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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/types.h>
18 #include <linux/netfilter.h>
19 #include <linux/module.h>
20 #include <linux/sched.h>
21 #include <linux/skbuff.h>
22 #include <linux/proc_fs.h>
23 #include <linux/vmalloc.h>
24 #include <linux/stddef.h>
25 #include <linux/slab.h>
26 #include <linux/random.h>
27 #include <linux/jhash.h>
28 #include <linux/err.h>
29 #include <linux/percpu.h>
30 #include <linux/moduleparam.h>
31 #include <linux/notifier.h>
32 #include <linux/kernel.h>
33 #include <linux/netdevice.h>
34 #include <linux/socket.h>
36 #include <linux/nsproxy.h>
37 #include <linux/rculist_nulls.h>
39 #include <net/netfilter/nf_conntrack.h>
40 #include <net/netfilter/nf_conntrack_l3proto.h>
41 #include <net/netfilter/nf_conntrack_l4proto.h>
42 #include <net/netfilter/nf_conntrack_expect.h>
43 #include <net/netfilter/nf_conntrack_helper.h>
44 #include <net/netfilter/nf_conntrack_seqadj.h>
45 #include <net/netfilter/nf_conntrack_core.h>
46 #include <net/netfilter/nf_conntrack_extend.h>
47 #include <net/netfilter/nf_conntrack_acct.h>
48 #include <net/netfilter/nf_conntrack_ecache.h>
49 #include <net/netfilter/nf_conntrack_zones.h>
50 #include <net/netfilter/nf_conntrack_timestamp.h>
51 #include <net/netfilter/nf_conntrack_timeout.h>
52 #include <net/netfilter/nf_conntrack_labels.h>
53 #include <net/netfilter/nf_conntrack_synproxy.h>
54 #include <net/netfilter/nf_nat.h>
55 #include <net/netfilter/nf_nat_core.h>
56 #include <net/netfilter/nf_nat_helper.h>
57 #include <net/netns/hash.h>
59 #define NF_CONNTRACK_VERSION "0.5.0"
61 int (*nfnetlink_parse_nat_setup_hook
)(struct nf_conn
*ct
,
62 enum nf_nat_manip_type manip
,
63 const struct nlattr
*attr
) __read_mostly
;
64 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook
);
66 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks
[CONNTRACK_LOCKS
];
67 EXPORT_SYMBOL_GPL(nf_conntrack_locks
);
69 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(nf_conntrack_expect_lock
);
70 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock
);
72 struct hlist_nulls_head
*nf_conntrack_hash __read_mostly
;
73 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
75 static __read_mostly
struct kmem_cache
*nf_conntrack_cachep
;
76 static __read_mostly spinlock_t nf_conntrack_locks_all_lock
;
77 static __read_mostly seqcount_t nf_conntrack_generation
;
78 static __read_mostly
DEFINE_SPINLOCK(nf_conntrack_locks_all_lock
);
79 static __read_mostly
bool nf_conntrack_locks_all
;
81 void nf_conntrack_lock(spinlock_t
*lock
) __acquires(lock
)
84 while (unlikely(nf_conntrack_locks_all
)) {
86 spin_unlock_wait(&nf_conntrack_locks_all_lock
);
90 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
92 static void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
94 h1
%= CONNTRACK_LOCKS
;
95 h2
%= CONNTRACK_LOCKS
;
96 spin_unlock(&nf_conntrack_locks
[h1
]);
98 spin_unlock(&nf_conntrack_locks
[h2
]);
101 /* return true if we need to recompute hashes (in case hash table was resized) */
102 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
103 unsigned int h2
, unsigned int sequence
)
105 h1
%= CONNTRACK_LOCKS
;
106 h2
%= CONNTRACK_LOCKS
;
108 nf_conntrack_lock(&nf_conntrack_locks
[h1
]);
110 spin_lock_nested(&nf_conntrack_locks
[h2
],
111 SINGLE_DEPTH_NESTING
);
113 nf_conntrack_lock(&nf_conntrack_locks
[h2
]);
114 spin_lock_nested(&nf_conntrack_locks
[h1
],
115 SINGLE_DEPTH_NESTING
);
117 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
118 nf_conntrack_double_unlock(h1
, h2
);
124 static void nf_conntrack_all_lock(void)
128 spin_lock(&nf_conntrack_locks_all_lock
);
129 nf_conntrack_locks_all
= true;
131 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++) {
132 spin_unlock_wait(&nf_conntrack_locks
[i
]);
136 static void nf_conntrack_all_unlock(void)
138 nf_conntrack_locks_all
= false;
139 spin_unlock(&nf_conntrack_locks_all_lock
);
142 unsigned int nf_conntrack_htable_size __read_mostly
;
143 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
145 unsigned int nf_conntrack_max __read_mostly
;
146 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
148 DEFINE_PER_CPU(struct nf_conn
, nf_conntrack_untracked
);
149 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked
);
151 static unsigned int nf_conntrack_hash_rnd __read_mostly
;
153 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
,
154 const struct net
*net
)
159 get_random_once(&nf_conntrack_hash_rnd
, sizeof(nf_conntrack_hash_rnd
));
161 /* The direction must be ignored, so we hash everything up to the
162 * destination ports (which is a multiple of 4) and treat the last
163 * three bytes manually.
