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
);
331 if (nf_ct_zone_add(tmpl
, flags
, zone
) < 0)
334 atomic_set(&tmpl
->ct_general
.use
, 0);
341 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
343 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
345 nf_ct_ext_destroy(tmpl
);
346 nf_ct_ext_free(tmpl
);
349 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
352 destroy_conntrack(struct nf_conntrack
*nfct
)
354 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
355 struct net
*net
= nf_ct_net(ct
);
356 struct nf_conntrack_l4proto
*l4proto
;
358 pr_debug("destroy_conntrack(%p)\n", ct
);
359 NF_CT_ASSERT(atomic_read(&nfct
->use
) == 0);
360 NF_CT_ASSERT(!timer_pending(&ct
->timeout
));
362 if (unlikely(nf_ct_is_template(ct
))) {
367 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
368 if (l4proto
->destroy
)
369 l4proto
->destroy(ct
);
374 /* Expectations will have been removed in clean_from_lists,
375 * except TFTP can create an expectation on the first packet,
376 * before connection is in the list, so we need to clean here,
379 nf_ct_remove_expectations(ct
);
381 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
383 NF_CT_STAT_INC(net
, delete);
387 nf_ct_put(ct
->master
);
389 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
390 nf_conntrack_free(ct
);
393 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
395 struct net
*net
= nf_ct_net(ct
);
396 unsigned int hash
, reply_hash
;
397 unsigned int sequence
;
399 nf_ct_helper_destroy(ct
);
403 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
404 hash
= hash_conntrack(net
,
405 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
406 reply_hash
= hash_conntrack(net
,
407 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
408 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
410 clean_from_lists(ct
);
411 nf_conntrack_double_unlock(hash
, reply_hash
);
413 nf_ct_add_to_dying_list(ct
);
415 NF_CT_STAT_INC(net
, delete_list
);
419 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
421 struct nf_conn_tstamp
*tstamp
;
423 tstamp
= nf_conn_tstamp_find(ct
);
424 if (tstamp
&& tstamp
->stop
== 0)
425 tstamp
->stop
= ktime_get_real_ns();
427 if (nf_ct_is_dying(ct
))
430 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
431 portid
, report
) < 0) {
432 /* destroy event was not delivered */
433 nf_ct_delete_from_lists(ct
);
434 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
438 nf_conntrack_ecache_work(nf_ct_net(ct
));
439 set_bit(IPS_DYING_BIT
, &ct
->status
);
441 nf_ct_delete_from_lists(ct
);
445 EXPORT_SYMBOL_GPL(nf_ct_delete
);
447 static void death_by_timeout(unsigned long ul_conntrack
)
449 nf_ct_delete((struct nf_conn
*)ul_conntrack
, 0, 0);
453 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
454 const struct nf_conntrack_tuple
*tuple
,
455 const struct nf_conntrack_zone
*zone
,
456 const struct net
*net
)
458 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
460 /* A conntrack can be recreated with the equal tuple,
461 * so we need to check that the conntrack is confirmed
463 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
464 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
465 nf_ct_is_confirmed(ct
) &&
466 net_eq(net
, nf_ct_net(ct
));
471 * - Caller must take a reference on returned object
472 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
474 static struct nf_conntrack_tuple_hash
*
475 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
476 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
478 struct nf_conntrack_tuple_hash
*h
;
479 struct hlist_nulls_head
*ct_hash
;
480 struct hlist_nulls_node
*n
;
481 unsigned int bucket
, sequence
;
485 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
486 bucket
= scale_hash(hash
);
487 ct_hash
= nf_conntrack_hash
;
488 } while (read_seqcount_retry(&nf_conntrack_generation
, sequence
));
490 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[bucket
], hnnode
) {
491 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
492 NF_CT_STAT_INC_ATOMIC(net
, found
);
495 NF_CT_STAT_INC_ATOMIC(net
, searched
);
498 * if the nulls value we got at the end of this lookup is
499 * not the expected one, we must restart lookup.
500 * We probably met an item that was moved to another chain.
