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 struct conntrack_gc_work
{
76 struct delayed_work dwork
;
82 static __read_mostly
struct kmem_cache
*nf_conntrack_cachep
;
83 static __read_mostly spinlock_t nf_conntrack_locks_all_lock
;
84 static __read_mostly
DEFINE_SPINLOCK(nf_conntrack_locks_all_lock
);
85 static __read_mostly
bool nf_conntrack_locks_all
;
87 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
88 #define GC_MAX_BUCKETS_DIV 64u
89 /* upper bound of scan intervals */
90 #define GC_INTERVAL_MAX (2 * HZ)
91 /* maximum conntracks to evict per gc run */
92 #define GC_MAX_EVICTS 256u
94 static struct conntrack_gc_work conntrack_gc_work
;
96 void nf_conntrack_lock(spinlock_t
*lock
) __acquires(lock
)
99 while (unlikely(nf_conntrack_locks_all
)) {
103 * Order the 'nf_conntrack_locks_all' load vs. the
104 * spin_unlock_wait() loads below, to ensure
105 * that 'nf_conntrack_locks_all_lock' is indeed held:
107 smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
108 spin_unlock_wait(&nf_conntrack_locks_all_lock
);
112 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
114 static void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
116 h1
%= CONNTRACK_LOCKS
;
117 h2
%= CONNTRACK_LOCKS
;
118 spin_unlock(&nf_conntrack_locks
[h1
]);
120 spin_unlock(&nf_conntrack_locks
[h2
]);
123 /* return true if we need to recompute hashes (in case hash table was resized) */
124 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
125 unsigned int h2
, unsigned int sequence
)
127 h1
%= CONNTRACK_LOCKS
;
128 h2
%= CONNTRACK_LOCKS
;
130 nf_conntrack_lock(&nf_conntrack_locks
[h1
]);
132 spin_lock_nested(&nf_conntrack_locks
[h2
],
133 SINGLE_DEPTH_NESTING
);
135 nf_conntrack_lock(&nf_conntrack_locks
[h2
]);
136 spin_lock_nested(&nf_conntrack_locks
[h1
],
137 SINGLE_DEPTH_NESTING
);
139 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
140 nf_conntrack_double_unlock(h1
, h2
);
146 static void nf_conntrack_all_lock(void)
150 spin_lock(&nf_conntrack_locks_all_lock
);
151 nf_conntrack_locks_all
= true;
154 * Order the above store of 'nf_conntrack_locks_all' against
155 * the spin_unlock_wait() loads below, such that if
156 * nf_conntrack_lock() observes 'nf_conntrack_locks_all'
157 * we must observe nf_conntrack_locks[] held:
159 smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
161 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++) {
162 spin_unlock_wait(&nf_conntrack_locks
[i
]);
166 static void nf_conntrack_all_unlock(void)
169 * All prior stores must be complete before we clear
170 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
171 * might observe the false value but not the entire
174 smp_store_release(&nf_conntrack_locks_all
, false);
175 spin_unlock(&nf_conntrack_locks_all_lock
);
178 unsigned int nf_conntrack_htable_size __read_mostly
;
179 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
181 unsigned int nf_conntrack_max __read_mostly
;
182 seqcount_t nf_conntrack_generation __read_mostly
;
184 DEFINE_PER_CPU(struct nf_conn
, nf_conntrack_untracked
);
185 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked
);
187 static unsigned int nf_conntrack_hash_rnd __read_mostly
;
189 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
,
190 const struct net
*net
)
195 get_random_once(&nf_conntrack_hash_rnd
, sizeof(nf_conntrack_hash_rnd
));
197 /* The direction must be ignored, so we hash everything up to the
198 * destination ports (which is a multiple of 4) and treat the last
199 * three bytes manually.
201 seed
= nf_conntrack_hash_rnd
^ net_hash_mix(net
);
202 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
203 return jhash2((u32
*)tuple
, n
, seed
^
204 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
205 tuple
->dst
.protonum
));
208 static u32
scale_hash(u32 hash
)
210 return reciprocal_scale(hash
, nf_conntrack_htable_size
);
213 static u32
__hash_conntrack(const struct net
*net
,
214 const struct nf_conntrack_tuple
*tuple
,
217 return reciprocal_scale(hash_conntrack_raw(tuple
, net
), size
);
220 static u32
hash_conntrack(const struct net
*net
,
221 const struct nf_conntrack_tuple
*tuple
)
223 return scale_hash(hash_conntrack_raw(tuple
, net
));
227 nf_ct_get_tuple(const struct sk_buff
*skb
,
229 unsigned int dataoff
,
233 struct nf_conntrack_tuple
*tuple
,
234 const struct nf_conntrack_l3proto
*l3proto
,
235 const struct nf_conntrack_l4proto
*l4proto
)
237 memset(tuple
, 0, sizeof(*tuple
));
239 tuple
->src
.l3num
= l3num
;
240 if (l3proto
->pkt_to_tuple(skb
, nhoff
, tuple
) == 0)
243 tuple
->dst
.protonum
= protonum
;
244 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
246 return l4proto
->pkt_to_tuple(skb
, dataoff
, net
, tuple
);
248 EXPORT_SYMBOL_GPL(nf_ct_get_tuple
);
250 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
252 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
254 struct nf_conntrack_l3proto
*l3proto
;
255 struct nf_conntrack_l4proto
*l4proto
;
256 unsigned int protoff
;
262 l3proto
= __nf_ct_l3proto_find(l3num
);
263 ret
= l3proto
->get_l4proto(skb
, nhoff
, &protoff
, &protonum
);
264 if (ret
!= NF_ACCEPT
) {
269 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
271 ret
= nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
,
277 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
280 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
281 const struct nf_conntrack_tuple
*orig
,
282 const struct nf_conntrack_l3proto
*l3proto
,
283 const struct nf_conntrack_l4proto
*l4proto
)
285 memset(inverse
, 0, sizeof(*inverse
));
287 inverse
->src
.l3num
= orig
->src
.l3num
;
288 if (l3proto
->invert_tuple(inverse
, orig
) == 0)
291 inverse
->dst
.dir
= !orig
->dst
.dir
;
293 inverse
->dst
.protonum
= orig
->dst
.protonum
;
294 return l4proto
->invert_tuple(inverse
, orig
);
296 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
299 clean_from_lists(struct nf_conn
*ct
)
