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/siphash.h>
29 #include <linux/err.h>
30 #include <linux/percpu.h>
31 #include <linux/moduleparam.h>
32 #include <linux/notifier.h>
33 #include <linux/kernel.h>
34 #include <linux/netdevice.h>
35 #include <linux/socket.h>
37 #include <linux/nsproxy.h>
38 #include <linux/rculist_nulls.h>
40 #include <net/netfilter/nf_conntrack.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_helper.h>
56 #include <net/netns/hash.h>
59 #include "nf_internals.h"
61 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks
[CONNTRACK_LOCKS
];
62 EXPORT_SYMBOL_GPL(nf_conntrack_locks
);
64 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(nf_conntrack_expect_lock
);
65 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock
);
67 struct hlist_nulls_head
*nf_conntrack_hash __read_mostly
;
68 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
70 struct conntrack_gc_work
{
71 struct delayed_work dwork
;
78 static __read_mostly
struct kmem_cache
*nf_conntrack_cachep
;
79 static __read_mostly spinlock_t nf_conntrack_locks_all_lock
;
80 static __read_mostly
DEFINE_SPINLOCK(nf_conntrack_locks_all_lock
);
81 static __read_mostly
bool nf_conntrack_locks_all
;
83 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
84 #define GC_MAX_BUCKETS_DIV 128u
85 /* upper bound of full table scan */
86 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
87 /* desired ratio of entries found to be expired */
88 #define GC_EVICT_RATIO 50u
90 static struct conntrack_gc_work conntrack_gc_work
;
92 void nf_conntrack_lock(spinlock_t
*lock
) __acquires(lock
)
94 /* 1) Acquire the lock */
97 /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
98 * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
100 if (likely(smp_load_acquire(&nf_conntrack_locks_all
) == false))
103 /* fast path failed, unlock */
106 /* Slow path 1) get global lock */
107 spin_lock(&nf_conntrack_locks_all_lock
);
109 /* Slow path 2) get the lock we want */
112 /* Slow path 3) release the global lock */
113 spin_unlock(&nf_conntrack_locks_all_lock
);
115 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
117 static void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
119 h1
%= CONNTRACK_LOCKS
;
120 h2
%= CONNTRACK_LOCKS
;
121 spin_unlock(&nf_conntrack_locks
[h1
]);
123 spin_unlock(&nf_conntrack_locks
[h2
]);
126 /* return true if we need to recompute hashes (in case hash table was resized) */
127 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
128 unsigned int h2
, unsigned int sequence
)
130 h1
%= CONNTRACK_LOCKS
;
131 h2
%= CONNTRACK_LOCKS
;
133 nf_conntrack_lock(&nf_conntrack_locks
[h1
]);
135 spin_lock_nested(&nf_conntrack_locks
[h2
],
136 SINGLE_DEPTH_NESTING
);
138 nf_conntrack_lock(&nf_conntrack_locks
[h2
]);
139 spin_lock_nested(&nf_conntrack_locks
[h1
],
140 SINGLE_DEPTH_NESTING
);
142 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
143 nf_conntrack_double_unlock(h1
, h2
);
149 static void nf_conntrack_all_lock(void)
153 spin_lock(&nf_conntrack_locks_all_lock
);
155 nf_conntrack_locks_all
= true;
157 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++) {
158 spin_lock(&nf_conntrack_locks
[i
]);
160 /* This spin_unlock provides the "release" to ensure that
161 * nf_conntrack_locks_all==true is visible to everyone that
162 * acquired spin_lock(&nf_conntrack_locks[]).
164 spin_unlock(&nf_conntrack_locks
[i
]);
168 static void nf_conntrack_all_unlock(void)
170 /* All prior stores must be complete before we clear
171 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
172 * might observe the false value but not the entire
174 * It pairs with the smp_load_acquire() in nf_conntrack_lock()
176 smp_store_release(&nf_conntrack_locks_all
, false);
177 spin_unlock(&nf_conntrack_locks_all_lock
);
180 unsigned int nf_conntrack_htable_size __read_mostly
;
181 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
183 unsigned int nf_conntrack_max __read_mostly
;
184 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
185 seqcount_t nf_conntrack_generation __read_mostly
;
186 static unsigned int nf_conntrack_hash_rnd __read_mostly
;
188 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
,
189 const struct net
*net
)
194 get_random_once(&nf_conntrack_hash_rnd
, sizeof(nf_conntrack_hash_rnd
));
196 /* The direction must be ignored, so we hash everything up to the
197 * destination ports (which is a multiple of 4) and treat the last
198 * three bytes manually.
