct nf_conntrack_tuple tuple
;
1044 struct nf_conntrack_tuple_hash
*h
;
1048 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1049 dataoff
, l3num
, protonum
, &tuple
, l3proto
,
1051 pr_debug("resolve_normal_ct: Can't get tuple\n");
1055 /* look for tuple match */
1056 zone
= nf_ct_zone_tmpl(tmpl
);
1057 hash
= hash_conntrack_raw(&tuple
);
1058 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1060 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1061 skb
, dataoff
, hash
);
1067 ct
= nf_ct_tuplehash_to_ctrack(h
);
1069 /* It exists; we have (non-exclusive) reference. */
1070 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1071 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1072 /* Please set reply bit if this packet OK */
1075 /* Once we've had two way comms, always ESTABLISHED. */
1076 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1077 pr_debug("nf_conntrack_in: normal packet for %p\n", ct
);
1078 *ctinfo
= IP_CT_ESTABLISHED
;
1079 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1080 pr_debug("nf_conntrack_in: related packet for %p\n",
1082 *ctinfo
= IP_CT_RELATED
;
1084 pr_debug("nf_conntrack_in: new packet for %p\n", ct
);
1085 *ctinfo
= IP_CT_NEW
;
1089 skb
->nfct
= &ct
->ct_general
;
1090 skb
->nfctinfo
= *ctinfo
;
1095 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1096 struct sk_buff
*skb
)
1098 struct nf_conn
*ct
, *tmpl
= NULL
;
1099 enum ip_conntrack_info ctinfo
;
1100 struct nf_conntrack_l3proto
*l3proto
;
1101 struct nf_conntrack_l4proto
*l4proto
;
1102 unsigned int *timeouts
;
1103 unsigned int dataoff
;
1109 /* Previously seen (loopback or untracked)? Ignore. */
1110 tmpl
= (struct nf_conn
*)skb
->nfct
;
1111 if (!nf_ct_is_template(tmpl
)) {
1112 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1118 /* rcu_read_lock()ed by nf_hook_slow */
1119 l3proto
= __nf_ct_l3proto_find(pf
);
1120 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1121 &dataoff
, &protonum
);
1123 pr_debug("not prepared to track yet or error occurred\n");
1124 NF_CT_STAT_INC_ATOMIC(net
, error
);
1125 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1130 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1132 /* It may be an special packet, error, unclean...
1133 * inverse of the return code tells to the netfilter
1134 * core what to do with the packet. */
1135 if (l4proto
->error
!= NULL
) {
1136 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
1139 NF_CT_STAT_INC_ATOMIC(net
, error
);
1140 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1144 /* ICMP[v6] protocol trackers may assign one conntrack. */
1149 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1150 l3proto
, l4proto
, &set_reply
, &ctinfo
);
1152 /* Not valid part of a connection */
1153 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1159 /* Too stressed to deal. */
1160 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1165 NF_CT_ASSERT(skb
->nfct
);
1167 /* Decide what timeout policy we want to apply to this flow. */
1168 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1170 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1172 /* Invalid: inverse of the return code tells
1173 * the netfilter core what to do */
1174 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1175 nf_conntrack_put(skb
->nfct
);
1177 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1178 if (ret
== -NF_DROP
)
1179 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1184 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1185 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1188 /* Special case: we have to repeat this hook, assign the
1189 * template again to this packet. We assume that this packet
1190 * has no conntrack assigned. This is used by nf_ct_tcp. */
1191 if (ret
== NF_REPEAT
)
1192 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
1199 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1201 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1202 const struct nf_conntrack_tuple
*orig
)
1207 ret
= nf_ct_invert_tuple(inverse
, orig
,
1208 __nf_ct_l3proto_find(orig
->src
.l3num
),
1209 __nf_ct_l4proto_find(orig
->src
.l3num
,
1210 orig
->dst
.protonum
));
1214 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1216 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1217 implicitly racy: see __nf_conntrack_confirm */
1218 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1219 const struct nf_conntrack_tuple
*newreply
)
1221 struct nf_conn_help
*help
= nfct_help(ct
);
1223 /* Should be unconfirmed, so not in hash table yet */
1224 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1226 pr_debug("Altering reply tuple of %p to ", ct
);
1227 nf_ct_dump_tuple(newreply
);
1229 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1230 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1234 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1237 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1239 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1240 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1241 enum ip_conntrack_info ctinfo
,
1242 const struct sk_buff
*skb
,
1243 unsigned long extra_jiffies
,
1246 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1249 /* Only update if this is not a fixed timeout */
1250 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1253 /* If not in hash table, timer will not be active yet */
1254 if (!nf_ct_is_confirmed(ct
)) {
1255 ct
->timeout
.expires
= extra_jiffies
;
1257 unsigned long newtime
= jiffies
+ extra_jiffies
;
1259 /* Only update the timeout if the new timeout is at least
1260 HZ jiffies from the old timeout. Need del_timer for race
1261 avoidance (may already be dying). */
1262 if (newtime
- ct
->timeout
.expires
>= HZ
)
1263 mod_timer_pending(&ct
->timeout
, newtime
);
1268 struct nf_conn_acct
*acct
;