165 seed
= nf_conntrack_hash_rnd
^ net_hash_mix(net
);
166 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
167 return jhash2((u32
*)tuple
, n
, seed
^
168 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
169 tuple
->dst
.protonum
));
172 static u32
scale_hash(u32 hash
)
174 return reciprocal_scale(hash
, nf_conntrack_htable_size
);
177 static u32
__hash_conntrack(const struct net
*net
,
178 const struct nf_conntrack_tuple
*tuple
,
181 return reciprocal_scale(hash_conntrack_raw(tuple
, net
), size
);
184 static u32
hash_conntrack(const struct net
*net
,
185 const struct nf_conntrack_tuple
*tuple
)
187 return scale_hash(hash_conntrack_raw(tuple
, net
));
191 nf_ct_get_tuple(const struct sk_buff
*skb
,
193 unsigned int dataoff
,
197 struct nf_conntrack_tuple
*tuple
,
198 const struct nf_conntrack_l3proto
*l3proto
,
199 const struct nf_conntrack_l4proto
*l4proto
)
201 memset(tuple
, 0, sizeof(*tuple
));
203 tuple
->src
.l3num
= l3num
;
204 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
207 tuple
->dst
.protonum
= protonum
;
208 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
210 return l4proto
->pkt_to_tuple(skb
, dataoff
, net
, tuple
);
212 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
214 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
216 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
218 struct nf_conntrack_l3proto
*l3proto
;
219 struct nf_conntrack_l4proto
*l4proto
;
220 unsigned int protoff
;
226 l3proto
= __nf_ct_l3proto_find(l3num
);
227 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
228 if (ret
!= NF_ACCEPT
) {
233 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
235 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
,
241 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
244 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
245 const struct nf_conntrack_tuple
*orig
,
246 const struct nf_conntrack_l3proto
*l3proto
,
247 const struct nf_conntrack_l4proto
*l4proto
)
249 memset(inverse
, 0, sizeof(*inverse
));
251 inverse
->src
.l3num
= orig
->src
.l3num
;
252 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
255 inverse
->dst
.dir
= !orig
->dst
.dir
;
257 inverse
->dst
.protonum
= orig
->dst
.protonum
;
258 return l4proto
->invert_tuple(inverse
, orig
);
260 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
263 clean_from_lists(struct nf_conn
*ct
)
265 pr_debug("clean_from_lists(%p)\n", ct
);
266 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
267 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
269 /* Destroy all pending expectations */
270 nf_ct_remove_expectations(ct
);
273 /* must be called with local_bh_disable */
274 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
276 struct ct_pcpu
*pcpu
;
278 /* add this conntrack to the (per cpu) dying list */
279 ct
->cpu
= smp_processor_id();
280 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
282 spin_lock(&pcpu
->lock
);
283 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
285 spin_unlock(&pcpu
->lock
);
288 /* must be called with local_bh_disable */
289 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
291 struct ct_pcpu
*pcpu
;
293 /* add this conntrack to the (per cpu) unconfirmed list */
294 ct
->cpu
= smp_processor_id();
295 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
297 spin_lock(&pcpu
->lock
);
298 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
300 spin_unlock(&pcpu
->lock
);
303 /* must be called with local_bh_disable */
304 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
306 struct ct_pcpu
*pcpu
;
308 /* We overload first tuple to link into unconfirmed or dying list.*/
309 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
311 spin_lock(&pcpu
->lock
);
312 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
313 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
314 spin_unlock(&pcpu
->lock
);
317 /* Released via destroy_conntrack() */
318 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
319 const struct nf_conntrack_zone
*zone
,
322 struct nf_conn
*tmpl
;
324 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
328 tmpl
->status
= IPS_TEMPLATE
;
329 write_pnet(&tmpl
->ct_net
, net
);
330 nf_ct_zone_add(tmpl
, zone
);
331 atomic_set(&tmpl
->ct_general
.use
, 0);
335 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
337 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
339 nf_ct_ext_destroy(tmpl
);
340 nf_ct_ext_free(tmpl
);
343 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
346 destroy_conntrack(struct nf_conntrack
*nfct
)
348 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
349 struct net
*net
= nf_ct_net(ct
);
350 struct nf_conntrack_l4proto
*l4proto
;
352 pr_debug("destroy_conntrack(%p)\n", ct
);
353 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
354 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
356 if (unlikely(nf_ct_is_template(ct
))) {
361 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
362 if (l4proto
->destroy
)
363 l4proto
->destroy(ct
);
368 /* Expectations will have been removed in clean_from_lists,
369 * except TFTP can create an expectation on the first packet,
370 * before connection is in the list, so we need to clean here,
373 nf_ct_remove_expectations(ct
);
375 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
377 NF_CT_STAT_INC(net
, delete);
381 nf_ct_put(ct
->master
);
383 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
384 nf_conntrack_free(ct
);
387 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
389 struct net
*net
= nf_ct_net(ct
);
390 unsigned int hash
, reply_hash
;
391 unsigned int sequence
;
393 nf_ct_helper_destroy(ct
);
397 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
398 hash
= hash_conntrack(net
,
399 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
400 reply_hash
= hash_conntrack(net
,
401 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
402 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
404 clean_from_lists(ct
);
405 nf_conntrack_double_unlock(hash
, reply_hash
);
407 nf_ct_add_to_dying_list(ct
);
409 NF_CT_STAT_INC(net
, delete_list
);
413 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
415 struct nf_conn_tstamp
*tstamp
;
417 tstamp
= nf_conn_tstamp_find(ct
);