200 seed
= nf_conntrack_hash_rnd
^ net_hash_mix(net
);
201 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
202 return jhash2((u32
*)tuple
, n
, seed
^
203 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
204 tuple
->dst
.protonum
));
207 static u32
scale_hash(u32 hash
)
209 return reciprocal_scale(hash
, nf_conntrack_htable_size
);
212 static u32
__hash_conntrack(const struct net
*net
,
213 const struct nf_conntrack_tuple
*tuple
,
216 return reciprocal_scale(hash_conntrack_raw(tuple
, net
), size
);
219 static u32
hash_conntrack(const struct net
*net
,
220 const struct nf_conntrack_tuple
*tuple
)
222 return scale_hash(hash_conntrack_raw(tuple
, net
));
225 static bool nf_ct_get_tuple_ports(const struct sk_buff
*skb
,
226 unsigned int dataoff
,
227 struct nf_conntrack_tuple
*tuple
)
231 } _inet_hdr
, *inet_hdr
;
233 /* Actually only need first 4 bytes to get ports. */
234 inet_hdr
= skb_header_pointer(skb
, dataoff
, sizeof(_inet_hdr
), &_inet_hdr
);
238 tuple
->src
.u
.udp
.port
= inet_hdr
->sport
;
239 tuple
->dst
.u
.udp
.port
= inet_hdr
->dport
;
244 nf_ct_get_tuple(const struct sk_buff
*skb
,
246 unsigned int dataoff
,
250 struct nf_conntrack_tuple
*tuple
)
256 memset(tuple
, 0, sizeof(*tuple
));
258 tuple
->src
.l3num
= l3num
;
261 nhoff
+= offsetof(struct iphdr
, saddr
);
262 size
= 2 * sizeof(__be32
);
265 nhoff
+= offsetof(struct ipv6hdr
, saddr
);
266 size
= sizeof(_addrs
);
272 ap
= skb_header_pointer(skb
, nhoff
, size
, _addrs
);
278 tuple
->src
.u3
.ip
= ap
[0];
279 tuple
->dst
.u3
.ip
= ap
[1];
282 memcpy(tuple
->src
.u3
.ip6
, ap
, sizeof(tuple
->src
.u3
.ip6
));
283 memcpy(tuple
->dst
.u3
.ip6
, ap
+ 4, sizeof(tuple
->dst
.u3
.ip6
));
287 tuple
->dst
.protonum
= protonum
;
288 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
291 #if IS_ENABLED(CONFIG_IPV6)
293 return icmpv6_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
296 return icmp_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
297 #ifdef CONFIG_NF_CT_PROTO_GRE
299 return gre_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
302 case IPPROTO_UDP
: /* fallthrough */
303 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
304 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
305 case IPPROTO_UDPLITE
:
306 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
308 #ifdef CONFIG_NF_CT_PROTO_SCTP
310 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
312 #ifdef CONFIG_NF_CT_PROTO_DCCP
314 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
323 static int ipv4_get_l4proto(const struct sk_buff
*skb
, unsigned int nhoff
,
327 const struct iphdr
*iph
;
330 iph
= skb_header_pointer(skb
, nhoff
, sizeof(_iph
), &_iph
);
334 /* Conntrack defragments packets, we might still see fragments
335 * inside ICMP packets though.
337 if (iph
->frag_off
& htons(IP_OFFSET
))
340 dataoff
= nhoff
+ (iph
->ihl
<< 2);
341 *protonum
= iph
->protocol
;
343 /* Check bogus IP headers */
344 if (dataoff
> skb
->len
) {
345 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
346 nhoff
, iph
->ihl
<< 2, skb
->len
);
352 #if IS_ENABLED(CONFIG_IPV6)
353 static int ipv6_get_l4proto(const struct sk_buff
*skb
, unsigned int nhoff
,
357 unsigned int extoff
= nhoff
+ sizeof(struct ipv6hdr
);
361 if (skb_copy_bits(skb
, nhoff
+ offsetof(struct ipv6hdr
, nexthdr
),
362 &nexthdr
, sizeof(nexthdr
)) != 0) {
363 pr_debug("can't get nexthdr\n");
366 protoff
= ipv6_skip_exthdr(skb
, extoff
, &nexthdr
, &frag_off
);
368 * (protoff == skb->len) means the packet has not data, just
369 * IPv6 and possibly extensions headers, but it is tracked anyway
371 if (protoff
< 0 || (frag_off
& htons(~0x7)) != 0) {
372 pr_debug("can't find proto in pkt\n");
381 static int get_l4proto(const struct sk_buff
*skb
,
382 unsigned int nhoff
, u8 pf
, u8
*l4num
)
386 return ipv4_get_l4proto(skb
, nhoff
, l4num
);
387 #if IS_ENABLED(CONFIG_IPV6)
389 return ipv6_get_l4proto(skb
, nhoff
, l4num
);
398 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
400 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
405 protoff
= get_l4proto(skb
, nhoff
, l3num
, &protonum
);
409 return nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
);
411 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
414 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
415 const struct nf_conntrack_tuple
*orig
)
417 memset(inverse
, 0, sizeof(*inverse
));
419 inverse
->src
.l3num
= orig
->src
.l3num
;
421 switch (orig
->src
.l3num
) {
423 inverse
->src
.u3
.ip
= orig
->dst
.u3
.ip
;
424 inverse
->dst
.u3
.ip
= orig
->src
.u3
.ip
;
427 inverse
->src
.u3
.in6
= orig
->dst
.u3
.in6
;
428 inverse
->dst
.u3
.in6
= orig
->src
.u3
.in6
;
434 inverse
->dst
.dir
= !orig
->dst
.dir
;
436 inverse
->dst
.protonum
= orig
->dst
.protonum
;
438 switch (orig
->dst
.protonum
) {
440 return nf_conntrack_invert_icmp_tuple(inverse
, orig
);
441 #if IS_ENABLED(CONFIG_IPV6)
443 return nf_conntrack_invert_icmpv6_tuple(inverse
, orig
);
447 inverse
->src
.u
.all
= orig
->dst
.u
.all
;
448 inverse
->dst
.u
.all
= orig
->src
.u
.all
;
451 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
453 /* Generate a almost-unique pseudo-id for a given conntrack.
455 * intentionally doesn't re-use any of the seeds used for hash
456 * table location, we assume id gets exposed to userspace.