301 pr_debug("clean_from_lists(%p)\n", ct
);
302 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
303 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
305 /* Destroy all pending expectations */
306 nf_ct_remove_expectations(ct
);
309 /* must be called with local_bh_disable */
310 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
312 struct ct_pcpu
*pcpu
;
314 /* add this conntrack to the (per cpu) dying list */
315 ct
->cpu
= smp_processor_id();
316 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
318 spin_lock(&pcpu
->lock
);
319 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
321 spin_unlock(&pcpu
->lock
);
324 /* must be called with local_bh_disable */
325 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
327 struct ct_pcpu
*pcpu
;
329 /* add this conntrack to the (per cpu) unconfirmed list */
330 ct
->cpu
= smp_processor_id();
331 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
333 spin_lock(&pcpu
->lock
);
334 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
336 spin_unlock(&pcpu
->lock
);
339 /* must be called with local_bh_disable */
340 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
342 struct ct_pcpu
*pcpu
;
344 /* We overload first tuple to link into unconfirmed or dying list.*/
345 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
347 spin_lock(&pcpu
->lock
);
348 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
349 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
350 spin_unlock(&pcpu
->lock
);
353 /* Released via destroy_conntrack() */
354 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
355 const struct nf_conntrack_zone
*zone
,
358 struct nf_conn
*tmpl
;
360 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
364 tmpl
->status
= IPS_TEMPLATE
;
365 write_pnet(&tmpl
->ct_net
, net
);
366 nf_ct_zone_add(tmpl
, zone
);
367 atomic_set(&tmpl
->ct_general
.use
, 0);
371 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
373 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
375 nf_ct_ext_destroy(tmpl
);
376 nf_ct_ext_free(tmpl
);
379 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
382 destroy_conntrack(struct nf_conntrack
*nfct
)
384 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
385 struct nf_conntrack_l4proto
*l4proto
;
387 pr_debug("destroy_conntrack(%p)\n", ct
);
388 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
390 if (unlikely(nf_ct_is_template(ct
))) {
395 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
396 if (l4proto
->destroy
)
397 l4proto
->destroy(ct
);
402 /* Expectations will have been removed in clean_from_lists,
403 * except TFTP can create an expectation on the first packet,
404 * before connection is in the list, so we need to clean here,
407 nf_ct_remove_expectations(ct
);
409 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
414 nf_ct_put(ct
->master
);
416 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
417 nf_conntrack_free(ct
);
420 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
422 struct net
*net
= nf_ct_net(ct
);
423 unsigned int hash
, reply_hash
;
424 unsigned int sequence
;
426 nf_ct_helper_destroy(ct
);
430 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
431 hash
= hash_conntrack(net
,
432 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
433 reply_hash
= hash_conntrack(net
,
434 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
435 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
437 clean_from_lists(ct
);
438 nf_conntrack_double_unlock(hash
, reply_hash
);
440 nf_ct_add_to_dying_list(ct
);
445 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
447 struct nf_conn_tstamp
*tstamp
;
449 if (test_and_set_bit(IPS_DYING_BIT
, &ct
->status
))
452 tstamp
= nf_conn_tstamp_find(ct
);
453 if (tstamp
&& tstamp
->stop
== 0)
454 tstamp
->stop
= ktime_get_real_ns();
456 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
457 portid
, report
) < 0) {
458 /* destroy event was not delivered. nf_ct_put will
459 * be done by event cache worker on redelivery.
461 nf_ct_delete_from_lists(ct
);
462 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
466 nf_conntrack_ecache_work(nf_ct_net(ct
));
467 nf_ct_delete_from_lists(ct
);
471 EXPORT_SYMBOL_GPL(nf_ct_delete
);
474 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
475 const struct nf_conntrack_tuple
*tuple
,
476 const struct nf_conntrack_zone
*zone
,
477 const struct net
*net
)
479 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
481 /* A conntrack can be recreated with the equal tuple,
482 * so we need to check that the conntrack is confirmed
484 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
485 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
486 nf_ct_is_confirmed(ct
) &&
487 net_eq(net
, nf_ct_net(ct
));
490 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
491 static void nf_ct_gc_expired(struct nf_conn
*ct
)
493 if (!atomic_inc_not_zero(&ct
->ct_general
.use
))
496 if (nf_ct_should_gc(ct
))
504 * - Caller must take a reference on returned object
505 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
507 static struct nf_conntrack_tuple_hash
*
508 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
509 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
511 struct nf_conntrack_tuple_hash
*h
;
512 struct hlist_nulls_head
*ct_hash
;
513 struct hlist_nulls_node
*n
;
514 unsigned int bucket
, hsize
;
517 nf_conntrack_get_ht(&ct_hash
, &hsize
);
518 bucket
= reciprocal_scale(hash
, hsize
);
520 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[bucket
], hnnode
) {
523 ct
= nf_ct_tuplehash_to_ctrack(h
);
524 if (nf_ct_is_expired(ct
)) {
525 nf_ct_gc_expired(ct
);
529 if (nf_ct_is_dying(ct
))
532 if (nf_ct_key_equal(h
, tuple
, zone
, net
))
536 * if the nulls value we got at the end of this lookup is
537 * not the expected one, we must restart lookup.
538 * We probably met an item that was moved to another chain.