418 if (tstamp
&& tstamp
->stop
== 0)
419 tstamp
->stop
= ktime_get_real_ns();
421 if (nf_ct_is_dying(ct
))
424 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
425 portid
, report
) < 0) {
426 /* destroy event was not delivered */
427 nf_ct_delete_from_lists(ct
);
428 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
432 nf_conntrack_ecache_work(nf_ct_net(ct
));
433 set_bit(IPS_DYING_BIT
, &ct
->status
);
435 nf_ct_delete_from_lists(ct
);
439 EXPORT_SYMBOL_GPL(nf_ct_delete
);
441 static void death_by_timeout(unsigned long ul_conntrack
)
443 nf_ct_delete((struct nf_conn
*)ul_conntrack
, 0, 0);
447 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
448 const struct nf_conntrack_tuple
*tuple
,
449 const struct nf_conntrack_zone
*zone
,
450 const struct net
*net
)
452 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
454 /* A conntrack can be recreated with the equal tuple,
455 * so we need to check that the conntrack is confirmed
457 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
458 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
459 nf_ct_is_confirmed(ct
) &&
460 net_eq(net
, nf_ct_net(ct
));
465 * - Caller must take a reference on returned object
466 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
468 static struct nf_conntrack_tuple_hash
*
469 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
470 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
472 struct nf_conntrack_tuple_hash
*h
;
473 struct hlist_nulls_head
*ct_hash
;
474 struct hlist_nulls_node
*n
;
475 unsigned int bucket
, sequence
;
479 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
480 bucket
= scale_hash(hash
);
481 ct_hash
= nf_conntrack_hash
;
482 } while (read_seqcount_retry(&nf_conntrack_generation
, sequence
));
484 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[bucket
], hnnode
) {
485 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
486 NF_CT_STAT_INC_ATOMIC(net
, found
);
489 NF_CT_STAT_INC_ATOMIC(net
, searched
);
492 * if the nulls value we got at the end of this lookup is
493 * not the expected one, we must restart lookup.
494 * We probably met an item that was moved to another chain.
496 if (get_nulls_value(n
) != bucket
) {
497 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
504 /* Find a connection corresponding to a tuple. */
505 static struct nf_conntrack_tuple_hash
*
506 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
507 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
509 struct nf_conntrack_tuple_hash
*h
;
514 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
516 ct
= nf_ct_tuplehash_to_ctrack(h
);
517 if (unlikely(nf_ct_is_dying(ct
) ||
518 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
521 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
, net
))) {
532 struct nf_conntrack_tuple_hash
*
533 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
534 const struct nf_conntrack_tuple
*tuple
)
536 return __nf_conntrack_find_get(net
, zone
, tuple
,
537 hash_conntrack_raw(tuple
, net
));
539 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
541 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
543 unsigned int reply_hash
)
545 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
546 &nf_conntrack_hash
[hash
]);
547 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
548 &nf_conntrack_hash
[reply_hash
]);
552 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
554 const struct nf_conntrack_zone
*zone
;
555 struct net
*net
= nf_ct_net(ct
);
556 unsigned int hash
, reply_hash
;
557 struct nf_conntrack_tuple_hash
*h
;
558 struct hlist_nulls_node
*n
;
559 unsigned int sequence
;
561 zone
= nf_ct_zone(ct
);
565 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
566 hash
= hash_conntrack(net
,
567 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
568 reply_hash
= hash_conntrack(net
,
569 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
570 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
572 /* See if there's one in the list already, including reverse */
573 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
574 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
578 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
579 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
583 add_timer(&ct
->timeout
);
585 /* The caller holds a reference to this object */
586 atomic_set(&ct
->ct_general
.use
, 2);
587 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
588 nf_conntrack_double_unlock(hash
, reply_hash
);
589 NF_CT_STAT_INC(net
, insert
);
594 nf_conntrack_double_unlock(hash
, reply_hash
);
595 NF_CT_STAT_INC(net
, insert_failed
);
599 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
601 static inline void nf_ct_acct_update(struct nf_conn
*ct
,
602 enum ip_conntrack_info ctinfo
,
605 struct nf_conn_acct
*acct
;
607 acct
= nf_conn_acct_find(ct
);
609 struct nf_conn_counter
*counter
= acct
->counter
;
611 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
612 atomic64_add(len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
616 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
617 const struct nf_conn
*loser_ct
)
619 struct nf_conn_acct
*acct
;
621 acct
= nf_conn_acct_find(loser_ct
);
623 struct nf_conn_counter
*counter
= acct
->counter
;
626 /* u32 should be fine since we must have seen one packet. */
627 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
628 nf_ct_acct_update(ct
, ctinfo
, bytes
);
632 /* Resolve race on insertion if this protocol allows this. */
633 static int nf_ct_resolve_clash(struct net
*net
, struct sk_buff
*skb
,
634 enum ip_conntrack_info ctinfo
,
635 struct nf_conntrack_tuple_hash
*h
)
637 /* This is the conntrack entry already in hashes that won race. */
638 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
639 struct nf_conntrack_l4proto
*l4proto
;
641 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
642 if (l4proto
->allow_clash
&&
643 !nf_ct_is_dying(ct
) &&
644 atomic_inc_not_zero(&ct
->ct_general
.use
)) {
645 nf_ct_acct_merge(ct
, ctinfo
, (struct nf_conn
*)skb
->nfct
);
646 nf_conntrack_put(skb
->nfct
);
647 /* Assign conntrack already in hashes to this skbuff. Don't
648 * modify skb->nfctinfo to ensure consistent stateful filtering.