458 * Following nf_conn items do not change throughout lifetime
459 * of the nf_conn after it has been committed to main hash table:
462 * 2. nf_conn->ext address
463 * 3. nf_conn->master address (normally NULL)
465 * 5. the associated net namespace
467 u32
nf_ct_get_id(const struct nf_conn
*ct
)
469 static __read_mostly siphash_key_t ct_id_seed
;
470 unsigned long a
, b
, c
, d
;
472 net_get_random_once(&ct_id_seed
, sizeof(ct_id_seed
));
474 a
= (unsigned long)ct
;
475 b
= (unsigned long)ct
->master
^ net_hash_mix(nf_ct_net(ct
));
476 c
= (unsigned long)ct
->ext
;
477 d
= (unsigned long)siphash(&ct
->tuplehash
, sizeof(ct
->tuplehash
),
480 return siphash_4u64((u64
)a
, (u64
)b
, (u64
)c
, (u64
)d
, &ct_id_seed
);
482 return siphash_4u32((u32
)a
, (u32
)b
, (u32
)c
, (u32
)d
, &ct_id_seed
);
485 EXPORT_SYMBOL_GPL(nf_ct_get_id
);
488 clean_from_lists(struct nf_conn
*ct
)
490 pr_debug("clean_from_lists(%p)\n", ct
);
491 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
492 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
494 /* Destroy all pending expectations */
495 nf_ct_remove_expectations(ct
);
498 /* must be called with local_bh_disable */
499 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
501 struct ct_pcpu
*pcpu
;
503 /* add this conntrack to the (per cpu) dying list */
504 ct
->cpu
= smp_processor_id();
505 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
507 spin_lock(&pcpu
->lock
);
508 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
510 spin_unlock(&pcpu
->lock
);
513 /* must be called with local_bh_disable */
514 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
516 struct ct_pcpu
*pcpu
;
518 /* add this conntrack to the (per cpu) unconfirmed list */
519 ct
->cpu
= smp_processor_id();
520 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
522 spin_lock(&pcpu
->lock
);
523 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
525 spin_unlock(&pcpu
->lock
);
528 /* must be called with local_bh_disable */
529 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
531 struct ct_pcpu
*pcpu
;
533 /* We overload first tuple to link into unconfirmed or dying list.*/
534 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
536 spin_lock(&pcpu
->lock
);
537 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
538 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
539 spin_unlock(&pcpu
->lock
);
542 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
544 /* Released via destroy_conntrack() */
545 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
546 const struct nf_conntrack_zone
*zone
,
549 struct nf_conn
*tmpl
, *p
;
551 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
) {
552 tmpl
= kzalloc(sizeof(*tmpl
) + NFCT_INFOMASK
, flags
);
557 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
559 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
560 tmpl
->proto
.tmpl_padto
= (char *)tmpl
- (char *)p
;
563 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
568 tmpl
->status
= IPS_TEMPLATE
;
569 write_pnet(&tmpl
->ct_net
, net
);
570 nf_ct_zone_add(tmpl
, zone
);
571 atomic_set(&tmpl
->ct_general
.use
, 0);
575 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
577 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
579 nf_ct_ext_destroy(tmpl
);
580 nf_ct_ext_free(tmpl
);
582 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
)
583 kfree((char *)tmpl
- tmpl
->proto
.tmpl_padto
);
587 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
589 static void destroy_gre_conntrack(struct nf_conn
*ct
)
591 #ifdef CONFIG_NF_CT_PROTO_GRE
592 struct nf_conn
*master
= ct
->master
;
595 nf_ct_gre_keymap_destroy(master
);
600 destroy_conntrack(struct nf_conntrack
*nfct
)
602 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
604 pr_debug("destroy_conntrack(%p)\n", ct
);
605 WARN_ON(atomic_read(&nfct
->use
) != 0);
607 if (unlikely(nf_ct_is_template(ct
))) {
612 if (unlikely(nf_ct_protonum(ct
) == IPPROTO_GRE
))
613 destroy_gre_conntrack(ct
);
616 /* Expectations will have been removed in clean_from_lists,
617 * except TFTP can create an expectation on the first packet,
618 * before connection is in the list, so we need to clean here,
621 nf_ct_remove_expectations(ct
);
623 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
628 nf_ct_put(ct
->master
);
630 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
631 nf_conntrack_free(ct
);
634 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
636 struct net
*net
= nf_ct_net(ct
);
637 unsigned int hash
, reply_hash
;
638 unsigned int sequence
;
640 nf_ct_helper_destroy(ct
);
644 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
645 hash
= hash_conntrack(net
,
646 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
647 reply_hash
= hash_conntrack(net
,
648 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
649 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
651 clean_from_lists(ct
);
652 nf_conntrack_double_unlock(hash
, reply_hash
);
654 nf_ct_add_to_dying_list(ct
);
659 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
661 struct nf_conn_tstamp
*tstamp
;
663 if (test_and_set_bit(IPS_DYING_BIT
, &ct
->status
))
666 tstamp
= nf_conn_tstamp_find(ct
);
667 if (tstamp
&& tstamp
->stop
== 0)
668 tstamp
->stop
= ktime_get_real_ns();
670 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
671 portid
, report
) < 0) {
672 /* destroy event was not delivered. nf_ct_put will
673 * be done by event cache worker on redelivery.