1270 acct
= nf_conn_acct_find(ct
);
1272 struct nf_conn_counter
*counter
= acct
->counter
;
1274 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1275 atomic64_add(skb
->len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1279 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1281 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1282 enum ip_conntrack_info ctinfo
,
1283 const struct sk_buff
*skb
,
1287 struct nf_conn_acct
*acct
;
1289 acct
= nf_conn_acct_find(ct
);
1291 struct nf_conn_counter
*counter
= acct
->counter
;
1293 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
1294 atomic64_add(skb
->len
- skb_network_offset(skb
),
1295 &counter
[CTINFO2DIR(ctinfo
)].bytes
);
1299 if (del_timer(&ct
->timeout
)) {
1300 ct
->timeout
.function((unsigned long)ct
);
1305 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1307 /* Built-in default zone used e.g. by modules. */
1308 const struct nf_conntrack_zone nf_ct_zone_dflt
= {
1309 .id
= NF_CT_DEFAULT_ZONE_ID
,
1310 .dir
= NF_CT_DEFAULT_ZONE_DIR
,
1312 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt
);
1314 #ifdef CONFIG_NF_CONNTRACK_ZONES
1315 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1316 .len
= sizeof(struct nf_conntrack_zone
),
1317 .align
= __alignof__(struct nf_conntrack_zone
),
1318 .id
= NF_CT_EXT_ZONE
,
1322 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1324 #include <linux/netfilter/nfnetlink.h>
1325 #include <linux/netfilter/nfnetlink_conntrack.h>
1326 #include <linux/mutex.h>
1328 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1329 * in ip_conntrack_core, since we don't want the protocols to autoload
1330 * or depend on ctnetlink */
1331 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1332 const struct nf_conntrack_tuple
*tuple
)
1334 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1335 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1336 goto nla_put_failure
;
1342 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1344 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1345 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1346 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1348 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1350 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1351 struct nf_conntrack_tuple
*t
)
1353 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1356 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1357 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1361 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1363 int nf_ct_port_nlattr_tuple_size(void)
1365 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1367 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1370 /* Used by ipt_REJECT and ip6t_REJECT. */
1371 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1374 enum ip_conntrack_info ctinfo
;
1376 /* This ICMP is in reverse direction to the packet which caused it */
1377 ct
= nf_ct_get(skb
, &ctinfo
);
1378 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1379 ctinfo
= IP_CT_RELATED_REPLY
;
1381 ctinfo
= IP_CT_RELATED
;
1383 /* Attach to new skbuff, and increment count */
1384 nskb
->nfct
= &ct
->ct_general
;
1385 nskb
->nfctinfo
= ctinfo
;
1386 nf_conntrack_get(nskb
->nfct
);
1389 /* Bring out ya dead! */
1390 static struct nf_conn
*
1391 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1392 void *data
, unsigned int *bucket
)
1394 struct nf_conntrack_tuple_hash
*h
;
1396 struct hlist_nulls_node
*n
;
1400 for (; *bucket
< net
->ct
.htable_size
; (*bucket
)++) {
1401 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1404 if (*bucket
< net
->ct
.htable_size
) {
1405 hlist_nulls_for_each_entry(h
, n
, &net
->ct
.hash
[*bucket
], hnnode
) {
1406 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1408 ct
= nf_ct_tuplehash_to_ctrack(h
);
1417 for_each_possible_cpu(cpu
) {
1418 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1420 spin_lock_bh(&pcpu
->lock
);
1421 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1422 ct
= nf_ct_tuplehash_to_ctrack(h
);
1424 set_bit(IPS_DYING_BIT
, &ct
->status
);
1426 spin_unlock_bh(&pcpu
->lock
);
1430 atomic_inc(&ct
->ct_general
.use
);
1436 void nf_ct_iterate_cleanup(struct net
*net
,
1437 int (*iter
)(struct nf_conn
*i
, void *data
),
1438 void *data
, u32 portid
, int report
)
1441 unsigned int bucket
= 0;
1443 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1444 /* Time to push up daises... */
1445 if (del_timer(&ct
->timeout
))
1446 nf_ct_delete(ct
, portid
, report
);
1448 /* ... else the timer will get him soon. */
1453 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1455 static int kill_all(struct nf_conn
*i
, void *data
)
1460 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1462 if (is_vmalloc_addr(hash
))
1465 free_pages((unsigned long)hash
,
1466 get_order(sizeof(struct hlist_head
) * size
));
1468 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1470 static int untrack_refs(void)
1474 for_each_possible_cpu(cpu
) {
1475 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1477 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1482 void nf_conntrack_cleanup_start(void)
1484 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1487 void nf_conntrack_cleanup_end(void)
1489 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1490 while (untrack_refs() > 0)
1493 #ifdef CONFIG_NF_CONNTRACK_ZONES
1494 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1496 nf_conntrack_proto_fini();
1497 nf_conntrack_seqadj_fini();
1498 nf_conntrack_labels_fini();
1499 nf_conntrack_helper_fini();
1500 nf_conntrack_timeout_fini();
1501 nf_conntrack_ecache_fini();
1502 nf_conntrack_tstamp_fini();
1503 nf_conntrack_acct_fini();
1504 nf_conntrack_expect_fini();
1508 * Mishearing the voices in his head, our hero wonders how he's
1509 * supposed to kill the mall.