650 skb
->nfct
= &ct
->ct_general
;
653 NF_CT_STAT_INC(net
, drop
);
657 /* Confirm a connection given skb; places it in hash table */
659 __nf_conntrack_confirm(struct sk_buff
*skb
)
661 const struct nf_conntrack_zone
*zone
;
662 unsigned int hash
, reply_hash
;
663 struct nf_conntrack_tuple_hash
*h
;
665 struct nf_conn_help
*help
;
666 struct nf_conn_tstamp
*tstamp
;
667 struct hlist_nulls_node
*n
;
668 enum ip_conntrack_info ctinfo
;
670 unsigned int sequence
;
673 ct
= nf_ct_get(skb
, &ctinfo
);
676 /* ipt_REJECT uses nf_conntrack_attach to attach related
677 ICMP/TCP RST packets in other direction. Actual packet
678 which created connection will be IP_CT_NEW or for an
679 expected connection, IP_CT_RELATED. */
680 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
683 zone
= nf_ct_zone(ct
);
687 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
688 /* reuse the hash saved before */
689 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
690 hash
= scale_hash(hash
);
691 reply_hash
= hash_conntrack(net
,
692 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
694 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
696 /* We're not in hash table, and we refuse to set up related
697 * connections for unconfirmed conns. But packet copies and
698 * REJECT will give spurious warnings here.
700 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
702 /* No external references means no one else could have
705 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
706 pr_debug("Confirming conntrack %p\n", ct
);
707 /* We have to check the DYING flag after unlink to prevent
708 * a race against nf_ct_get_next_corpse() possibly called from
709 * user context, else we insert an already 'dead' hash, blocking
710 * further use of that particular connection -JM.
712 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
714 if (unlikely(nf_ct_is_dying(ct
))) {
715 nf_ct_add_to_dying_list(ct
);
719 /* See if there's one in the list already, including reverse:
720 NAT could have grabbed it without realizing, since we're
721 not in the hash. If there is, we lost race. */
722 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
723 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
727 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
728 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
732 /* Timer relative to confirmation time, not original
733 setting time, otherwise we'd get timer wrap in
734 weird delay cases. */
735 ct
->timeout
.expires
+= jiffies
;
736 add_timer(&ct
->timeout
);
737 atomic_inc(&ct
->ct_general
.use
);
738 ct
->status
|= IPS_CONFIRMED
;
740 /* set conntrack timestamp, if enabled. */
741 tstamp
= nf_conn_tstamp_find(ct
);
743 if (skb
->tstamp
.tv64
== 0)
744 __net_timestamp(skb
);
746 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
748 /* Since the lookup is lockless, hash insertion must be done after
749 * starting the timer and setting the CONFIRMED bit. The RCU barriers
750 * guarantee that no other CPU can find the conntrack before the above
751 * stores are visible.
753 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
754 nf_conntrack_double_unlock(hash
, reply_hash
);
755 NF_CT_STAT_INC(net
, insert
);
758 help
= nfct_help(ct
);
759 if (help
&& help
->helper
)
760 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
762 nf_conntrack_event_cache(master_ct(ct
) ?
763 IPCT_RELATED
: IPCT_NEW
, ct
);
767 nf_ct_add_to_dying_list(ct
);
768 ret
= nf_ct_resolve_clash(net
, skb
, ctinfo
, h
);
770 nf_conntrack_double_unlock(hash
, reply_hash
);
771 NF_CT_STAT_INC(net
, insert_failed
);
775 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
777 /* Returns true if a connection correspondings to the tuple (required
780 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
781 const struct nf_conn
*ignored_conntrack
)
783 struct net
*net
= nf_ct_net(ignored_conntrack
);
784 const struct nf_conntrack_zone
*zone
;
785 struct nf_conntrack_tuple_hash
*h
;
786 struct hlist_nulls_head
*ct_hash
;
787 unsigned int hash
, sequence
;
788 struct hlist_nulls_node
*n
;
791 zone
= nf_ct_zone(ignored_conntrack
);
795 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
796 hash
= hash_conntrack(net
, tuple
);
797 ct_hash
= nf_conntrack_hash
;
798 } while (read_seqcount_retry(&nf_conntrack_generation
, sequence
));
800 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
801 ct
= nf_ct_tuplehash_to_ctrack(h
);
802 if (ct
!= ignored_conntrack
&&
803 nf_ct_key_equal(h
, tuple
, zone
, net
)) {
804 NF_CT_STAT_INC_ATOMIC(net
, found
);
808 NF_CT_STAT_INC_ATOMIC(net
, searched
);
814 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
816 #define NF_CT_EVICTION_RANGE 8
818 /* There's a small race here where we may free a just-assured
819 connection. Too bad: we're in trouble anyway. */
820 static noinline
int early_drop(struct net
*net
, unsigned int _hash
)
822 /* Use oldest entry, which is roughly LRU */
823 struct nf_conntrack_tuple_hash
*h
;
825 struct hlist_nulls_node
*n
;
826 unsigned int i
, hash
, sequence
;
827 struct nf_conn
*ct
= NULL
;
835 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
836 for (; i
< NF_CT_EVICTION_RANGE
; i
++) {
837 hash
= scale_hash(_hash
++);
838 lockp
= &nf_conntrack_locks
[hash
% CONNTRACK_LOCKS
];
839 nf_conntrack_lock(lockp
);
840 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
844 hlist_nulls_for_each_entry_rcu(h
, n
, &nf_conntrack_hash
[hash
],
846 tmp
= nf_ct_tuplehash_to_ctrack(h
);
848 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
849 !net_eq(nf_ct_net(tmp
), net
) ||
853 if (atomic_inc_not_zero(&tmp
->ct_general
.use
)) {
869 /* kill only if in same netns -- might have moved due to
870 * SLAB_DESTROY_BY_RCU rules
872 if (net_eq(nf_ct_net(ct
), net
) && del_timer(&ct
->timeout
)) {
873 if (nf_ct_delete(ct
, 0, 0)) {
874 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
883 static struct nf_conn
*
884 __nf_conntrack_alloc(struct net
*net
,
885 const struct nf_conntrack_zone
*zone
,
886 const struct nf_conntrack_tuple
*orig
,
887 const struct nf_conntrack_tuple
*repl
,
892 /* We don't want any race condition at early drop stage */
893 atomic_inc(&net
->ct
.count
);
895 if (nf_conntrack_max
&&
896 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
897 if (!early_drop(net
, hash
)) {
898 atomic_dec(&net
->ct
.count
);
899 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
900 return ERR_PTR(-ENOMEM
);
905 * Do not use kmem_cache_zalloc(), as this cache uses
906 * SLAB_DESTROY_BY_RCU.