675 nf_ct_delete_from_lists(ct
);
676 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
680 nf_conntrack_ecache_work(nf_ct_net(ct
));
681 nf_ct_delete_from_lists(ct
);
685 EXPORT_SYMBOL_GPL(nf_ct_delete
);
688 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
689 const struct nf_conntrack_tuple
*tuple
,
690 const struct nf_conntrack_zone
*zone
,
691 const struct net
*net
)
693 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
695 /* A conntrack can be recreated with the equal tuple,
696 * so we need to check that the conntrack is confirmed
698 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
699 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
700 nf_ct_is_confirmed(ct
) &&
701 net_eq(net
, nf_ct_net(ct
));
705 nf_ct_match(const struct nf_conn
*ct1
, const struct nf_conn
*ct2
)
707 return nf_ct_tuple_equal(&ct1
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
708 &ct2
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
) &&
709 nf_ct_tuple_equal(&ct1
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
710 &ct2
->tuplehash
[IP_CT_DIR_REPLY
].tuple
) &&
711 nf_ct_zone_equal(ct1
, nf_ct_zone(ct2
), IP_CT_DIR_ORIGINAL
) &&
712 nf_ct_zone_equal(ct1
, nf_ct_zone(ct2
), IP_CT_DIR_REPLY
) &&
713 net_eq(nf_ct_net(ct1
), nf_ct_net(ct2
));
716 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
717 static void nf_ct_gc_expired(struct nf_conn
*ct
)
719 if (!atomic_inc_not_zero(&ct
->ct_general
.use
))
722 if (nf_ct_should_gc(ct
))
730 * - Caller must take a reference on returned object
731 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
733 static struct nf_conntrack_tuple_hash
*
734 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
735 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
737 struct nf_conntrack_tuple_hash
*h
;
738 struct hlist_nulls_head
*ct_hash
;
739 struct hlist_nulls_node
*n
;
740 unsigned int bucket
, hsize
;
743 nf_conntrack_get_ht(&ct_hash
, &hsize
);
744 bucket
= reciprocal_scale(hash
, hsize
);
746 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[bucket
], hnnode
) {
749 ct
= nf_ct_tuplehash_to_ctrack(h
);
750 if (nf_ct_is_expired(ct
)) {
751 nf_ct_gc_expired(ct
);
755 if (nf_ct_is_dying(ct
))
758 if (nf_ct_key_equal(h
, tuple
, zone
, net
))
762 * if the nulls value we got at the end of this lookup is
763 * not the expected one, we must restart lookup.
764 * We probably met an item that was moved to another chain.
766 if (get_nulls_value(n
) != bucket
) {
767 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
774 /* Find a connection corresponding to a tuple. */
775 static struct nf_conntrack_tuple_hash
*
776 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
777 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
779 struct nf_conntrack_tuple_hash
*h
;
784 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
786 ct
= nf_ct_tuplehash_to_ctrack(h
);
787 if (unlikely(nf_ct_is_dying(ct
) ||
788 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
791 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
, net
))) {
802 struct nf_conntrack_tuple_hash
*
803 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
804 const struct nf_conntrack_tuple
*tuple
)
806 return __nf_conntrack_find_get(net
, zone
, tuple
,
807 hash_conntrack_raw(tuple
, net
));
809 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
811 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
813 unsigned int reply_hash
)
815 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
816 &nf_conntrack_hash
[hash
]);
817 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
818 &nf_conntrack_hash
[reply_hash
]);
822 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
824 const struct nf_conntrack_zone
*zone
;
825 struct net
*net
= nf_ct_net(ct
);
826 unsigned int hash
, reply_hash
;
827 struct nf_conntrack_tuple_hash
*h
;
828 struct hlist_nulls_node
*n
;
829 unsigned int sequence
;
831 zone
= nf_ct_zone(ct
);
835 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
836 hash
= hash_conntrack(net
,
837 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
838 reply_hash
= hash_conntrack(net
,
839 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
840 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
842 /* See if there's one in the list already, including reverse */
843 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
844 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
848 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
849 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
854 /* The caller holds a reference to this object */
855 atomic_set(&ct
->ct_general
.use
, 2);
856 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
857 nf_conntrack_double_unlock(hash
, reply_hash
);
858 NF_CT_STAT_INC(net
, insert
);
863 nf_conntrack_double_unlock(hash
, reply_hash
);
864 NF_CT_STAT_INC(net
, insert_failed
);
868 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
870 static inline void nf_ct_acct_update(struct nf_conn
*ct
,
871 enum ip_conntrack_info ctinfo
,
874 struct nf_conn_acct
*acct
;
876 acct
= nf_conn_acct_find(ct
);
878 struct nf_conn_counter
*counter
= acct
->counter
;
880 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
881 atomic64_add(len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
885 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
886 const struct nf_conn
*loser_ct
)
888 struct nf_conn_acct
*acct
;
890 acct
= nf_conn_acct_find(loser_ct
);
892 struct nf_conn_counter
*counter
= acct
->counter
;
895 /* u32 should be fine since we must have seen one packet. */
896 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
897 nf_ct_acct_update(ct
, ctinfo
, bytes
);
901 /* Resolve race on insertion if this protocol allows this. */
902 static int nf_ct_resolve_clash(struct net
*net
, struct sk_buff
*skb
,
903 enum ip_conntrack_info ctinfo
,
904 struct nf_conntrack_tuple_hash
*h
)
906 /* This is the conntrack entry already in hashes that won race. */
907 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
908 const struct nf_conntrack_l4proto
*l4proto
;
909 enum ip_conntrack_info oldinfo
;
910 struct nf_conn
*loser_ct
= nf_ct_get(skb
, &oldinfo
);
912 l4proto
= nf_ct_l4proto_find(nf_ct_protonum(ct
));
913 if (l4proto
->allow_clash
&&
914 !nf_ct_is_dying(ct
) &&
915 atomic_inc_not_zero(&ct
->ct_general
.use
)) {
916 if (((ct
->status
& IPS_NAT_DONE_MASK
) == 0) ||
917 nf_ct_match(ct
, loser_ct
)) {
918 nf_ct_acct_merge(ct
, ctinfo
, loser_ct
);
919 nf_conntrack_put(&loser_ct
->ct_general
);
920 nf_ct_set(skb
, ct
, oldinfo
);
925 NF_CT_STAT_INC(net
, drop
);
929 /* Confirm a connection given skb; places it in hash table */
931 __nf_conntrack_confirm(struct sk_buff
*skb
)
933 const struct nf_conntrack_zone
*zone
;
934 unsigned int hash
, reply_hash
;
935 struct nf_conntrack_tuple_hash
*h
;
937 struct nf_conn_help
*help
;
938 struct nf_conn_tstamp
*tstamp
;
939 struct hlist_nulls_node
*n
;
940 enum ip_conntrack_info ctinfo
;
942 unsigned int sequence
;
945 ct
= nf_ct_get(skb
, &ctinfo
);
948 /* ipt_REJECT uses nf_conntrack_attach to attach related
949 ICMP/TCP RST packets in other direction. Actual packet
950 which created connection will be IP_CT_NEW or for an
951 expected connection, IP_CT_RELATED. */
952 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
955 zone
= nf_ct_zone(ct
);
959 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
960 /* reuse the hash saved before */
961 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
962 hash
= scale_hash(hash
);
963 reply_hash
= hash_conntrack(net
,
964 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
966 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
968 /* We're not in hash table, and we refuse to set up related
969 * connections for unconfirmed conns. But packet copies and
970 * REJECT will give spurious warnings here.