1511 void nf_conntrack_cleanup_net(struct net
*net
)
1515 list_add(&net
->exit_list
, &single
);
1516 nf_conntrack_cleanup_net_list(&single
);
1519 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1525 * This makes sure all current packets have passed through
1526 * netfilter framework. Roll on, two-stage module
1532 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1533 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1534 if (atomic_read(&net
->ct
.count
) != 0)
1539 goto i_see_dead_people
;
1542 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1543 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1544 nf_conntrack_proto_pernet_fini(net
);
1545 nf_conntrack_helper_pernet_fini(net
);
1546 nf_conntrack_ecache_pernet_fini(net
);
1547 nf_conntrack_tstamp_pernet_fini(net
);
1548 nf_conntrack_acct_pernet_fini(net
);
1549 nf_conntrack_expect_pernet_fini(net
);
1550 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1551 kfree(net
->ct
.slabname
);
1552 free_percpu(net
->ct
.stat
);
1553 free_percpu(net
->ct
.pcpu_lists
);
1557 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1559 struct hlist_nulls_head
*hash
;
1560 unsigned int nr_slots
, i
;
1563 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1564 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1565 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1566 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1569 printk(KERN_WARNING
"nf_conntrack: falling back to vmalloc.\n");
1574 for (i
= 0; i
< nr_slots
; i
++)
1575 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1579 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1581 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1584 unsigned int hashsize
, old_size
;
1585 struct hlist_nulls_head
*hash
, *old_hash
;
1586 struct nf_conntrack_tuple_hash
*h
;
1589 if (current
->nsproxy
->net_ns
!= &init_net
)
1592 /* On boot, we can set this without any fancy locking. */
1593 if (!nf_conntrack_htable_size
)
1594 return param_set_uint(val
, kp
);
1596 rc
= kstrtouint(val
, 0, &hashsize
);
1602 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1607 nf_conntrack_all_lock();
1608 write_seqcount_begin(&init_net
.ct
.generation
);
1610 /* Lookups in the old hash might happen in parallel, which means we
1611 * might get false negatives during connection lookup. New connections
1612 * created because of a false negative won't make it into the hash
1613 * though since that required taking the locks.
1616 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
1617 while (!hlist_nulls_empty(&init_net
.ct
.hash
[i
])) {
1618 h
= hlist_nulls_entry(init_net
.ct
.hash
[i
].first
,
1619 struct nf_conntrack_tuple_hash
, hnnode
);
1620 ct
= nf_ct_tuplehash_to_ctrack(h
);
1621 hlist_nulls_del_rcu(&h
->hnnode
);
1622 bucket
= __hash_conntrack(&h
->tuple
, hashsize
);
1623 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1626 old_size
= init_net
.ct
.htable_size
;
1627 old_hash
= init_net
.ct
.hash
;
1629 init_net
.ct
.htable_size
= nf_conntrack_htable_size
= hashsize
;
1630 init_net
.ct
.hash
= hash
;
1632 write_seqcount_end(&init_net
.ct
.generation
);
1633 nf_conntrack_all_unlock();
1636 nf_ct_free_hashtable(old_hash
, old_size
);
1639 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1641 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1642 &nf_conntrack_htable_size
, 0600);
1644 void nf_ct_untracked_status_or(unsigned long bits
)
1648 for_each_possible_cpu(cpu
)
1649 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1651 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1653 int nf_conntrack_init_start(void)
1658 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
1659 spin_lock_init(&nf_conntrack_locks
[i
]);
1661 if (!nf_conntrack_htable_size
) {
1662 /* Idea from tcp.c: use 1/16384 of memory.
1663 * On i386: 32MB machine has 512 buckets.