502 if (get_nulls_value(n
) != bucket
) {
503 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
510 /* Find a connection corresponding to a tuple. */
511 static struct nf_conntrack_tuple_hash
*
512 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
513 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
515 struct nf_conntrack_tuple_hash
*h
;
520 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
522 ct
= nf_ct_tuplehash_to_ctrack(h
);
523 if (unlikely(nf_ct_is_dying(ct
) ||
524 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
527 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
, net
))) {
538 struct nf_conntrack_tuple_hash
*
539 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
540 const struct nf_conntrack_tuple
*tuple
)
542 return __nf_conntrack_find_get(net
, zone
, tuple
,
543 hash_conntrack_raw(tuple
, net
));
545 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
547 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
549 unsigned int reply_hash
)
551 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
552 &nf_conntrack_hash
[hash
]);
553 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
554 &nf_conntrack_hash
[reply_hash
]);
558 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
560 const struct nf_conntrack_zone
*zone
;
561 struct net
*net
= nf_ct_net(ct
);
562 unsigned int hash
, reply_hash
;
563 struct nf_conntrack_tuple_hash
*h
;
564 struct hlist_nulls_node
*n
;
565 unsigned int sequence
;
567 zone
= nf_ct_zone(ct
);
571 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
572 hash
= hash_conntrack(net
,
573 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
574 reply_hash
= hash_conntrack(net
,
575 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
576 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
578 /* See if there's one in the list already, including reverse */
579 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
580 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
584 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
585 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
589 add_timer(&ct
->timeout
);
591 /* The caller holds a reference to this object */
592 atomic_set(&ct
->ct_general
.use
, 2);
593 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
594 nf_conntrack_double_unlock(hash
, reply_hash
);
595 NF_CT_STAT_INC(net
, insert
);
600 nf_conntrack_double_unlock(hash
, reply_hash
);
601 NF_CT_STAT_INC(net
, insert_failed
);
605 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
607 static inline void nf_ct_acct_update(struct nf_conn
*ct
,
608 enum ip_conntrack_info ctinfo
,
611 struct nf_conn_acct
*acct
;
613 acct
= nf_conn_acct_find(ct
);
615 struct nf_conn_counter
*counter
= acct
->counter
;
617 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
618 atomic64_add(len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
622 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
623 const struct nf_conn
*loser_ct
)
625 struct nf_conn_acct
*acct
;
627 acct
= nf_conn_acct_find(loser_ct
);
629 struct nf_conn_counter
*counter
= acct
->counter
;
632 /* u32 should be fine since we must have seen one packet. */
633 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
634 nf_ct_acct_update(ct
, ctinfo
, bytes
);
638 /* Resolve race on insertion if this protocol allows this. */
639 static int nf_ct_resolve_clash(struct net
*net
, struct sk_buff
*skb
,
640 enum ip_conntrack_info ctinfo
,
641 struct nf_conntrack_tuple_hash
*h
)
643 /* This is the conntrack entry already in hashes that won race. */
644 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
645 struct nf_conntrack_l4proto
*l4proto
;
647 l4proto
= __nf_ct_l4proto_find(nf_ct_l3num(ct
), nf_ct_protonum(ct
));
648 if (l4proto
->allow_clash
&&
649 !nf_ct_is_dying(ct
) &&
650 atomic_inc_not_zero(&ct
->ct_general
.use
)) {
651 nf_ct_acct_merge(ct
, ctinfo
, (struct nf_conn
*)skb
->nfct
);
652 nf_conntrack_put(skb
->nfct
);
653 /* Assign conntrack already in hashes to this skbuff. Don't
654 * modify skb->nfctinfo to ensure consistent stateful filtering.
656 skb
->nfct
= &ct
->ct_general
;
659 NF_CT_STAT_INC(net
, drop
);
663 /* Confirm a connection given skb; places it in hash table */
665 __nf_conntrack_confirm(struct sk_buff
*skb
)
667 const struct nf_conntrack_zone
*zone
;
668 unsigned int hash
, reply_hash
;
669 struct nf_conntrack_tuple_hash
*h
;
671 struct nf_conn_help
*help
;
672 struct nf_conn_tstamp
*tstamp
;
673 struct hlist_nulls_node
*n
;
674 enum ip_conntrack_info ctinfo
;
676 unsigned int sequence
;
679 ct
= nf_ct_get(skb
, &ctinfo
);
682 /* ipt_REJECT uses nf_conntrack_attach to attach related
683 ICMP/TCP RST packets in other direction. Actual packet
684 which created connection will be IP_CT_NEW or for an
685 expected connection, IP_CT_RELATED. */
686 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
689 zone
= nf_ct_zone(ct
);
693 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
694 /* reuse the hash saved before */
695 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
696 hash
= scale_hash(hash
);
697 reply_hash
= hash_conntrack(net
,
698 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
700 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
702 /* We're not in hash table, and we refuse to set up related
703 * connections for unconfirmed conns. But packet copies and
704 * REJECT will give spurious warnings here.
706 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
708 /* No external references means no one else could have
711 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
712 pr_debug("Confirming conntrack %p\n", ct
);
713 /* We have to check the DYING flag after unlink to prevent
714 * a race against nf_ct_get_next_corpse() possibly called from
715 * user context, else we insert an already 'dead' hash, blocking
716 * further use of that particular connection -JM.
718 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
720 if (unlikely(nf_ct_is_dying(ct
))) {
721 nf_ct_add_to_dying_list(ct
);
725 /* See if there's one in the list already, including reverse:
726 NAT could have grabbed it without realizing, since we're
727 not in the hash. If there is, we lost race. */
728 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
729 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
733 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
734 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
738 /* Timer relative to confirmation time, not original
739 setting time, otherwise we'd get timer wrap in
740 weird delay cases. */
741 ct
->timeout
.expires
+= jiffies
;
742 add_timer(&ct
->timeout
);
743 atomic_inc(&ct
->ct_general
.use
);
744 ct
->status
|= IPS_CONFIRMED
;
746 /* set conntrack timestamp, if enabled. */
747 tstamp
= nf_conn_tstamp_find(ct
);
749 if (skb
->tstamp
.tv64
== 0)
750 __net_timestamp(skb
);
752 tstamp
->start
= ktime_to_ns(skb
->tstamp
);
754 /* Since the lookup is lockless, hash insertion must be done after
755 * starting the timer and setting the CONFIRMED bit. The RCU barriers
756 * guarantee that no other CPU can find the conntrack before the above
757 * stores are visible.