973 /* Another skb with the same unconfirmed conntrack may
974 * win the race. This may happen for bridge(br_flood)
975 * or broadcast/multicast packets do skb_clone with
976 * unconfirmed conntrack.
978 if (unlikely(nf_ct_is_confirmed(ct
))) {
980 nf_conntrack_double_unlock(hash
, reply_hash
);
985 pr_debug("Confirming conntrack %p\n", ct
);
986 /* We have to check the DYING flag after unlink to prevent
987 * a race against nf_ct_get_next_corpse() possibly called from
988 * user context, else we insert an already 'dead' hash, blocking
989 * further use of that particular connection -JM.
991 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
993 if (unlikely(nf_ct_is_dying(ct
))) {
994 nf_ct_add_to_dying_list(ct
);
998 /* See if there's one in the list already, including reverse:
999 NAT could have grabbed it without realizing, since we're
1000 not in the hash. If there is, we lost race. */
1001 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
1002 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
1006 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
1007 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
1011 /* Timer relative to confirmation time, not original
1012 setting time, otherwise we'd get timer wrap in
1013 weird delay cases. */
1014 ct
->timeout
+= nfct_time_stamp
;
1015 atomic_inc(&ct
->ct_general
.use
);
1016 ct
->status
|= IPS_CONFIRMED
;
1018 /* set conntrack timestamp, if enabled. */
1019 tstamp
= nf_conn_tstamp_find(ct
);
1021 tstamp
->start
= ktime_get_real_ns();
1023 /* Since the lookup is lockless, hash insertion must be done after
1024 * starting the timer and setting the CONFIRMED bit. The RCU barriers
1025 * guarantee that no other CPU can find the conntrack before the above
1026 * stores are visible.
1028 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
1029 nf_conntrack_double_unlock(hash
, reply_hash
);
1032 help
= nfct_help(ct
);
1033 if (help
&& help
->helper
)
1034 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
1036 nf_conntrack_event_cache(master_ct(ct
) ?
1037 IPCT_RELATED
: IPCT_NEW
, ct
);
1041 nf_ct_add_to_dying_list(ct
);
1042 ret
= nf_ct_resolve_clash(net
, skb
, ctinfo
, h
);
1044 nf_conntrack_double_unlock(hash
, reply_hash
);
1045 NF_CT_STAT_INC(net
, insert_failed
);
1049 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
1051 /* Returns true if a connection correspondings to the tuple (required
1054 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
1055 const struct nf_conn
*ignored_conntrack
)
1057 struct net
*net
= nf_ct_net(ignored_conntrack
);
1058 const struct nf_conntrack_zone
*zone
;
1059 struct nf_conntrack_tuple_hash
*h
;
1060 struct hlist_nulls_head
*ct_hash
;
1061 unsigned int hash
, hsize
;
1062 struct hlist_nulls_node
*n
;
1065 zone
= nf_ct_zone(ignored_conntrack
);
1069 nf_conntrack_get_ht(&ct_hash
, &hsize
);
1070 hash
= __hash_conntrack(net
, tuple
, hsize
);
1072 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
1073 ct
= nf_ct_tuplehash_to_ctrack(h
);
1075 if (ct
== ignored_conntrack
)
1078 if (nf_ct_is_expired(ct
)) {
1079 nf_ct_gc_expired(ct
);
1083 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
1084 /* Tuple is taken already, so caller will need to find
1085 * a new source port to use.
1088 * If the *original tuples* are identical, then both
1089 * conntracks refer to the same flow.
1090 * This is a rare situation, it can occur e.g. when
1091 * more than one UDP packet is sent from same socket
1092 * in different threads.
1094 * Let nf_ct_resolve_clash() deal with this later.