540 if (get_nulls_value(n
) != bucket
) {
541 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
548 /* Find a connection corresponding to a tuple. */
549 static struct nf_conntrack_tuple_hash
*
550 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
551 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
553 struct nf_conntrack_tuple_hash
*h
;
558 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
560 ct
= nf_ct_tuplehash_to_ctrack(h
);
561 if (unlikely(nf_ct_is_dying(ct
) ||
562 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
565 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
, net
))) {
576 struct nf_conntrack_tuple_hash
*
577 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
578 const struct nf_conntrack_tuple
*tuple
)
580 return __nf_conntrack_find_get(net
, zone
, tuple
,
581 hash_conntrack_raw(tuple
, net
));
583 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
585 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
587 unsigned int reply_hash
)
589 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
590 &nf_conntrack_hash
[hash
]);
591 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
592 &nf_conntrack_hash
[reply_hash
]);
596 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
598 const struct nf_conntrack_zone
*zone
;
599 struct net
*net
= nf_ct_net(ct
);
600 unsigned int hash
, reply_hash
;
601 struct nf_conntrack_tuple_hash
*h
;
602 struct hlist_nulls_node
*n
;
603 unsigned int sequence
;
605 zone
= nf_ct_zone(ct
);
609 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
610 hash
= hash_conntrack(net
,
611 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
612 reply_hash
= hash_conntrack(net
,
613 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
614 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
616 /* See if there's one in the list already, including reverse */
617 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
618 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
622 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
623 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
628 /* The caller holds a reference to this object */
629 atomic_set(&ct
->ct_general
.use
, 2);
630 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
631 nf_conntrack_double_unlock(hash
, reply_hash
);
632 NF_CT_STAT_INC(net
, insert
);
637 nf_conntrack_double_unlock(hash
, reply_hash
);
638 NF_CT_STAT_INC(net
, insert_failed
);
642 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
644 static inline void nf_ct_acct_update(struct nf_conn
*ct
,
645 enum ip_conntrack_info ctinfo
,
648 struct nf_conn_acct
*acct
;
650 acct
= nf_conn_acct_find(ct
);
652 struct nf_conn_counter
*counter
= acct
->counter
;
654 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
655 atomic64_add(len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
659 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
660 const struct nf_conn
*loser_ct
)
662 struct nf_conn_acct
*acct
;
664 acct
= nf_conn_acct_find(loser_ct
);
666 struct nf_conn_counter
*counter
= acct
->counter
;
669 /* u32 should be fine since we must have seen one packet. */
670 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
671 nf_ct_acct_update(ct
, ctinfo
, bytes
);
675 /* Resolve race on insertion if this protocol allows this. */
676 static int nf_ct_resolve_clash(struct net
*net
, struct sk_buff
*skb
,
677 enum ip_conntrack_info ctinfo
,
678 struct nf_conntrack_tuple_hash
*h
)
680 /* This is the conntrack entry already in hashes that won race. */
681 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
682 struct nf_conntrack_l4proto
*l4proto
;
684 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
685 if (l4proto
->allow_clash
&&
687 !nf_ct_is_dying(ct
) &&
688 atomic_inc_not_zero(&ct
->ct_general
.use
)) {
689 enum ip_conntrack_info oldinfo
;
690 struct nf_conn
*loser_ct
= nf_ct_get(skb
, &oldinfo
);
692 nf_ct_acct_merge(ct
, ctinfo
, loser_ct
);
693 nf_conntrack_put(&loser_ct
->ct_general
);
694 /* Assign conntrack already in hashes to this skbuff. Don't
695 * modify skb->nfctinfo to ensure consistent stateful filtering.
697 skb
->nfct
= &ct
->ct_general
;
700 NF_CT_STAT_INC(net
, drop
);
704 /* Confirm a connection given skb; places it in hash table */
706 __nf_conntrack_confirm(struct sk_buff
*skb
)
708 const struct nf_conntrack_zone
*zone
;
709 unsigned int hash
, reply_hash
;
710 struct nf_conntrack_tuple_hash
*h
;
712 struct nf_conn_help
*help
;
713 struct nf_conn_tstamp
*tstamp
;
714 struct hlist_nulls_node
*n
;
715 enum ip_conntrack_info ctinfo
;
717 unsigned int sequence
;
720 ct
= nf_ct_get(skb
, &ctinfo
);
723 /* ipt_REJECT uses nf_conntrack_attach to attach related
724 ICMP/TCP RST packets in other direction. Actual packet
725 which created connection will be IP_CT_NEW or for an
726 expected connection, IP_CT_RELATED. */
727 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
730 zone
= nf_ct_zone(ct
);
734 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
735 /* reuse the hash saved before */
736 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
737 hash
= scale_hash(hash
);
738 reply_hash
= hash_conntrack(net
,
739 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
741 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
743 /* We're not in hash table, and we refuse to set up related
744 * connections for unconfirmed conns. But packet copies and
745 * REJECT will give spurious warnings here.
747 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
749 /* No external references means no one else could have
752 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
753 pr_debug("Confirming conntrack %p\n", ct
);
754 /* We have to check the DYING flag after unlink to prevent
755 * a race against nf_ct_get_next_corpse() possibly called from
756 * user context, else we insert an already 'dead' hash, blocking
757 * further use of that particular connection -JM.
759 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
761 if (unlikely(nf_ct_is_dying(ct
))) {
762 nf_ct_add_to_dying_list(ct
);
766 /* See if there's one in the list already, including reverse:
767 NAT could have grabbed it without realizing, since we're
768 not in the hash. If there is, we lost race. */
769 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
770 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
774 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
775 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
779 /* Timer relative to confirmation time, not original
780 setting time, otherwise we'd get timer wrap in
781 weird delay cases. */
782 ct
->timeout
+= nfct_time_stamp
;
783 atomic_inc(&ct
->ct_general
.use
);
784 ct
->status
|= IPS_CONFIRMED
;
786 /* set conntrack timestamp, if enabled. */
787 tstamp
= nf_conn_tstamp_find(ct
);
789 if (skb
->tstamp
== 0)
790 __net_timestamp(skb
);
792 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
794 /* Since the lookup is lockless, hash insertion must be done after
795 * starting the timer and setting the CONFIRMED bit. The RCU barriers
796 * guarantee that no other CPU can find the conntrack before the above
797 * stores are visible.