759 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
760 nf_conntrack_double_unlock(hash
, reply_hash
);
761 NF_CT_STAT_INC(net
, insert
);
764 help
= nfct_help(ct
);
765 if (help
&& help
->helper
)
766 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
768 nf_conntrack_event_cache(master_ct(ct
) ?
769 IPCT_RELATED
: IPCT_NEW
, ct
);
773 nf_ct_add_to_dying_list(ct
);
774 ret
= nf_ct_resolve_clash(net
, skb
, ctinfo
, h
);
776 nf_conntrack_double_unlock(hash
, reply_hash
);
777 NF_CT_STAT_INC(net
, insert_failed
);
781 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
783 /* Returns true if a connection correspondings to the tuple (required
786 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
787 const struct nf_conn
*ignored_conntrack
)
789 struct net
*net
= nf_ct_net(ignored_conntrack
);
790 const struct nf_conntrack_zone
*zone
;
791 struct nf_conntrack_tuple_hash
*h
;
792 struct hlist_nulls_head
*ct_hash
;
793 unsigned int hash
, sequence
;
794 struct hlist_nulls_node
*n
;
797 zone
= nf_ct_zone(ignored_conntrack
);
801 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
802 hash
= hash_conntrack(net
, tuple
);
803 ct_hash
= nf_conntrack_hash
;
804 } while (read_seqcount_retry(&nf_conntrack_generation
, sequence
));
806 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
807 ct
= nf_ct_tuplehash_to_ctrack(h
);
808 if (ct
!= ignored_conntrack
&&
809 nf_ct_key_equal(h
, tuple
, zone
, net
)) {
810 NF_CT_STAT_INC_ATOMIC(net
, found
);
814 NF_CT_STAT_INC_ATOMIC(net
, searched
);
820 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
822 #define NF_CT_EVICTION_RANGE 8
824 /* There's a small race here where we may free a just-assured
825 connection. Too bad: we're in trouble anyway. */
826 static noinline
int early_drop(struct net
*net
, unsigned int _hash
)
828 /* Use oldest entry, which is roughly LRU */
829 struct nf_conntrack_tuple_hash
*h
;
831 struct hlist_nulls_node
*n
;
832 unsigned int i
, hash
, sequence
;
833 struct nf_conn
*ct
= NULL
;
841 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
842 for (; i
< NF_CT_EVICTION_RANGE
; i
++) {
843 hash
= scale_hash(_hash
++);
844 lockp
= &nf_conntrack_locks
[hash
% CONNTRACK_LOCKS
];
845 nf_conntrack_lock(lockp
);
846 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
850 hlist_nulls_for_each_entry_rcu(h
, n
, &nf_conntrack_hash
[hash
],
852 tmp
= nf_ct_tuplehash_to_ctrack(h
);
854 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
855 !net_eq(nf_ct_net(tmp
), net
) ||
859 if (atomic_inc_not_zero(&tmp
->ct_general
.use
)) {
875 /* kill only if in same netns -- might have moved due to
876 * SLAB_DESTROY_BY_RCU rules
878 if (net_eq(nf_ct_net(ct
), net
) && del_timer(&ct
->timeout
)) {
879 if (nf_ct_delete(ct
, 0, 0)) {
880 NF_CT_STAT_INC_ATOMIC(net
, early_drop
);
889 static struct nf_conn
*
890 __nf_conntrack_alloc(struct net
*net
,
891 const struct nf_conntrack_zone
*zone
,
892 const struct nf_conntrack_tuple
*orig
,
893 const struct nf_conntrack_tuple
*repl
,
898 /* We don't want any race condition at early drop stage */
899 atomic_inc(&net
->ct
.count
);
901 if (nf_conntrack_max
&&
902 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
903 if (!early_drop(net
, hash
)) {
904 atomic_dec(&net
->ct
.count
);
905 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
906 return ERR_PTR(-ENOMEM
);
911 * Do not use kmem_cache_zalloc(), as this cache uses
912 * SLAB_DESTROY_BY_RCU.
914 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
918 spin_lock_init(&ct
->lock
);
919 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
920 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
921 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
922 /* save hash for reusing when confirming */
923 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
925 /* Don't set timer yet: wait for confirmation */
926 setup_timer(&ct
->timeout
, death_by_timeout
, (unsigned long)ct
);
927 write_pnet(&ct
->ct_net
, net
);
928 memset(&ct
->__nfct_init_offset
[0], 0,
929 offsetof(struct nf_conn
, proto
) -
930 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
932 if (zone
&& nf_ct_zone_add(ct
, GFP_ATOMIC
, zone
) < 0)
935 /* Because we use RCU lookups, we set ct_general.use to zero before
936 * this is inserted in any list.