908 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
912 spin_lock_init(&ct
->lock
);
913 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
914 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
915 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
916 /* save hash for reusing when confirming */
917 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
919 /* Don't set timer yet: wait for confirmation */
920 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
921 write_pnet(&ct
->ct_net
, net
);
922 memset(&ct
->__nfct_init_offset
[0], 0,
923 offsetof(struct nf_conn
, proto
) -
924 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
926 nf_ct_zone_add(ct
, zone
);
928 /* Because we use RCU lookups, we set ct_general.use to zero before
929 * this is inserted in any list.
931 atomic_set(&ct
->ct_general
.use
, 0);
934 atomic_dec(&net
->ct
.count
);
935 return ERR_PTR(-ENOMEM
);
938 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
939 const struct nf_conntrack_zone
*zone
,
940 const struct nf_conntrack_tuple
*orig
,
941 const struct nf_conntrack_tuple
*repl
,
944 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
946 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
948 void nf_conntrack_free(struct nf_conn
*ct
)
950 struct net
*net
= nf_ct_net(ct
);
952 /* A freed object has refcnt == 0, that's
953 * the golden rule for SLAB_DESTROY_BY_RCU
955 NF_CT_ASSERT(atomic_read(&ct
->ct_general
.use
) == 0);
957 nf_ct_ext_destroy(ct
);
959 kmem_cache_free(nf_conntrack_cachep
, ct
);
960 smp_mb__before_atomic();
961 atomic_dec(&net
->ct
.count
);
963 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
966 /* Allocate a new conntrack: we return -ENOMEM if classification
967 failed due to stress. Otherwise it really is unclassifiable. */
968 static struct nf_conntrack_tuple_hash
*
969 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
970 const struct nf_conntrack_tuple
*tuple
,
971 struct nf_conntrack_l3proto
*l3proto
,
972 struct nf_conntrack_l4proto
*l4proto
,
974 unsigned int dataoff
, u32 hash
)
977 struct nf_conn_help
*help
;
978 struct nf_conntrack_tuple repl_tuple
;
979 struct nf_conntrack_ecache
*ecache
;
980 struct nf_conntrack_expect
*exp
= NULL
;
981 const struct nf_conntrack_zone
*zone
;
982 struct nf_conn_timeout
*timeout_ext
;
983 struct nf_conntrack_zone tmp
;
984 unsigned int *timeouts
;
986 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
987 pr_debug("Can't invert tuple.\n");
991 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
992 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
995 return (struct nf_conntrack_tuple_hash
*)ct
;
997 if (tmpl
&& nfct_synproxy(tmpl
)) {
998 nfct_seqadj_ext_add(ct
);
999 nfct_synproxy_ext_add(ct
);
1002 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1004 timeouts
= nf_ct_timeout_data(timeout_ext
);
1005 if (unlikely(!timeouts
))
1006 timeouts
= l4proto
->get_timeouts(net
);
1008 timeouts
= l4proto
->get_timeouts(net
);
1011 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
1012 nf_conntrack_free(ct
);
1013 pr_debug("can't track with proto module\n");
1018 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1021 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1022 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1023 nf_ct_labels_ext_add(ct
);
1025 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1026 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1027 ecache
? ecache
->expmask
: 0,
1031 if (net
->ct
.expect_count
) {
1032 spin_lock(&nf_conntrack_expect_lock
);
1033 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1035 pr_debug("expectation arrives ct=%p exp=%p\n",
1037 /* Welcome, Mr. Bond. We've been expecting you... */
1038 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1039 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1040 ct
->master
= exp
->master
;
1042 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
1045 rcu_assign_pointer(help
->helper
, exp
->helper
);
1048 #ifdef CONFIG_NF_CONNTRACK_MARK
1049 ct
->mark
= exp
->master
->mark
;
1051 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1052 ct
->secmark
= exp
->master
->secmark
;
1054 NF_CT_STAT_INC(net
, expect_new
);
1056 spin_unlock(&nf_conntrack_expect_lock
);
1059 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1060 NF_CT_STAT_INC(net
, new);
1063 /* Now it is inserted into the unconfirmed list, bump refcount */
1064 nf_conntrack_get(&ct
->ct_general
);
1065 nf_ct_add_to_unconfirmed_list(ct
);
1071 exp
->expectfn(ct
, exp
);
1072 nf_ct_expect_put(exp
);
1075 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1078 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1079 static inline struct nf_conn
*
1080 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1081 struct sk_buff
*skb
,
1082 unsigned int dataoff
,
1085 struct nf_conntrack_l3proto
*l3proto
,
1086 struct nf_conntrack_l4proto
*l4proto
,
1088 enum ip_conntrack_info
*ctinfo
)
1090 const struct nf_conntrack_zone
*zone
;
1091 struct nf_conntrack_tuple tuple
;
1092 struct nf_conntrack_tuple_hash
*h
;
1093 struct nf_conntrack_zone tmp
;
1097 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1098 dataoff
, l3num
, protonum
, net
, &tuple
, l3proto
,
1100 pr_debug("Can't get tuple\n");
1104 /* look for tuple match */
1105 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1106 hash
= hash_conntrack_raw(&tuple
, net
);
1107 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1109 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1110 skb
, dataoff
, hash
);
1116 ct
= nf_ct_tuplehash_to_ctrack(h
);
1118 /* It exists; we have (non-exclusive) reference. */
1119 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1120 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1121 /* Please set reply bit if this packet OK */
1124 /* Once we've had two way comms, always ESTABLISHED. */
1125 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1126 pr_debug("normal packet for %p\n", ct
);
1127 *ctinfo