799 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
800 nf_conntrack_double_unlock(hash
, reply_hash
);
803 help
= nfct_help(ct
);
804 if (help
&& help
->helper
)
805 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
807 nf_conntrack_event_cache(master_ct(ct
) ?
808 IPCT_RELATED
: IPCT_NEW
, ct
);
812 nf_ct_add_to_dying_list(ct
);
813 ret
= nf_ct_resolve_clash(net
, skb
, ctinfo
, h
);
815 nf_conntrack_double_unlock(hash
, reply_hash
);
816 NF_CT_STAT_INC(net
, insert_failed
);
820 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
822 /* Returns true if a connection correspondings to the tuple (required
825 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
826 const struct nf_conn
*ignored_conntrack
)
828 struct net
*net
= nf_ct_net(ignored_conntrack
);
829 const struct nf_conntrack_zone
*zone
;
830 struct nf_conntrack_tuple_hash
*h
;
831 struct hlist_nulls_head
*ct_hash
;
832 unsigned int hash
, hsize
;
833 struct hlist_nulls_node
*n
;
836 zone
= nf_ct_zone(ignored_conntrack
);
840 nf_conntrack_get_ht(&ct_hash
, &hsize
);
841 hash
= __hash_conntrack(net
, tuple
, hsize
);
843 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
844 ct
= nf_ct_tuplehash_to_ctrack(h
);
846 if (ct
== ignored_conntrack
)
849 if (nf_ct_is_expired(ct
)) {
850 nf_ct_gc_expired(ct
);
854 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
855 NF_CT_STAT_INC_ATOMIC(net
, found
);
861 if (get_nulls_value(n
) != hash
) {
862 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
870 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
872 #define NF_CT_EVICTION_RANGE 8
874 /* There's a small race here where we may free a just-assured
875 connection. Too bad: we're in trouble anyway. */
876 static unsigned int early_drop_list(struct net
*net
,
877 struct hlist_nulls_head
*head
)
879 struct nf_conntrack_tuple_hash
*h
;
880 struct hlist_nulls_node
*n
;
881 unsigned int drops
= 0;
884 hlist_nulls_for_each_entry_rcu(h
, n
, head
, hnnode
) {
885 tmp
= nf_ct_tuplehash_to_ctrack(h
);
887 if (nf_ct_is_expired(tmp
)) {
888 nf_ct_gc_expired(tmp
);
892 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
893 !net_eq(nf_ct_net(tmp
), net
) ||
897 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
900 /* kill only if still in same netns -- might have moved due to
901 * SLAB_DESTROY_BY_RCU rules.
903 * We steal the timer reference. If that fails timer has
904 * already fired or someone else deleted it. Just drop ref
905 * and move to next entry.
907 if (net_eq(nf_ct_net(tmp
), net
) &&
908 nf_ct_is_confirmed(tmp
) &&
909 nf_ct_delete(tmp
, 0, 0))
918 static noinline
int early_drop(struct net
*net
, unsigned int _hash
)
922 for (i
= 0; i
< NF_CT_EVICTION_RANGE
; i
++) {
923 struct hlist_nulls_head
*ct_hash
;
924 unsigned int hash
, hsize
, drops
;
927 nf_conntrack_get_ht(&ct_hash
, &hsize
);
928 hash
= reciprocal_scale(_hash
++, hsize
);
930 drops
= early_drop_list(net
, &ct_hash
[hash
]);
934 NF_CT_STAT_ADD_ATOMIC(net
, early_drop
, drops
);
942 static void gc_worker(struct work_struct
*work
)
944 unsigned int i
, goal
, buckets
= 0, expired_count
= 0;
945 struct conntrack_gc_work
*gc_work
;
946 unsigned int ratio
, scanned
= 0;
947 unsigned long next_run
;
949 gc_work
= container_of(work
, struct conntrack_gc_work
, dwork
.work
);
951 goal
= nf_conntrack_htable_size
/ GC_MAX_BUCKETS_DIV
;
952 i
= gc_work
->last_bucket
;
955 struct nf_conntrack_tuple_hash
*h
;
956 struct hlist_nulls_head
*ct_hash
;
957 struct hlist_nulls_node
*n
;
964 nf_conntrack_get_ht(&ct_hash
, &hashsz
);
968 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[i
], hnnode
) {
969 tmp
= nf_ct_tuplehash_to_ctrack(h
);
972 if (nf_ct_is_expired(tmp
)) {
973 nf_ct_gc_expired(tmp
);
979 /* could check get_nulls_value() here and restart if ct
980 * was moved to another chain. But given gc is best-effort
981 * we will just continue with next hash slot.
984 cond_resched_rcu_qs();
985 } while (++buckets
< goal
&&
986 expired_count
< GC_MAX_EVICTS
);
988 if (gc_work
->exiting
)
992 * Eviction will normally happen from the packet path, and not
993 * from this gc worker.
995 * This worker is only here to reap expired entries when system went
996 * idle after a busy period.
998 * The heuristics below are supposed to balance conflicting goals:
1000 * 1. Minimize time until we notice a stale entry
1001 * 2. Maximize scan intervals to not waste cycles
1003 * Normally, expired_count will be 0, this increases the next_run time
1004 * to priorize 2) above.
1006 * As soon as a timed-out entry is found, move towards 1) and increase
1007 * the scan frequency.