1096 if (nf_ct_tuple_equal(&ignored_conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
1097 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
))
1100 NF_CT_STAT_INC_ATOMIC(net
, found
);
1106 if (get_nulls_value(n
) != hash
) {
1107 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
1115 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
1117 #define NF_CT_EVICTION_RANGE 8
1119 /* There's a small race here where we may free a just-assured
1120 connection. Too bad: we're in trouble anyway. */
1121 static unsigned int early_drop_list(struct net
*net
,
1122 struct hlist_nulls_head
*head
)
1124 struct nf_conntrack_tuple_hash
*h
;
1125 struct hlist_nulls_node
*n
;
1126 unsigned int drops
= 0;
1127 struct nf_conn
*tmp
;
1129 hlist_nulls_for_each_entry_rcu(h
, n
, head
, hnnode
) {
1130 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1132 if (test_bit(IPS_OFFLOAD_BIT
, &tmp
->status
))
1135 if (nf_ct_is_expired(tmp
)) {
1136 nf_ct_gc_expired(tmp
);
1140 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
1141 !net_eq(nf_ct_net(tmp
), net
) ||
1142 nf_ct_is_dying(tmp
))
1145 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1148 /* kill only if still in same netns -- might have moved due to
1149 * SLAB_TYPESAFE_BY_RCU rules.
1151 * We steal the timer reference. If that fails timer has
1152 * already fired or someone else deleted it. Just drop ref
1153 * and move to next entry.
1155 if (net_eq(nf_ct_net(tmp
), net
) &&
1156 nf_ct_is_confirmed(tmp
) &&
1157 nf_ct_delete(tmp
, 0, 0))
1166 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
1168 unsigned int i
, bucket
;
1170 for (i
= 0; i
< NF_CT_EVICTION_RANGE
; i
++) {
1171 struct hlist_nulls_head
*ct_hash
;
1172 unsigned int hsize
, drops
;
1175 nf_conntrack_get_ht(&ct_hash
, &hsize
);
1177 bucket
= reciprocal_scale(hash
, hsize
);
1179 bucket
= (bucket
+ 1) % hsize
;
1181 drops
= early_drop_list(net
, &ct_hash
[bucket
]);
1185 NF_CT_STAT_ADD_ATOMIC(net
, early_drop
, drops
);
1193 static bool gc_worker_skip_ct(const struct nf_conn
*ct
)
1195 return !nf_ct_is_confirmed(ct
) || nf_ct_is_dying(ct
);
1198 static bool gc_worker_can_early_drop(const struct nf_conn
*ct
)
1200 const struct nf_conntrack_l4proto
*l4proto
;
1202 if (!test_bit(IPS_ASSURED_BIT
, &ct
->status
))
1205 l4proto
= nf_ct_l4proto_find(nf_ct_protonum(ct
));
1206 if (l4proto
->can_early_drop
&& l4proto
->can_early_drop(ct
))
1212 #define DAY (86400 * HZ)
1214 /* Set an arbitrary timeout large enough not to ever expire, this save
1215 * us a check for the IPS_OFFLOAD_BIT from the packet path via
1216 * nf_ct_is_expired().
1218 static void nf_ct_offload_timeout(struct nf_conn
*ct
)
1220 if (nf_ct_expires(ct
) < DAY
/ 2)
1221 ct
->timeout
= nfct_time_stamp
+ DAY
;
1224 static void gc_worker(struct work_struct
*work
)
1226 unsigned int min_interval
= max(HZ
/ GC_MAX_BUCKETS_DIV
, 1u);
1227 unsigned int i
, goal
, buckets
= 0, expired_count
= 0;
1228 unsigned int nf_conntrack_max95
= 0;
1229 struct conntrack_gc_work
*gc_work
;
1230 unsigned int ratio
, scanned
= 0;
1231 unsigned long next_run
;
1233 gc_work
= container_of(work
, struct conntrack_gc_work
, dwork
.work
);
1235 goal
= nf_conntrack_htable_size
/ GC_MAX_BUCKETS_DIV
;
1236 i
= gc_work
->last_bucket
;
1237 if (gc_work
->early_drop
)
1238 nf_conntrack_max95
= nf_conntrack_max
/ 100u * 95u;
1241 struct nf_conntrack_tuple_hash
*h
;
1242 struct hlist_nulls_head
*ct_hash
;
1243 struct hlist_nulls_node
*n
;
1244 unsigned int hashsz
;
1245 struct nf_conn
*tmp
;
1250 nf_conntrack_get_ht(&ct_hash
, &hashsz
);
1254 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[i
], hnnode
) {
1257 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1260 if (test_bit(IPS_OFFLOAD_BIT
, &tmp
->status
)) {
1261 nf_ct_offload_timeout(tmp
);
1265 if (nf_ct_is_expired(tmp
)) {
1266 nf_ct_gc_expired(tmp
);
1271 if (nf_conntrack_max95
== 0 || gc_worker_skip_ct(tmp
))
1274 net
= nf_ct_net(tmp
);
1275 if (atomic_read(&net
->ct
.count
) < nf_conntrack_max95
)
1278 /* need to take reference to avoid possible races */
1279 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1282 if (gc_worker_skip_ct(tmp
)) {
1287 if (gc_worker_can_early_drop(tmp
))
1293 /* could check get_nulls_value() here and restart if ct
1294 * was moved to another chain. But given gc is best-effort
1295 * we will just continue with next hash slot.
1299 } while (++buckets
< goal
);
1301 if (gc_work
->exiting
)
1305 * Eviction will normally happen from the packet path, and not
1306 * from this gc worker.
1308 * This worker is only here to reap expired entries when system went
1309 * idle after a busy period.
1311 * The heuristics below are supposed to balance conflicting goals:
1313 * 1. Minimize time until we notice a stale entry
1314 * 2. Maximize scan intervals to not waste cycles
1316 * Normally, expire ratio will be close to 0.