938 atomic_set(&ct
->ct_general
.use
, 0);
941 kmem_cache_free(nf_conntrack_cachep
, ct
);
943 atomic_dec(&net
->ct
.count
);
944 return ERR_PTR(-ENOMEM
);
947 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
948 const struct nf_conntrack_zone
*zone
,
949 const struct nf_conntrack_tuple
*orig
,
950 const struct nf_conntrack_tuple
*repl
,
953 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
955 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
957 void nf_conntrack_free(struct nf_conn
*ct
)
959 struct net
*net
= nf_ct_net(ct
);
961 /* A freed object has refcnt == 0, that's
962 * the golden rule for SLAB_DESTROY_BY_RCU
964 NF_CT_ASSERT(atomic_read(&ct
->ct_general
.use
) == 0);
966 nf_ct_ext_destroy(ct
);
968 kmem_cache_free(nf_conntrack_cachep
, ct
);
969 smp_mb__before_atomic();
970 atomic_dec(&net
->ct
.count
);
972 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
975 /* Allocate a new conntrack: we return -ENOMEM if classification
976 failed due to stress. Otherwise it really is unclassifiable. */
977 static struct nf_conntrack_tuple_hash
*
978 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
979 const struct nf_conntrack_tuple
*tuple
,
980 struct nf_conntrack_l3proto
*l3proto
,
981 struct nf_conntrack_l4proto
*l4proto
,
983 unsigned int dataoff
, u32 hash
)
986 struct nf_conn_help
*help
;
987 struct nf_conntrack_tuple repl_tuple
;
988 struct nf_conntrack_ecache
*ecache
;
989 struct nf_conntrack_expect
*exp
= NULL
;
990 const struct nf_conntrack_zone
*zone
;
991 struct nf_conn_timeout
*timeout_ext
;
992 struct nf_conntrack_zone tmp
;
993 unsigned int *timeouts
;
995 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
, l3proto
, l4proto
)) {
996 pr_debug("Can't invert tuple.\n");
1000 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1001 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
1004 return (struct nf_conntrack_tuple_hash
*)ct
;
1006 if (tmpl
&& nfct_synproxy(tmpl
)) {
1007 nfct_seqadj_ext_add(ct
);
1008 nfct_synproxy_ext_add(ct
);
1011 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1013 timeouts
= nf_ct_timeout_data(timeout_ext
);
1014 if (unlikely(!timeouts
))
1015 timeouts
= l4proto
->get_timeouts(net
);
1017 timeouts
= l4proto
->get_timeouts(net
);
1020 if (!l4proto
->new(ct
, skb
, dataoff
, timeouts
)) {
1021 nf_conntrack_free(ct
);
1022 pr_debug("can't track with proto module\n");
1027 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1030 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1031 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1032 nf_ct_labels_ext_add(ct
);
1034 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1035 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1036 ecache
? ecache
->expmask
: 0,
1040 if (net
->ct
.expect_count
) {
1041 spin_lock(&nf_conntrack_expect_lock
);
1042 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1044 pr_debug("expectation arrives ct=%p exp=%p\n",
1046 /* Welcome, Mr. Bond. We've been expecting you... */
1047 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1048 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1049 ct
->master
= exp
->master
;
1051 help
= nf_ct_helper_ext_add(ct
, exp
->helper
,
1054 rcu_assign_pointer(help
->helper
, exp
->helper
);
1057 #ifdef CONFIG_NF_CONNTRACK_MARK
1058 ct
->mark
= exp
->master
->mark
;
1060 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1061 ct
->secmark
= exp
->master
->secmark
;
1063 NF_CT_STAT_INC(net
, expect_new
);
1065 spin_unlock(&nf_conntrack_expect_lock
);
1068 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1069 NF_CT_STAT_INC(net
, new);
1072 /* Now it is inserted into the unconfirmed list, bump refcount */
1073 nf_conntrack_get(&ct
->ct_general
);
1074 nf_ct_add_to_unconfirmed_list(ct
);
1080 exp
->expectfn(ct
, exp
);
1081 nf_ct_expect_put(exp
);
1084 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1087 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1088 static inline struct nf_conn
*
1089 resolve_normal_ct(struct net
*net
, struct nf_conn
*tmpl
,
1090 struct sk_buff
*skb
,
1091 unsigned int dataoff
,
1094 struct nf_conntrack_l3proto
*l3proto
,
1095 struct nf_conntrack_l4proto
*l4proto
,
1097 enum ip_conntrack_info
*ctinfo
)
1099 const struct nf_conntrack_zone
*zone
;
1100 struct nf_conntrack_tuple tuple
;
1101 struct nf_conntrack_tuple_hash
*h
;
1102 struct nf_conntrack_zone tmp
;
1106 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1107 dataoff
, l3num
, protonum
, net
, &tuple
, l3proto
,
1109 pr_debug("Can't get tuple\n");
1113 /* look for tuple match */
1114 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1115 hash
= hash_conntrack_raw(&tuple
, net
);
1116 h
= __nf_conntrack_find_get(net
, zone
, &tuple
, hash
);
1118 h
= init_conntrack(net
, tmpl
, &tuple
, l3proto
, l4proto
,
1119 skb
, dataoff
, hash
);
1125 ct
= nf_ct_tuplehash_to_ctrack(h
);
1127 /* It exists; we have (non-exclusive) reference. */
1128 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1129 *ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1130 /* Please set reply bit if this packet OK */
1133 /* Once we've had two way comms, always ESTABLISHED. */