= IP_CT_ESTABLISHED
;
1128 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1129 pr_debug("related packet for %p\n", ct
);
1130 *ctinfo
= IP_CT_RELATED
;
1132 pr_debug("new packet for %p\n", ct
);
1133 *ctinfo
= IP_CT_NEW
;
1137 skb
->nfct
= &ct
->ct_general
;
1138 skb
->nfctinfo
= *ctinfo
;
1143 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1144 struct sk_buff
*skb
)
1146 struct nf_conn
*ct
, *tmpl
= NULL
;
1147 enum ip_conntrack_info ctinfo
;
1148 struct nf_conntrack_l3proto
*l3proto
;
1149 struct nf_conntrack_l4proto
*l4proto
;
1150 unsigned int *timeouts
;
1151 unsigned int dataoff
;
1157 /* Previously seen (loopback or untracked)? Ignore. */
1158 tmpl
= (struct nf_conn
*)skb
->nfct
;
1159 if (!nf_ct_is_template(tmpl
)) {
1160 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1166 /* rcu_read_lock()ed by nf_hook_slow */
1167 l3proto
= __nf_ct_l3proto_find(pf
);
1168 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1169 &dataoff
, &protonum
);
1171 pr_debug("not prepared to track yet or error occurred\n");
1172 NF_CT_STAT_INC_ATOMIC(net
, error
);
1173 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1178 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1180 /* It may be an special packet, error, unclean...
1181 * inverse of the return code tells to the netfilter
1182 * core what to do with the packet. */
1183 if (l4proto
->error
!= NULL
) {
1184 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
1187 NF_CT_STAT_INC_ATOMIC(net
, error
);
1188 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1192 /* ICMP[v6] protocol trackers may assign one conntrack. */
1197 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1198 l3proto
, l4proto
, &set_reply
, &ctinfo
);
1200 /* Not valid part of a connection */
1201 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1207 /* Too stressed to deal. */
1208 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1213 NF_CT_ASSERT(skb
->nfct
);
1215 /* Decide what timeout policy we want to apply to this flow. */
1216 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1218 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1220 /* Invalid: inverse of the return code tells
1221 * the netfilter core what to do */
1222 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1223 nf_conntrack_put(skb
->nfct
);
1225 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1226 if (ret
== -NF_DROP
)
1227 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1232 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1233 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1236 /* Special case: we have to repeat this hook, assign the
1237 * template again to this packet. We assume that this packet
1238 * has no conntrack assigned. This is used by nf_ct_tcp. */
1239 if (ret
== NF_REPEAT
)
1240 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
1247 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1249 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1250 const struct nf_conntrack_tuple
*orig
)
1255 ret
= nf_ct_invert_tuple(inverse
, orig
,
1256 __nf_ct_l3proto_find(orig
->src
.l3num
),
1257 __nf_ct_l4proto_find(orig
->src
.l3num
,
1258 orig
->dst
.protonum
));
1262 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1264 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1265 implicitly racy: see __nf_conntrack_confirm */
1266 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1267 const struct nf_conntrack_tuple
*newreply
)
1269 struct nf_conn_help
*help
= nfct_help(ct
);
1271 /* Should be unconfirmed, so not in hash table yet */
1272 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1274 pr_debug("Altering reply tuple of %p to ", ct
);
1275 nf_ct_dump_tuple(newreply
);
1277 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1278 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1282 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1285 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1287 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1288 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1289 enum ip_conntrack_info ctinfo
,
1290 const struct sk_buff
*skb
,
1291 unsigned long extra_jiffies
,
1294 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1297 /* Only update if this is not a fixed timeout */
1298 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1301 /* If not in hash table, timer will not be active yet */
1302 if (!nf_ct_is_confirmed(ct
)) {
1303 ct
->timeout
.expires
= extra_jiffies
;
1305 unsigned long newtime
= jiffies
+ extra_jiffies
;
1307 /* Only update the timeout if the new timeout is at least
1308 HZ jiffies from the old timeout. Need del_timer for race
1309 avoidance (may already be dying). */
1310 if (newtime
- ct
->timeout
.expires
>= HZ
)
1311 mod_timer_pending(&ct
->timeout
, newtime
);
1316 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1318 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1320 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1321 enum ip_conntrack_info ctinfo
,
1322 const struct sk_buff
*skb
,
1326 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1328 if (del_timer(&ct
->timeout
)) {
1329 ct
->timeout
.function((unsigned long)ct
);
1334 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1336 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1338 #include <linux/netfilter/nfnetlink.h>
1339 #include <linux/netfilter/nfnetlink_conntrack.h>
1340 #include <linux/mutex.h>
1342 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1343 * in ip_conntrack_core, since we don't want the protocols to autoload
1344 * or depend on ctnetlink */
1345 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1346 const struct nf_conntrack_tuple
*tuple
)
1348 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1349 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1350 goto nla_put_failure
;