1008 * In case we have lots of evictions next scan is done immediately.
1010 ratio
= scanned
? expired_count
* 100 / scanned
: 0;
1011 if (ratio
>= 90 || expired_count
== GC_MAX_EVICTS
) {
1012 gc_work
->next_gc_run
= 0;
1014 } else if (expired_count
) {
1015 gc_work
->next_gc_run
/= 2U;
1016 next_run
= msecs_to_jiffies(1);
1018 if (gc_work
->next_gc_run
< GC_INTERVAL_MAX
)
1019 gc_work
->next_gc_run
+= msecs_to_jiffies(1);
1021 next_run
= gc_work
->next_gc_run
;
1024 gc_work
->last_bucket
= i
;
1025 queue_delayed_work(system_long_wq
, &gc_work
->dwork
, next_run
);
1028 static void conntrack_gc_work_init(struct conntrack_gc_work
*gc_work
)
1030 INIT_DELAYED_WORK(&gc_work
->dwork
, gc_worker
);
1031 gc_work
->next_gc_run
= GC_INTERVAL_MAX
;
1032 gc_work
->exiting
= false;
1035 static struct nf_conn
*
1036 __nf_conntrack_alloc(struct net
*net
,
1037 const struct nf_conntrack_zone
*zone
,
1038 const struct nf_conntrack_tuple
*orig
,
1039 const struct nf_conntrack_tuple
*repl
,
1040 gfp_t gfp
, u32 hash
)
1044 /* We don't want any race condition at early drop stage */
1045 atomic_inc(&net
->ct
.count
);
1047 if (nf_conntrack_max
&&
1048 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
1049 if (!early_drop(net
, hash
)) {
1050 atomic_dec(&net
->ct
.count
);
1051 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1052 return ERR_PTR(-ENOMEM
);
1057 * Do not use kmem_cache_zalloc(), as this cache uses
1058 * SLAB_DESTROY_BY_RCU.
1060 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
1064 spin_lock_init(&ct
->lock
);
1065 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
1066 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
1067 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
1068 /* save hash for reusing when confirming */
1069 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
1071 write_pnet(&ct
->ct_net
, net
);
1072 memset(&ct
->__nfct_init_offset
[0], 0,
1073 offsetof(struct nf_conn
, proto
) -
1074 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
1076 nf_ct_zone_add(ct
, zone
);
1078 /* Because we use RCU lookups, we set ct_general.use to zero before
1079 * this is inserted in any list.
1081 atomic_set(&ct
->ct_general
.use
, 0);
1084 atomic_dec(&net
->ct
.count
);
1085 return ERR_PTR(-ENOMEM
);
1088 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
1089 const struct nf_conntrack_zone
*zone
,
1090 const struct nf_conntrack_tuple
*orig
,
1091 const struct nf_conntrack_tuple
*repl
,
1094 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
1096 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
1098 void nf_conntrack_free(struct nf_conn
*ct
)
1100 struct net
*net
= nf_ct_net(ct
);
1102 /* A freed object has refcnt == 0, that's
1103 * the golden rule for SLAB_DESTROY_BY_RCU
1105 NF_CT_ASSERT(atomic_read(&ct
->ct_general
.use
) == 0);
1107 nf_ct_ext_destroy(ct
);
1109 kmem_cache_free(nf_conntrack_cachep
, ct
);
1110 smp_mb__before_atomic();
1111 atomic_dec(&net
->ct
.count
);
1113 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
1116 /* Allocate a new conntrack: we return -ENOMEM if classification
1117 failed due to stress. Otherwise it really is unclassifiable. */
1118 static struct nf_conntrack_tuple_hash
*
1119 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
1120 const struct nf_conntrack_tuple
*tuple
,
1121 struct nf_conntrack_l3proto
*l3proto
,
1122 struct nf_conntrack_l4proto
*l4proto
,
1123 struct sk_buff
*skb
,
1124 unsigned int dataoff
, u32 hash
)
1127 struct nf_conn_help
*help
;
1128 struct nf_conntrack_tuple repl_tuple
;
1129 struct nf_conntrack_ecache
*ecache
;
1130 struct nf_conntrack_expect
*exp
= NULL
;
1131 const struct nf_conntrack_zone
*zone
;
1132 struct nf_conn_timeout
*timeout_ext
;
1133 struct nf_conntrack_zone tmp
;
1134 unsigned int *timeouts
;
1136 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
1137 pr_debug("Can't invert tuple.\n");
1141 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1142 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
1145 return (struct nf_conntrack_tuple_hash
*)ct
;
1147 if (!nf_ct_add_synproxy(ct
, tmpl
)) {
1148 nf_conntrack_free(ct
);
1149 return ERR_PTR(-ENOMEM
);
1152 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1154 timeouts
= nf_ct_timeout_data(timeout_ext
);
1155 if (unlikely(!timeouts
))
1156 timeouts
= l4proto
->get_timeouts(net
);
1158 timeouts
= l4proto
->get_timeouts(net
);
1161 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
1162 nf_conntrack_free(ct
);
1163 pr_debug("can't track with proto module\n");
1168 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1171 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1172 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1173 nf_ct_labels_ext_add(ct
);
1175 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1176 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1177 ecache
? ecache
->expmask
: 0,
1181 if (net
->ct
.expect_count
) {
1182 spin_lock(&nf_conntrack_expect_lock
);
1183 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1185 pr_debug("expectation arrives ct=%p exp=%p\n",
1187 /* Welcome, Mr. Bond. We've been expecting you... */
1188 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1189 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1190 ct
->master
= exp
->master
;
1192 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
1195 rcu_assign_pointer(help
->helper
, exp
->helper
);
1198 #ifdef CONFIG_NF_CONNTRACK_MARK
1199 ct
->mark
= exp
->master
->mark
;
1201 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1202 ct
->secmark
= exp
->master
->secmark
;
1204 NF_CT_STAT_INC(net
, expect_new
);
1206 spin_unlock(&nf_conntrack_expect_lock
);
1209 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1211 /* Now it is inserted into the unconfirmed list, bump refcount */
1212 nf_conntrack_get(&ct
->ct_general
);
1213 nf_ct_add_to_unconfirmed_list(ct
);
1219 exp
->expectfn(ct
, exp
);
1220 nf_ct_expect_put(exp
);
1223 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1226 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1227 static inline struct nf_conn
*
1228 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1229 struct sk_buff
*skb
,
1230 unsigned int dataoff
,
1233 struct nf_conntrack_l3proto
*l3proto
,
1234 struct nf_conntrack_l4proto
*l4proto
,
1236 enum ip_conntrack_info
*ctinfo