1134 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1135 pr_debug("normal packet for %p\n", ct
);
1136 *ctinfo
= IP_CT_ESTABLISHED
;
1137 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1138 pr_debug("related packet for %p\n", ct
);
1139 *ctinfo
= IP_CT_RELATED
;
1141 pr_debug("new packet for %p\n", ct
);
1142 *ctinfo
= IP_CT_NEW
;
1146 skb
->nfct
= &ct
->ct_general
;
1147 skb
->nfctinfo
= *ctinfo
;
1152 nf_conntrack_in(struct net
*net
, u_int8_t pf
, unsigned int hooknum
,
1153 struct sk_buff
*skb
)
1155 struct nf_conn
*ct
, *tmpl
= NULL
;
1156 enum ip_conntrack_info ctinfo
;
1157 struct nf_conntrack_l3proto
*l3proto
;
1158 struct nf_conntrack_l4proto
*l4proto
;
1159 unsigned int *timeouts
;
1160 unsigned int dataoff
;
1166 /* Previously seen (loopback or untracked)? Ignore. */
1167 tmpl
= (struct nf_conn
*)skb
->nfct
;
1168 if (!nf_ct_is_template(tmpl
)) {
1169 NF_CT_STAT_INC_ATOMIC(net
, ignore
);
1175 /* rcu_read_lock()ed by nf_hook_slow */
1176 l3proto
= __nf_ct_l3proto_find(pf
);
1177 ret
= l3proto
->get_l4proto(skb
, skb_network_offset(skb
),
1178 &dataoff
, &protonum
);
1180 pr_debug("not prepared to track yet or error occurred\n");
1181 NF_CT_STAT_INC_ATOMIC(net
, error
);
1182 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1187 l4proto
= __nf_ct_l4proto_find(pf
, protonum
);
1189 /* It may be an special packet, error, unclean...
1190 * inverse of the return code tells to the netfilter
1191 * core what to do with the packet. */
1192 if (l4proto
->error
!= NULL
) {
1193 ret
= l4proto
->error(net
, tmpl
, skb
, dataoff
, &ctinfo
,
1196 NF_CT_STAT_INC_ATOMIC(net
, error
);
1197 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1201 /* ICMP[v6] protocol trackers may assign one conntrack. */
1206 ct
= resolve_normal_ct(net
, tmpl
, skb
, dataoff
, pf
, protonum
,
1207 l3proto
, l4proto
, &set_reply
, &ctinfo
);
1209 /* Not valid part of a connection */
1210 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1216 /* Too stressed to deal. */
1217 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1222 NF_CT_ASSERT(skb
->nfct
);
1224 /* Decide what timeout policy we want to apply to this flow. */
1225 timeouts
= nf_ct_timeout_lookup(net
, ct
, l4proto
);
1227 ret
= l4proto
->packet(ct
, skb
, dataoff
, ctinfo
, pf
, hooknum
, timeouts
);
1229 /* Invalid: inverse of the return code tells
1230 * the netfilter core what to do */
1231 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1232 nf_conntrack_put(skb
->nfct
);
1234 NF_CT_STAT_INC_ATOMIC(net
, invalid
);
1235 if (ret
== -NF_DROP
)
1236 NF_CT_STAT_INC_ATOMIC(net
, drop
);
1241 if (set_reply
&& !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1242 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1245 /* Special case: we have to repeat this hook, assign the
1246 * template again to this packet. We assume that this packet
1247 * has no conntrack assigned. This is used by nf_ct_tcp. */
1248 if (ret
== NF_REPEAT
)
1249 skb
->nfct
= (struct nf_conntrack
*)tmpl
;
1256 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1258 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple
*inverse
,
1259 const struct nf_conntrack_tuple
*orig
)
1264 ret
= nf_ct_invert_tuple(inverse
, orig
,
1265 __nf_ct_l3proto_find(orig
->src
.l3num
),
1266 __nf_ct_l4proto_find(orig
->src
.l3num
,
1267 orig
->dst
.protonum
));
1271 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr
);
1273 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1274 implicitly racy: see __nf_conntrack_confirm */
1275 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1276 const struct nf_conntrack_tuple
*newreply
)
1278 struct nf_conn_help
*help
= nfct_help(ct
);
1280 /* Should be unconfirmed, so not in hash table yet */
1281 NF_CT_ASSERT(!nf_ct_is_confirmed(ct
));
1283 pr_debug("Altering reply tuple of %p to ", ct
);
1284 nf_ct_dump_tuple(newreply
);
1286 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1287 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1291 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1294 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1296 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1297 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1298 enum ip_conntrack_info ctinfo
,
1299 const struct sk_buff
*skb
,
1300 unsigned long extra_jiffies
,
1303 NF_CT_ASSERT(ct
->timeout
.data
== (unsigned long)ct
);
1306 /* Only update if this is not a fixed timeout */
1307 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1310 /* If not in hash table, timer will not be active yet */
1311 if (!nf_ct_is_confirmed(ct
)) {
1312 ct
->timeout
.expires
= extra_jiffies
;
1314 unsigned long newtime
= jiffies
+ extra_jiffies
;
1316 /* Only update the timeout if the new timeout is at least
1317 HZ jiffies from the old timeout. Need del_timer for race
1318 avoidance (may already be dying). */
1319 if (newtime
- ct
->timeout
.expires
>= HZ
)
1320 mod_timer_pending(&ct
->timeout
, newtime
);
1325 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1327 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1329 bool __nf_ct_kill_acct(struct nf_conn
*ct
,
1330 enum ip_conntrack_info ctinfo
,
1331 const struct sk_buff
*skb