1318 * As soon as a sizeable fraction of the entries have expired
1319 * increase scan frequency.
1321 ratio
= scanned
? expired_count
* 100 / scanned
: 0;
1322 if (ratio
> GC_EVICT_RATIO
) {
1323 gc_work
->next_gc_run
= min_interval
;
1325 unsigned int max
= GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
;
1327 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
) == 0);
1329 gc_work
->next_gc_run
+= min_interval
;
1330 if (gc_work
->next_gc_run
> max
)
1331 gc_work
->next_gc_run
= max
;
1334 next_run
= gc_work
->next_gc_run
;
1335 gc_work
->last_bucket
= i
;
1336 gc_work
->early_drop
= false;
1337 queue_delayed_work(system_power_efficient_wq
, &gc_work
->dwork
, next_run
);
1340 static void conntrack_gc_work_init(struct conntrack_gc_work
*gc_work
)
1342 INIT_DEFERRABLE_WORK(&gc_work
->dwork
, gc_worker
);
1343 gc_work
->next_gc_run
= HZ
;
1344 gc_work
->exiting
= false;
1347 static struct nf_conn
*
1348 __nf_conntrack_alloc(struct net
*net
,
1349 const struct nf_conntrack_zone
*zone
,
1350 const struct nf_conntrack_tuple
*orig
,
1351 const struct nf_conntrack_tuple
*repl
,
1352 gfp_t gfp
, u32 hash
)
1356 /* We don't want any race condition at early drop stage */
1357 atomic_inc(&net
->ct
.count
);
1359 if (nf_conntrack_max
&&
1360 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
1361 if (!early_drop(net
, hash
)) {
1362 if (!conntrack_gc_work
.early_drop
)
1363 conntrack_gc_work
.early_drop
= true;
1364 atomic_dec(&net
->ct
.count
);
1365 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1366 return ERR_PTR(-ENOMEM
);
1371 * Do not use kmem_cache_zalloc(), as this cache uses
1372 * SLAB_TYPESAFE_BY_RCU.
1374 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
1378 spin_lock_init(&ct
->lock
);
1379 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
1380 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
1381 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
1382 /* save hash for reusing when confirming */
1383 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
1386 write_pnet(&ct
->ct_net
, net
);
1387 memset(&ct
->__nfct_init_offset
[0], 0,
1388 offsetof(struct nf_conn
, proto
) -
1389 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
1391 nf_ct_zone_add(ct
, zone
);
1393 /* Because we use RCU lookups, we set ct_general.use to zero before
1394 * this is inserted in any list.
1396 atomic_set(&ct
->ct_general
.use
, 0);
1399 atomic_dec(&net
->ct
.count
);
1400 return ERR_PTR(-ENOMEM
);
1403 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
1404 const struct nf_conntrack_zone
*zone
,
1405 const struct nf_conntrack_tuple
*orig
,
1406 const struct nf_conntrack_tuple
*repl
,
1409 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
1411 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
1413 void nf_conntrack_free(struct nf_conn
*ct
)
1415 struct net
*net
= nf_ct_net(ct
);
1417 /* A freed object has refcnt == 0, that's
1418 * the golden rule for SLAB_TYPESAFE_BY_RCU
1420 WARN_ON(atomic_read(&ct
->ct_general
.use
) != 0);
1422 nf_ct_ext_destroy(ct
);
1424 kmem_cache_free(nf_conntrack_cachep
, ct
);
1425 smp_mb__before_atomic();
1426 atomic_dec(&net
->ct
.count
);
1428 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
1431 /* Allocate a new conntrack: we return -ENOMEM if classification
1432 failed due to stress. Otherwise it really is unclassifiable. */
1433 static noinline
struct nf_conntrack_tuple_hash
*
1434 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
1435 const struct nf_conntrack_tuple
*tuple
,
1436 struct sk_buff
*skb
,
1437 unsigned int dataoff
, u32 hash
)
1440 struct nf_conn_help
*help
;
1441 struct nf_conntrack_tuple repl_tuple
;
1442 struct nf_conntrack_ecache
*ecache
;
1443 struct nf_conntrack_expect
*exp
= NULL
;
1444 const struct nf_conntrack_zone
*zone
;
1445 struct nf_conn_timeout
*timeout_ext
;
1446 struct nf_conntrack_zone tmp
;
1448 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
)) {
1449 pr_debug("Can't invert tuple.\n");
1453 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1454 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
1457 return (struct nf_conntrack_tuple_hash
*)ct
;
1459 if (!nf_ct_add_synproxy(ct
, tmpl
)) {
1460 nf_conntrack_free(ct
);
1461 return ERR_PTR(-ENOMEM
);
1464 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1467 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1470 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1471 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1472 nf_ct_labels_ext_add(ct
);
1474 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1475 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1476 ecache
? ecache
->expmask
: 0,
1480 if (net
->ct
.expect_count
) {
1481 spin_lock(&nf_conntrack_expect_lock
);
1482 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1484 pr_debug("expectation arrives ct=%p exp=%p\n",
1486 /* Welcome, Mr. Bond. We've been expecting you... */