)
1238 const struct nf_conntrack_zone
*zone
;
1239 struct nf_conntrack_tuple tuple
;
1240 struct nf_conntrack_tuple_hash
*h
;
1241 struct nf_conntrack_zone tmp
;
1245 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1246 dataoff
, l3num
, protonum
, net
, &tuple
, l3proto
,
1248 pr_debug("Can't get tuple\n");
1252 /* look for tuple match */
1253 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1254 hash
= hash_conntrack_raw(&tuple
, net
);
1255 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1257 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1258 skb
, dataoff
, hash
);
1264 ct
= nf_ct_tuplehash_to_ctrack(h
);
1266 /* It exists; we have (non-exclusive) reference. */
1267 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1268 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1269 /* Please set reply bit if this packet OK */
1272 /* Once we've had two way comms, always ESTABLISHED. */
1273 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1274 pr_debug("normal packet for %p\n", ct
);
1275 *ctinfo
= IP_CT_ESTABLISHED
;
1276 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1277 pr_debug("related packet for %p\n", ct
);
1278 *ctinfo
= IP_CT_RELATED
;
1280 pr_debug("new packet for %p\n", ct
);
1281 *ctinfo
= IP_CT_NEW
;
1285 skb
->nfct
= &ct
->ct_general
;
1286 skb
->nfctinfo
= *ctinfo
;
1291 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1292 struct sk_buff
*skb
)
1294 struct nf_conn
*ct
, *tmpl
;
1295 enum ip_conntrack_info ctinfo
;
1296 struct nf_conntrack_l3proto
*l3proto
;
1297 struct nf_conntrack_l4proto
*l4proto
;
1298 unsigned int *timeouts
;
1299 unsigned int dataoff
;
1304 tmpl
= nf_ct_get(skb
, &ctinfo
);
1306 /* Previously seen (loopback or untracked)? Ignore. */
1307 if (!nf_ct_is_template(tmpl
)) {
1308 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1314 /* rcu_read_lock()ed by nf_hook_thresh */
1315 l3proto
= __nf_ct_l3proto_find(pf
);
1316 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1317 &dataoff
, &protonum
);
1319 pr_debug("not prepared to track yet or error occurred\n");
1320 NF_CT_STAT_INC_ATOMIC(net
, error
);
1321 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1326 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1328 /* It may be an special packet, error, unclean...
1329 * inverse of the return code tells to the netfilter
1330 * core what to do with the packet. */
1331 if (l4proto
->error
!= NULL
) {
1332 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, pf
, hooknum
);
1334 NF_CT_STAT_INC_ATOMIC(net
, error
);
1335 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1339 /* ICMP[v6] protocol trackers may assign one conntrack. */
1344 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1345 l3proto
, l4proto
, &set_reply
, &ctinfo
);
1347 /* Not valid part of a connection */
1348 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1354 /* Too stressed to deal. */
1355 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1360 NF_CT_ASSERT(skb
->nfct
);
1362 /* Decide what timeout policy we want to apply to this flow. */
1363 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1365 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1367 /* Invalid: inverse of the return code tells
1368 * the netfilter core what to do */
1369 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1370 nf_conntrack_put(&ct
->ct_general
);
1372 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1373 if (ret
== -NF_DROP
)
1374 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1375 /* Special case: TCP tracker reports an attempt to reopen a
1376 * closed/aborted connection. We have to go back and create a
1379 if (ret
== -NF_REPEAT
)
1385 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1386 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1393 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1395 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1396 const struct nf_conntrack_tuple
*orig
)
1401 ret
= nf_ct_invert_tuple(inverse
, orig
,
1402 __nf_ct_l3proto_find(orig
->src
.l3num
),
1403 __nf_ct_l4proto_find(orig
->src
.l3num
,
1404 orig
->dst
.protonum
));
1408 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1410 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1411 implicitly racy: see __nf_conntrack_confirm */
1412 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1413 const struct nf_conntrack_tuple
*newreply
)
1415 struct nf_conn_help
*help
= nfct_help(ct
);
1417 /* Should be unconfirmed, so not in hash table yet */
1418 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1420 pr_debug("Altering reply tuple of %p to ", ct
);
1421 nf_ct_dump_tuple(newreply
);
1423 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1424 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1428 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1431 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1433 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1434 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1435 enum ip_conntrack_info ctinfo
,
1436 const struct sk_buff
*skb
,
1437 unsigned long extra_jiffies
,
1442 /* Only update if this is not a fixed timeout */
1443 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1446 /* If not in hash table, timer will not be active yet */
1447 if (nf_ct_is_confirmed(ct
))
1448 extra_jiffies
+= nfct_time_stamp
;
1450 ct
->timeout
= extra_jiffies
;
1453 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1455 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1457 bool nf_ct_kill_acct(struct nf_conn
*ct
,
1458 enum ip_conntrack_info ctinfo
,
1459 const struct sk_buff
*skb
)
1461 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1463 return nf_ct_delete(ct
, 0, 0);
1465 EXPORT_SYMBOL_GPL(nf_ct_kill_acct
);
1467 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1469 #include <linux/netfilter/nfnetlink.h>
1470 #include <linux/netfilter/nfnetlink_conntrack.h>
1471 #include <linux/mutex.h>
1473 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1474 * in ip_conntrack_core, since we don't want the protocols to autoload
1475 * or depend on ctnetlink */
1476 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1477 const struct nf_conntrack_tuple
*tuple
)
1479 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1480 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1481 goto nla_put_failure