,
1335 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1337 if (del_timer(&ct
->timeout
)) {
1338 ct
->timeout
.function((unsigned long)ct
);
1343 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct
);
1345 #ifdef CONFIG_NF_CONNTRACK_ZONES
1346 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly
= {
1347 .len
= sizeof(struct nf_conntrack_zone
),
1348 .align
= __alignof__(struct nf_conntrack_zone
),
1349 .id
= NF_CT_EXT_ZONE
,
1353 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1355 #include <linux/netfilter/nfnetlink.h>
1356 #include <linux/netfilter/nfnetlink_conntrack.h>
1357 #include <linux/mutex.h>
1359 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1360 * in ip_conntrack_core, since we don't want the protocols to autoload
1361 * or depend on ctnetlink */
1362 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1363 const struct nf_conntrack_tuple
*tuple
)
1365 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1366 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1367 goto nla_put_failure
;
1373 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1375 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1376 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1377 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1379 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1381 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1382 struct nf_conntrack_tuple
*t
)
1384 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1387 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1388 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1392 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1394 int nf_ct_port_nlattr_tuple_size(void)
1396 return nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1398 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1401 /* Used by ipt_REJECT and ip6t_REJECT. */
1402 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1405 enum ip_conntrack_info ctinfo
;
1407 /* This ICMP is in reverse direction to the packet which caused it */
1408 ct
= nf_ct_get(skb
, &ctinfo
);
1409 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1410 ctinfo
= IP_CT_RELATED_REPLY
;
1412 ctinfo
= IP_CT_RELATED
;
1414 /* Attach to new skbuff, and increment count */
1415 nskb
->nfct
= &ct
->ct_general
;
1416 nskb
->nfctinfo
= ctinfo
;
1417 nf_conntrack_get(nskb
->nfct
);
1420 /* Bring out ya dead! */
1421 static struct nf_conn
*
1422 get_next_corpse(struct net
*net
, int (*iter
)(struct nf_conn
*i
, void *data
),
1423 void *data
, unsigned int *bucket
)
1425 struct nf_conntrack_tuple_hash
*h
;
1427 struct hlist_nulls_node
*n
;
1431 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
1432 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
1434 nf_conntrack_lock(lockp
);
1435 if (*bucket
< nf_conntrack_htable_size
) {
1436 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
1437 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
1439 ct
= nf_ct_tuplehash_to_ctrack(h
);
1440 if (net_eq(nf_ct_net(ct
), net
) &&
1450 for_each_possible_cpu(cpu
) {
1451 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1453 spin_lock_bh(&pcpu
->lock
);
1454 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
1455 ct
= nf_ct_tuplehash_to_ctrack(h
);
1457 set_bit(IPS_DYING_BIT
, &ct
->status
);
1459 spin_unlock_bh(&pcpu
->lock
);
1464 atomic_inc(&ct
->ct_general
.use
);
1470 void nf_ct_iterate_cleanup(struct net
*net
,
1471 int (*iter
)(struct nf_conn
*i
, void *data
),
1472 void *data
, u32 portid
, int report
)
1475 unsigned int bucket
= 0;
1479 if (atomic_read(&net
->ct
.count
) == 0)
1482 while ((ct
= get_next_corpse(net
, iter
, data
, &bucket
)) != NULL
) {
1483 /* Time to push up daises... */
1484 if (del_timer(&ct
->timeout
))
1485 nf_ct_delete(ct
, portid
, report
);
1487 /* ... else the timer will get him soon. */
1493 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup
);
1495 static int kill_all(struct nf_conn
*i
, void *data
)
1500 void nf_ct_free_hashtable(void *hash
, unsigned int size
)
1502 if (is_vmalloc_addr(hash
))
1505 free_pages((unsigned long)hash
,
1506 get_order(sizeof(struct hlist_head
) * size
));
1508 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable
);
1510 static int untrack_refs(void)
1514 for_each_possible_cpu(cpu
) {
1515 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1517 cnt
+= atomic_read(&ct
->ct_general
.use
) - 1;
1522 void nf_conntrack_cleanup_start(void)
1524 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
1527 void nf_conntrack_cleanup_end(void)
1529 RCU_INIT_POINTER(nf_ct_destroy
, NULL
);
1530 while (untrack_refs() > 0)
1533 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1535 #ifdef CONFIG_NF_CONNTRACK_ZONES
1536 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1538 nf_conntrack_proto_fini();
1539 nf_conntrack_seqadj_fini();
1540 nf_conntrack_labels_fini();
1541 nf_conntrack_helper_fini();
1542 nf_conntrack_timeout_fini();
1543 nf_conntrack_ecache_fini();
1544 nf_conntrack_tstamp_fini();
1545 nf_conntrack_acct_fini();
1546 nf_conntrack_expect_fini();
1550 * Mishearing the voices in his head, our hero wonders how he's
1551 * supposed to kill the mall.