1356 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1358 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1359 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1360 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1362 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1364 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1365 struct nf_conntrack_tuple
*t
)
1367 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1370 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1371 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1375 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1377 int nf_ct_port_nlattr_tuple_size(void)
1379 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1381 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1384 /* Used by ipt_REJECT and ip6t_REJECT. */
1385 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1388 enum ip_conntrack_info ctinfo
;
1390 /* This ICMP is in reverse direction to the packet which caused it */
1391 ct
= nf_ct_get(skb
, &ctinfo
);
1392 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1393 ctinfo
= IP_CT_RELATED_REPLY
;
1395 ctinfo
= IP_CT_RELATED
;
1397 /* Attach to new skbuff, and increment count */
1398 nskb
->nfct
= &ct
->ct_general
;
1399 nskb
->nfctinfo
= ctinfo
;
1400 nf_conntrack_get(nskb
->nfct
);
1403 /* Bring out ya dead! */
1404 static struct nf_conn
*
1405 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1406 void *data
, unsigned int *bucket
)
1408 struct nf_conntrack_tuple_hash
*h
;
1410 struct hlist_nulls_node
*n
;
1414 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1415 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1417 nf_conntrack_lock(lockp
);
1418 if (*bucket
< nf_conntrack_htable_size
) {
1419 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
1420 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1422 ct
= nf_ct_tuplehash_to_ctrack(h
);
1423 if (net_eq(nf_ct_net(ct
), net
) &&
1433 for_each_possible_cpu(cpu
) {
1434 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1436 spin_lock_bh(&pcpu
->lock
);
1437 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1438 ct
= nf_ct_tuplehash_to_ctrack(h
);
1440 set_bit(IPS_DYING_BIT
, &ct
->status
);
1442 spin_unlock_bh(&pcpu
->lock
);
1447 atomic_inc(&ct
->ct_general
.use
);
1453 void nf_ct_iterate_cleanup(struct net
*net
,
1454 int (*iter
)(struct nf_conn
*i
, void *data
),
1455 void *data
, u32 portid
, int report
)
1458 unsigned int bucket
= 0;
1462 if (atomic_read(&net
->ct
.count
) == 0)
1465 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1466 /* Time to push up daises... */
1467 if (del_timer(&ct
->timeout
))
1468 nf_ct_delete(ct
, portid
, report
);
1470 /* ... else the timer will get him soon. */
1476 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1478 static int kill_all(struct nf_conn
*i
, void *data
)
1483 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1485 if (is_vmalloc_addr(hash
))
1488 free_pages((unsigned long)hash
,
1489 get_order(sizeof(struct hlist_head
) * size
));
1491 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1493 static int untrack_refs(void)
1497 for_each_possible_cpu(cpu
) {
1498 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1500 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1505 void nf_conntrack_cleanup_start(void)
1507 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1510 void nf_conntrack_cleanup_end(void)
1512 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1513 while (untrack_refs() > 0)
1516 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1518 nf_conntrack_proto_fini();
1519 nf_conntrack_seqadj_fini();
1520 nf_conntrack_labels_fini();
1521 nf_conntrack_helper_fini();
1522 nf_conntrack_timeout_fini();
1523 nf_conntrack_ecache_fini();
1524 nf_conntrack_tstamp_fini();
1525 nf_conntrack_acct_fini();
1526 nf_conntrack_expect_fini();
1530 * Mishearing the voices in his head, our hero wonders how he's
1531 * supposed to kill the mall.
1533 void nf_conntrack_cleanup_net(struct net
*net
)
1537 list_add(&net
->exit_list
, &single
);
1538 nf_conntrack_cleanup_net_list(&single
);
1541 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1547 * This makes sure all current packets have passed through
1548 * netfilter framework. Roll on, two-stage module
1554 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1555 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1556 if (atomic_read(&net
->ct
.count
) != 0)
1561 goto i_see_dead_people
;
1564 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1565 nf_conntrack_proto_pernet_fini(net
);
1566 nf_conntrack_helper_pernet_fini(net
);
1567 nf_conntrack_ecache_pernet_fini(net
);
1568 nf_conntrack_tstamp_pernet_fini(net
);
1569 nf_conntrack_acct_pernet_fini(net
);
1570 nf_conntrack_expect_pernet_fini(net
);
1571 free_percpu(net
->ct
.stat
);
1572 free_percpu(net
->ct
.pcpu_lists
);
1576 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1578 struct hlist_nulls_head
*hash
;
1579 unsigned int nr_slots
, i
;
1582 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1583 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1584 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1585 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1591 for (i
= 0; i
< nr_slots
; i
++)
1592 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1596 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1598 int nf_conntrack_hash_resize(unsigned int hashsize
)
1601 unsigned int old_size
;
1602 struct hlist_nulls_head
*hash
, *old_hash
;
1603 struct nf_conntrack_tuple_hash
*h
;
1609 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1613 old_size
= nf_conntrack_htable_size
;
1614 if (old_size
== hashsize
) {
1615 nf_ct_free_hashtable(hash
, hashsize
);
1620 nf_conntrack_all_lock();
1621 write_seqcount_begin(&nf_conntrack_generation
);
1623 /* Lookups in the old hash might happen in parallel, which means we
1624 * might get false negatives during connection lookup. New connections
1625 * created because of a false negative won't make it into the hash
1626 * though since that required taking the locks.