;
1487 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1489 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1490 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1491 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1493 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1495 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1496 struct nf_conntrack_tuple
*t
)
1498 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1501 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1502 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1506 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1508 int nf_ct_port_nlattr_tuple_size(void)
1510 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1512 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1515 /* Used by ipt_REJECT and ip6t_REJECT. */
1516 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1519 enum ip_conntrack_info ctinfo
;
1521 /* This ICMP is in reverse direction to the packet which caused it */
1522 ct
= nf_ct_get(skb
, &ctinfo
);
1523 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1524 ctinfo
= IP_CT_RELATED_REPLY
;
1526 ctinfo
= IP_CT_RELATED
;
1528 /* Attach to new skbuff, and increment count */
1529 nskb
->nfct
= &ct
->ct_general
;
1530 nskb
->nfctinfo
= ctinfo
;
1531 nf_conntrack_get(nskb
->nfct
);
1534 /* Bring out ya dead! */
1535 static struct nf_conn
*
1536 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1537 void *data
, unsigned int *bucket
)
1539 struct nf_conntrack_tuple_hash
*h
;
1541 struct hlist_nulls_node
*n
;
1545 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1546 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1548 nf_conntrack_lock(lockp
);
1549 if (*bucket
< nf_conntrack_htable_size
) {
1550 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
1551 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1553 ct
= nf_ct_tuplehash_to_ctrack(h
);
1554 if (net_eq(nf_ct_net(ct
), net
) &&
1564 for_each_possible_cpu(cpu
) {
1565 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1567 spin_lock_bh(&pcpu
->lock
);
1568 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1569 ct
= nf_ct_tuplehash_to_ctrack(h
);
1571 set_bit(IPS_DYING_BIT
, &ct
->status
);
1573 spin_unlock_bh(&pcpu
->lock
);
1578 atomic_inc(&ct
->ct_general
.use
);
1584 void nf_ct_iterate_cleanup(struct net
*net
,
1585 int (*iter
)(struct nf_conn
*i
, void *data
),
1586 void *data
, u32 portid
, int report
)
1589 unsigned int bucket
= 0;
1593 if (atomic_read(&net
->ct
.count
) == 0)
1596 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1597 /* Time to push up daises... */
1599 nf_ct_delete(ct
, portid
, report
);
1604 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1606 static int kill_all(struct nf_conn
*i
, void *data
)
1611 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1613 if (is_vmalloc_addr(hash
))
1616 free_pages((unsigned long)hash
,
1617 get_order(sizeof(struct hlist_head
) * size
));
1619 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1621 static int untrack_refs(void)
1625 for_each_possible_cpu(cpu
) {
1626 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1628 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1633 void nf_conntrack_cleanup_start(void)
1635 conntrack_gc_work
.exiting
= true;
1636 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1639 void nf_conntrack_cleanup_end(void)
1641 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1642 while (untrack_refs() > 0)
1645 cancel_delayed_work_sync(&conntrack_gc_work
.dwork
);
1646 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1648 nf_conntrack_proto_fini();
1649 nf_conntrack_seqadj_fini();
1650 nf_conntrack_labels_fini();
1651 nf_conntrack_helper_fini();
1652 nf_conntrack_timeout_fini();
1653 nf_conntrack_ecache_fini();
1654 nf_conntrack_tstamp_fini();
1655 nf_conntrack_acct_fini();
1656 nf_conntrack_expect_fini();
1658 kmem_cache_destroy(nf_conntrack_cachep
);
1662 * Mishearing the voices in his head, our hero wonders how he's
1663 * supposed to kill the mall.
1665 void nf_conntrack_cleanup_net(struct net
*net
)
1669 list_add(&net
->exit_list
, &single
);
1670 nf_conntrack_cleanup_net_list(&single
);
1673 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1679 * This makes sure all current packets have passed through
1680 * netfilter framework. Roll on, two-stage module
1686 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1687 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1688 if (atomic_read(&net
->ct
.count
) != 0)
1693 goto i_see_dead_people
;
1696 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1697 nf_conntrack_proto_pernet_fini(net
);
1698 nf_conntrack_helper_pernet_fini(net
);
1699 nf_conntrack_ecache_pernet_fini(net
);
1700 nf_conntrack_tstamp_pernet_fini(net
);
1701 nf_conntrack_acct_pernet_fini(net
);
1702 nf_conntrack_expect_pernet_fini(net
);
1703 free_percpu(net
->ct
.stat
);
1704 free_percpu(net
->ct
.pcpu_lists
);
1708 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1710 struct hlist_nulls_head
*hash
;
1711 unsigned int nr_slots
, i
;
1714 if (*sizep
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
1717 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1718 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1720 if (nr_slots
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
1723 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1724 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1730 for (i
= 0; i
< nr_slots
; i
++)
1731 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1735 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1737 int nf_conntrack_hash_resize(unsigned int hashsize
)
1740 unsigned int old_size
;
1741 struct hlist_nulls_head
*hash
, *old_hash
;
1742 struct nf_conntrack_tuple_hash
*h
;
1748 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1752 old_size
= nf_conntrack_htable_size
;
1753 if (old_size
== hashsize
) {
1754 nf_ct_free_hashtable(hash
, hashsize
);
1759 nf_conntrack_all_lock();
1760 write_seqcount_begin(&nf_conntrack_generation
);
1762 /* Lookups in the old hash might happen in parallel, which means we
1763 * might get false negatives during connection lookup. New connections
1764 * created because of a false negative won't make it into the hash
1765 * though since that required taking the locks.