1553 void nf_conntrack_cleanup_net(struct net
*net
)
1557 list_add(&net
->exit_list
, &single
);
1558 nf_conntrack_cleanup_net_list(&single
);
1561 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
1567 * This makes sure all current packets have passed through
1568 * netfilter framework. Roll on, two-stage module
1574 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1575 nf_ct_iterate_cleanup(net
, kill_all
, NULL
, 0, 0);
1576 if (atomic_read(&net
->ct
.count
) != 0)
1581 goto i_see_dead_people
;
1584 list_for_each_entry(net
, net_exit_list
, exit_list
) {
1585 nf_conntrack_proto_pernet_fini(net
);
1586 nf_conntrack_helper_pernet_fini(net
);
1587 nf_conntrack_ecache_pernet_fini(net
);
1588 nf_conntrack_tstamp_pernet_fini(net
);
1589 nf_conntrack_acct_pernet_fini(net
);
1590 nf_conntrack_expect_pernet_fini(net
);
1591 free_percpu(net
->ct
.stat
);
1592 free_percpu(net
->ct
.pcpu_lists
);
1596 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
1598 struct hlist_nulls_head
*hash
;
1599 unsigned int nr_slots
, i
;
1602 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
1603 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
1604 sz
= nr_slots
* sizeof(struct hlist_nulls_head
);
1605 hash
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_NOWARN
| __GFP_ZERO
,
1611 for (i
= 0; i
< nr_slots
; i
++)
1612 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
1616 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
1618 int nf_conntrack_set_hashsize(const char *val
, struct kernel_param
*kp
)
1621 unsigned int hashsize
, old_size
;
1622 struct hlist_nulls_head
*hash
, *old_hash
;
1623 struct nf_conntrack_tuple_hash
*h
;
1626 if (current
->nsproxy
->net_ns
!= &init_net
)
1629 /* On boot, we can set this without any fancy locking. */
1630 if (!nf_conntrack_htable_size
)
1631 return param_set_uint(val
, kp
);
1633 rc
= kstrtouint(val
, 0, &hashsize
);
1639 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
1644 nf_conntrack_all_lock();
1645 write_seqcount_begin(&nf_conntrack_generation
);
1647 /* Lookups in the old hash might happen in parallel, which means we
1648 * might get false negatives during connection lookup. New connections
1649 * created because of a false negative won't make it into the hash
1650 * though since that required taking the locks.
1653 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
1654 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
1655 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
1656 struct nf_conntrack_tuple_hash
, hnnode
);
1657 ct
= nf_ct_tuplehash_to_ctrack(h
);
1658 hlist_nulls_del_rcu(&h
->hnnode
);
1659 bucket
= __hash_conntrack(nf_ct_net(ct
),
1660 &h
->tuple
, hashsize
);
1661 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
1664 old_size
= nf_conntrack_htable_size
;
1665 old_hash
= nf_conntrack_hash
;
1667 nf_conntrack_hash
= hash
;
1668 nf_conntrack_htable_size
= hashsize
;
1670 write_seqcount_end(&nf_conntrack_generation
);
1671 nf_conntrack_all_unlock();
1675 nf_ct_free_hashtable(old_hash
, old_size
);
1678 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
1680 module_param_call(hashsize
, nf_conntrack_set_hashsize
, param_get_uint
,
1681 &nf_conntrack_htable_size
, 0600);
1683 void nf_ct_untracked_status_or(unsigned long bits
)
1687 for_each_possible_cpu(cpu
)
1688 per_cpu(nf_conntrack_untracked
, cpu
).status
|= bits
;
1690 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or
);
1692 int nf_conntrack_init_start(void)
1698 seqcount_init(&nf_conntrack_generation
);
1700 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
1701 spin_lock_init(&nf_conntrack_locks
[i
]);
1703 if (!nf_conntrack_htable_size
) {
1704 /* Idea from tcp.c: use 1/16384 of memory.