1664 * >= 1GB machines have 16384 buckets.
1665 * >= 4GB machines have 65536 buckets.
1667 nf_conntrack_htable_size
1668 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1669 / sizeof(struct hlist_head
));
1670 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
1671 nf_conntrack_htable_size
= 65536;
1672 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1673 nf_conntrack_htable_size
= 16384;
1674 if (nf_conntrack_htable_size
< 32)
1675 nf_conntrack_htable_size
= 32;
1677 /* Use a max. factor of four by default to get the same max as
1678 * with the old struct list_heads. When a table size is given
1679 * we use the old value of 8 to avoid reducing the max.
1683 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1685 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1686 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1689 ret
= nf_conntrack_expect_init();
1693 ret
= nf_conntrack_acct_init();
1697 ret
= nf_conntrack_tstamp_init();
1701 ret
= nf_conntrack_ecache_init();
1705 ret
= nf_conntrack_timeout_init();
1709 ret
= nf_conntrack_helper_init();
1713 ret
= nf_conntrack_labels_init();
1717 ret
= nf_conntrack_seqadj_init();
1721 #ifdef CONFIG_NF_CONNTRACK_ZONES
1722 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1726 ret
= nf_conntrack_proto_init();
1730 /* Set up fake conntrack: to never be deleted, not in any hashes */
1731 for_each_possible_cpu(cpu
) {
1732 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1733 write_pnet(&ct
->ct_net
, &init_net
);
1734 atomic_set(&ct
->ct_general
.use
, 1);
1736 /* - and look it like as a confirmed connection */
1737 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1741 #ifdef CONFIG_NF_CONNTRACK_ZONES
1742 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1745 nf_conntrack_seqadj_fini();
1747 nf_conntrack_labels_fini();
1749 nf_conntrack_helper_fini();
1751 nf_conntrack_timeout_fini();
1753 nf_conntrack_ecache_fini();
1755 nf_conntrack_tstamp_fini();
1757 nf_conntrack_acct_fini();
1759 nf_conntrack_expect_fini();
1764 void nf_conntrack_init_end(void)
1766 /* For use by REJECT target */
1767 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1768 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1772 * We need to use special "null" values, not used in hash table
1774 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1775 #define DYING_NULLS_VAL ((1<<30)+1)
1776 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1778 int nf_conntrack_init_net(struct net
*net
)
1783 atomic_set(&net
->ct
.count
, 0);
1784 seqcount_init(&net
->ct
.generation
);
1786 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
1787 if (!net
->ct
.pcpu_lists
)
1790 for_each_possible_cpu(cpu
) {
1791 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1793 spin_lock_init(&pcpu
->lock
);
1794 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1795 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
1798 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1800 goto err_pcpu_lists
;
1802 net
->ct
.slabname
= kasprintf(GFP_KERNEL
, "nf_conntrack_%p", net
);
1803 if (!net
->ct
.slabname
)
1806 net
->ct
.nf_conntrack_cachep
= kmem_cache_create(net
->ct
.slabname
,
1807 sizeof(struct nf_conn
), 0,
1808 SLAB_DESTROY_BY_RCU
, NULL
);
1809 if (!net
->ct
.nf_conntrack_cachep
) {
1810 printk(KERN_ERR
"Unable to create nf_conn slab cache\n");
1814 net
->ct
.htable_size
= nf_conntrack_htable_size
;
1815 net
->ct
.hash
= nf_ct_alloc_hashtable(&net
->ct
.htable_size
, 1);
1816 if (!net
->ct
.hash
) {
1817 printk(KERN_ERR
"Unable to create nf_conntrack_hash\n");
1820 ret
= nf_conntrack_expect_pernet_init(net
);
1823 ret
= nf_conntrack_acct_pernet_init(net
);
1826 ret
= nf_conntrack_tstamp_pernet_init(net
);
1829 ret
= nf_conntrack_ecache_pernet_init(net
);
1832 ret
= nf_conntrack_helper_pernet_init(net
);
1835 ret
= nf_conntrack_proto_pernet_init(net
);
1841 nf_conntrack_helper_pernet_fini(net
);
1843 nf_conntrack_ecache_pernet_fini(net
);
1845 nf_conntrack_tstamp_pernet_fini(net
);
1847 nf_conntrack_acct_pernet_fini(net
);
1849 nf_conntrack_expect_pernet_fini(net
);
1851 nf_ct_free_hashtable(net
->ct
.hash
, net
->ct
.htable_size
);
1853 kmem_cache_destroy(net
->ct
.nf_conntrack_cachep
);
1855 kfree(net
->ct
.slabname
);
1857 free_percpu(net
->ct
.stat
);
1859 free_percpu(net
->ct
.pcpu_lists
);