1629 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1630 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
1631 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
1632 struct nf_conntrack_tuple_hash
, hnnode
);
1633 ct
= nf_ct_tuplehash_to_ctrack(h
);
1634 hlist_nulls_del_rcu(&h
->hnnode
);
1635 bucket
= __hash_conntrack(nf_ct_net(ct
),
1636 &h
->tuple
, hashsize
);
1637 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1640 old_size
= nf_conntrack_htable_size
;
1641 old_hash
= nf_conntrack_hash
;
1643 nf_conntrack_hash
= hash
;
1644 nf_conntrack_htable_size
= hashsize
;
1646 write_seqcount_end(&nf_conntrack_generation
);
1647 nf_conntrack_all_unlock();
1651 nf_ct_free_hashtable(old_hash
, old_size
);
1655 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1657 unsigned int hashsize
;
1660 if (current
->nsproxy
->net_ns
!= &init_net
)
1663 /* On boot, we can set this without any fancy locking. */
1664 if (!nf_conntrack_htable_size
)
1665 return param_set_uint(val
, kp
);
1667 rc
= kstrtouint(val
, 0, &hashsize
);
1671 return nf_conntrack_hash_resize(hashsize
);
1673 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1675 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1676 &nf_conntrack_htable_size
, 0600);
1678 void nf_ct_untracked_status_or(unsigned long bits
)
1682 for_each_possible_cpu(cpu
)
1683 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1685 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1687 int nf_conntrack_init_start(void)
1693 seqcount_init(&nf_conntrack_generation
);
1695 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
1696 spin_lock_init(&nf_conntrack_locks
[i
]);
1698 if (!nf_conntrack_htable_size
) {
1699 /* Idea from tcp.c: use 1/16384 of memory.
1700 * On i386: 32MB machine has 512 buckets.
1701 * >= 1GB machines have 16384 buckets.
1702 * >= 4GB machines have 65536 buckets.
1704 nf_conntrack_htable_size
1705 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1706 / sizeof(struct hlist_head
));
1707 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
1708 nf_conntrack_htable_size
= 65536;
1709 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1710 nf_conntrack_htable_size
= 16384;
1711 if (nf_conntrack_htable_size
< 32)
1712 nf_conntrack_htable_size
= 32;
1714 /* Use a max. factor of four by default to get the same max as
1715 * with the old struct list_heads. When a table size is given
1716 * we use the old value of 8 to avoid reducing the max.
1721 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
1722 if (!nf_conntrack_hash
)
1725 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1727 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1728 sizeof(struct nf_conn
), 0,
1729 SLAB_DESTROY_BY_RCU
| SLAB_HWCACHE_ALIGN
, NULL
);
1730 if (!nf_conntrack_cachep
)
1733 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1734 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1737 ret
= nf_conntrack_expect_init();
1741 ret
= nf_conntrack_acct_init();
1745 ret
= nf_conntrack_tstamp_init();
1749 ret
= nf_conntrack_ecache_init();
1753 ret
= nf_conntrack_timeout_init();
1757 ret
= nf_conntrack_helper_init();
1761 ret
= nf_conntrack_labels_init();
1765 ret
= nf_conntrack_seqadj_init();
1769 ret
= nf_conntrack_proto_init();
1773 /* Set up fake conntrack: to never be deleted, not in any hashes */
1774 for_each_possible_cpu(cpu
) {
1775 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1776 write_pnet(&ct
->ct_net
, &init_net
);
1777 atomic_set(&ct
->ct_general
.use
, 1);
1779 /* - and look it like as a confirmed connection */
1780 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1784 nf_conntrack_seqadj_fini();
1786 nf_conntrack_labels_fini();
1788 nf_conntrack_helper_fini();
1790 nf_conntrack_timeout_fini();
1792 nf_conntrack_ecache_fini();
1794 nf_conntrack_tstamp_fini();
1796 nf_conntrack_acct_fini();
1798 nf_conntrack_expect_fini();
1800 kmem_cache_destroy(nf_conntrack_cachep
);
1802 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1806 void nf_conntrack_init_end(void)
1808 /* For use by REJECT target */
1809 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1810 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1814 * We need to use special "null" values, not used in hash table
1816 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1817 #define DYING_NULLS_VAL ((1<<30)+1)
1818 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1820 int nf_conntrack_init_net(struct net
*net
)
1825 atomic_set(&net
->ct
.count
, 0);
1827 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
1828 if (!net
->ct
.pcpu_lists
)
1831 for_each_possible_cpu(cpu
) {
1832 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1834 spin_lock_init(&pcpu
->lock
);
1835 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1836 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
1839 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1841 goto err_pcpu_lists
;
1843 ret
= nf_conntrack_expect_pernet_init(net
);
1846 ret
= nf_conntrack_acct_pernet_init(net
);
1849 ret
= nf_conntrack_tstamp_pernet_init(net
);
1852 ret
= nf_conntrack_ecache_pernet_init(net
);
1855 ret
= nf_conntrack_helper_pernet_init(net
);
1858 ret
= nf_conntrack_proto_pernet_init(net
);
1864 nf_conntrack_helper_pernet_fini(net
);
1866 nf_conntrack_ecache_pernet_fini(net
);
1868 nf_conntrack_tstamp_pernet_fini(net
);
1870 nf_conntrack_acct_pernet_fini(net
);
1872 nf_conntrack_expect_pernet_fini(net
);
1874 free_percpu(net
->ct
.stat
);
1876 free_percpu(net
->ct
.pcpu_lists
);