1705 * On i386: 32MB machine has 512 buckets.
1706 * >= 1GB machines have 16384 buckets.
1707 * >= 4GB machines have 65536 buckets.
1709 nf_conntrack_htable_size
1710 = (((totalram_pages
<< PAGE_SHIFT
) / 16384)
1711 / sizeof(struct hlist_head
));
1712 if (totalram_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
1713 nf_conntrack_htable_size
= 65536;
1714 else if (totalram_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
1715 nf_conntrack_htable_size
= 16384;
1716 if (nf_conntrack_htable_size
< 32)
1717 nf_conntrack_htable_size
= 32;
1719 /* Use a max. factor of four by default to get the same max as
1720 * with the old struct list_heads. When a table size is given
1721 * we use the old value of 8 to avoid reducing the max.
1726 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
1727 if (!nf_conntrack_hash
)
1730 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
1732 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
1733 sizeof(struct nf_conn
), 0,
1734 SLAB_DESTROY_BY_RCU
, NULL
);
1735 if (!nf_conntrack_cachep
)
1738 printk(KERN_INFO
"nf_conntrack version %s (%u buckets, %d max)\n",
1739 NF_CONNTRACK_VERSION
, nf_conntrack_htable_size
,
1742 ret
= nf_conntrack_expect_init();
1746 ret
= nf_conntrack_acct_init();
1750 ret
= nf_conntrack_tstamp_init();
1754 ret
= nf_conntrack_ecache_init();
1758 ret
= nf_conntrack_timeout_init();
1762 ret
= nf_conntrack_helper_init();
1766 ret
= nf_conntrack_labels_init();
1770 ret
= nf_conntrack_seqadj_init();
1774 #ifdef CONFIG_NF_CONNTRACK_ZONES
1775 ret
= nf_ct_extend_register(&nf_ct_zone_extend
);
1779 ret
= nf_conntrack_proto_init();
1783 /* Set up fake conntrack: to never be deleted, not in any hashes */
1784 for_each_possible_cpu(cpu
) {
1785 struct nf_conn
*ct
= &per_cpu(nf_conntrack_untracked
, cpu
);
1786 write_pnet(&ct
->ct_net
, &init_net
);
1787 atomic_set(&ct
->ct_general
.use
, 1);
1789 /* - and look it like as a confirmed connection */
1790 nf_ct_untracked_status_or(IPS_CONFIRMED
| IPS_UNTRACKED
);
1794 #ifdef CONFIG_NF_CONNTRACK_ZONES
1795 nf_ct_extend_unregister(&nf_ct_zone_extend
);
1798 nf_conntrack_seqadj_fini();
1800 nf_conntrack_labels_fini();
1802 nf_conntrack_helper_fini();
1804 nf_conntrack_timeout_fini();
1806 nf_conntrack_ecache_fini();
1808 nf_conntrack_tstamp_fini();
1810 nf_conntrack_acct_fini();
1812 nf_conntrack_expect_fini();
1814 kmem_cache_destroy(nf_conntrack_cachep
);
1816 nf_ct_free_hashtable(nf_conntrack_hash
, nf_conntrack_htable_size
);
1820 void nf_conntrack_init_end(void)
1822 /* For use by REJECT target */
1823 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
1824 RCU_INIT_POINTER(nf_ct_destroy
, destroy_conntrack
);
1828 * We need to use special "null" values, not used in hash table
1830 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1831 #define DYING_NULLS_VAL ((1<<30)+1)
1832 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1834 int nf_conntrack_init_net(struct net
*net
)
1839 atomic_set(&net
->ct
.count
, 0);
1841 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
1842 if (!net
->ct
.pcpu_lists
)
1845 for_each_possible_cpu(cpu
) {
1846 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
1848 spin_lock_init(&pcpu
->lock
);
1849 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
1850 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
1853 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
1855 goto err_pcpu_lists
;
1857 ret
= nf_conntrack_expect_pernet_init(net
);
1860 ret
= nf_conntrack_acct_pernet_init(net
);
1863 ret
= nf_conntrack_tstamp_pernet_init(net
);
1866 ret
= nf_conntrack_ecache_pernet_init(net
);
1869 ret
= nf_conntrack_helper_pernet_init(net
);
1872 ret
= nf_conntrack_proto_pernet_init(net
);
1878 nf_conntrack_helper_pernet_fini(net
);
1880 nf_conntrack_ecache_pernet_fini(net
);
1882 nf_conntrack_tstamp_pernet_fini(net
);
1884 nf_conntrack_acct_pernet_fini(net
);
1886 nf_conntrack_expect_pernet_fini(net
);
1888 free_percpu(net
->ct
.stat
);
1890 free_percpu(net
->ct
.pcpu_lists
);