1768 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1769 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
1770 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
1771 struct nf_conntrack_tuple_hash
, hnnode
);
1772 ct
= nf_ct_tuplehash_to_ctrack(h
);
1773 hlist_nulls_del_rcu(&h
->hnnode
);
1774 bucket
= __hash_conntrack(nf_ct_net(ct
),
1775 &h
->tuple
, hashsize
);
1776 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1779 old_size
= nf_conntrack_htable_size
;
1780 old_hash
= nf_conntrack_hash
;
1782 nf_conntrack_hash
= hash
;
1783 nf_conntrack_htable_size
= hashsize
;
1785 write_seqcount_end(&nf_conntrack_generation
);
1786 nf_conntrack_all_unlock();
1790 nf_ct_free_hashtable(old_hash
, old_size
);
1794 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1796 unsigned int hashsize
;
1799 if (current
->nsproxy
->net_ns
!= &init_net
)
1802 /* On boot, we can set this without any fancy locking. */
1803 if (!nf_conntrack_htable_size
)
1804 return param_set_uint(val
, kp
);
1806 rc
= kstrtouint(val
, 0, &hashsize
);
1810 return nf_conntrack_hash_resize(hashsize
);
1812 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1814 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1815 &nf_conntrack_htable_size
, 0600);
1817 void nf_ct_untracked_status_or(unsigned long bits
)
1821 for_each_possible_cpu(cpu
)
1822 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1824 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1826 int nf_conntrack_init_start(void)
1832 seqcount_init(&nf_conntrack_generation
);
1834 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
1835 spin_lock_init(&nf_conntrack_locks
[i
]);
1837 if (!nf_conntrack_htable_size
) {
1838 /* Idea from tcp.c: use 1/16384 of memory.
1839 * On i386: 32MB machine has 512 buckets.
1840 * >= 1GB machines have 16384 buckets.
1841 * >= 4GB machines have 65536 buckets.
1843 nf_conntrack_htable_size
1844 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1845 / sizeof(struct hlist_head
));
1846 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
1847 nf_conntrack_htable_size
= 65536;
1848 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1849 nf_conntrack_htable_size
= 16384;
1850 if (nf_conntrack_htable_size
< 32)
1851 nf_conntrack_htable_size
= 32;
1853 /* Use a max. factor of four by default to get the same max as
1854 * with the old struct list_heads. When a table size is given
1855 * we use the old value of 8 to avoid reducing the max.
1860 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
1861 if (!nf_conntrack_hash
)
1864 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1866 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1867 sizeof(struct nf_conn
), 0,
1868 SLAB_DESTROY_BY_RCU
| SLAB_HWCACHE_ALIGN
, NULL
);
1869 if (!nf_conntrack_cachep
)
1872 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1873 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1876 ret
= nf_conntrack_expect_init();
1880 ret
= nf_conntrack_acct_init();
1884 ret
= nf_conntrack_tstamp_init();
1888 ret
= nf_conntrack_ecache_init();
1892 ret
= nf_conntrack_timeout_init();
1896 ret
= nf_conntrack_helper_init();
1900 ret
= nf_conntrack_labels_init();
1904 ret
= nf_conntrack_seqadj_init();
1908 ret
= nf_conntrack_proto_init();
1912 /* Set up fake conntrack: to never be deleted, not in any hashes */
1913 for_each_possible_cpu(cpu
) {
1914 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1915 write_pnet(&ct
->ct_net
, &init_net
);
1916 atomic_set(&ct
->ct_general
.use
, 1);
1918 /* - and look it like as a confirmed connection */
1919 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1921 conntrack_gc_work_init(&conntrack_gc_work
);
1922 queue_delayed_work(system_long_wq
, &conntrack_gc_work
.dwork
, GC_INTERVAL_MAX
);
1927 nf_conntrack_seqadj_fini();
1929 nf_conntrack_labels_fini();
1931 nf_conntrack_helper_fini();
1933 nf_conntrack_timeout_fini();
1935 nf_conntrack_ecache_fini();
1937 nf_conntrack_tstamp_fini();
1939 nf_conntrack_acct_fini();
1941 nf_conntrack_expect_fini();
1943 kmem_cache_destroy(nf_conntrack_cachep
);
1945 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1949 void nf_conntrack_init_end(void)
1951 /* For use by REJECT target */
1952 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1953 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1957 * We need to use special "null" values, not used in hash table
1959 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1960 #define DYING_NULLS_VAL ((1<<30)+1)
1961 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1963 int nf_conntrack_init_net(struct net
*net
)
1968 atomic_set(&net
->ct
.count
, 0);
1970 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
1971 if (!net
->ct
.pcpu_lists
)
1974 for_each_possible_cpu(cpu
) {
1975 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1977 spin_lock_init(&pcpu
->lock
);
1978 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1979 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
1982 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1984 goto err_pcpu_lists
;
1986 ret
= nf_conntrack_expect_pernet_init(net
);
1989 ret
= nf_conntrack_acct_pernet_init(net
);
1992 ret
= nf_conntrack_tstamp_pernet_init(net
);
1995 ret
= nf_conntrack_ecache_pernet_init(net
);
1998 ret
= nf_conntrack_helper_pernet_init(net
);
2001 ret
= nf_conntrack_proto_pernet_init(net
);
2007 nf_conntrack_helper_pernet_fini(net
);
2009 nf_conntrack_ecache_pernet_fini(net
);
2011 nf_conntrack_tstamp_pernet_fini(net
);
2013 nf_conntrack_acct_pernet_fini(net
);
2015 nf_conntrack_expect_pernet_fini(net
);
2017 free_percpu(net
->ct
.stat
);
2019 free_percpu(net
->ct
.pcpu_lists
);