1487 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1488 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1489 ct
->master
= exp
->master
;
1491 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
1493 rcu_assign_pointer(help
->helper
, exp
->helper
);
1496 #ifdef CONFIG_NF_CONNTRACK_MARK
1497 ct
->mark
= exp
->master
->mark
;
1499 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1500 ct
->secmark
= exp
->master
->secmark
;
1502 NF_CT_STAT_INC(net
, expect_new
);
1504 spin_unlock(&nf_conntrack_expect_lock
);
1507 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1509 /* Now it is inserted into the unconfirmed list, bump refcount */
1510 nf_conntrack_get(&ct
->ct_general
);
1511 nf_ct_add_to_unconfirmed_list(ct
);
1517 exp
->expectfn(ct
, exp
);
1518 nf_ct_expect_put(exp
);
1521 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1524 /* On success, returns 0, sets skb->_nfct | ctinfo */
1526 resolve_normal_ct(struct nf_conn
*tmpl
,
1527 struct sk_buff
*skb
,
1528 unsigned int dataoff
,
1530 const struct nf_hook_state
*state
)
1532 const struct nf_conntrack_zone
*zone
;
1533 struct nf_conntrack_tuple tuple
;
1534 struct nf_conntrack_tuple_hash
*h
;
1535 enum ip_conntrack_info ctinfo
;
1536 struct nf_conntrack_zone tmp
;
1540 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1541 dataoff
, state
->pf
, protonum
, state
->net
,
1543 pr_debug("Can't get tuple\n");
1547 /* look for tuple match */
1548 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1549 hash
= hash_conntrack_raw(&tuple
, state
->net
);
1550 h
= __nf_conntrack_find_get(state
->net
, zone
, &tuple
, hash
);
1552 h
= init_conntrack(state
->net
, tmpl
, &tuple
,
1553 skb
, dataoff
, hash
);
1559 ct
= nf_ct_tuplehash_to_ctrack(h
);
1561 /* It exists; we have (non-exclusive) reference. */
1562 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1563 ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1565 /* Once we've had two way comms, always ESTABLISHED. */
1566 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1567 pr_debug("normal packet for %p\n", ct
);
1568 ctinfo
= IP_CT_ESTABLISHED
;
1569 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1570 pr_debug("related packet for %p\n", ct
);
1571 ctinfo
= IP_CT_RELATED
;
1573 pr_debug("new packet for %p\n", ct
);
1577 nf_ct_set(skb
, ct
, ctinfo
);
1582 * icmp packets need special treatment to handle error messages that are
1583 * related to a connection.
1585 * Callers need to check if skb has a conntrack assigned when this
1586 * helper returns; in such case skb belongs to an already known connection.
1588 static unsigned int __cold
1589 nf_conntrack_handle_icmp(struct nf_conn
*tmpl
,
1590 struct sk_buff
*skb
,
1591 unsigned int dataoff
,
1593 const struct nf_hook_state
*state
)
1597 if (state
->pf
== NFPROTO_IPV4
&& protonum
== IPPROTO_ICMP
)
1598 ret
= nf_conntrack_icmpv4_error(tmpl
, skb
, dataoff
, state
);
1599 #if IS_ENABLED(CONFIG_IPV6)
1600 else if (state
->pf
== NFPROTO_IPV6
&& protonum
== IPPROTO_ICMPV6
)
1601 ret
= nf_conntrack_icmpv6_error(tmpl
, skb
, dataoff
, state
);
1607 NF_CT_STAT_INC_ATOMIC(state
->net
, error
);
1608 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1614 static int generic_packet(struct nf_conn
*ct
, struct sk_buff
*skb
,
1615 enum ip_conntrack_info ctinfo
)
1617 const unsigned int *timeout
= nf_ct_timeout_lookup(ct
);
1620 timeout
= &nf_generic_pernet(nf_ct_net(ct
))->timeout
;
1622 nf_ct_refresh_acct(ct
, ctinfo
, skb
, *timeout
);
1626 /* Returns verdict for packet, or -1 for invalid. */
1627 static int nf_conntrack_handle_packet(struct nf_conn
*ct
,
1628 struct sk_buff
*skb
,
1629 unsigned int dataoff
,
1630 enum ip_conntrack_info ctinfo
,
1631 const struct nf_hook_state
*state
)
1633 switch (nf_ct_protonum(ct
)) {
1635 return nf_conntrack_tcp_packet(ct
, skb
, dataoff
,
1638 return nf_conntrack_udp_packet(ct
, skb
, dataoff
,
1641 return nf_conntrack_icmp_packet(ct
, skb
, ctinfo
, state
);
1642 #if IS_ENABLED(CONFIG_IPV6)
1643 case IPPROTO_ICMPV6
:
1644 return nf_conntrack_icmpv6_packet(ct
, skb
, ctinfo
, state
);
1646 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
1647 case IPPROTO_UDPLITE
:
1648 return nf_conntrack_udplite_packet(ct
, skb
, dataoff
,
1651 #ifdef CONFIG_NF_CT_PROTO_SCTP
1653 return nf_conntrack_sctp_packet(ct
, skb
, dataoff
,
1656 #ifdef CONFIG_NF_CT_PROTO_DCCP
1658 return nf_conntrack_dccp_packet(ct
, skb
, dataoff
,
1661 #ifdef CONFIG_NF_CT_PROTO_GRE
1663 return nf_conntrack_gre_packet(ct
, skb
, dataoff
,
1668 return generic_packet(ct
, skb
, ctinfo
);
1672 nf_conntrack_in(struct sk_buff
*skb
, const struct nf_hook_state
*state
)
1674 enum ip_conntrack_info ctinfo
;
1675 struct nf_conn
*ct
, *tmpl
;
1679 tmpl
= nf_ct_get(skb
, &ctinfo
);
1680 if (tmpl
|| ctinfo
== IP_CT_UNTRACKED
) {
1681 /* Previously seen (loopback or untracked)? Ignore. */
1682 if ((tmpl
&& !nf_ct_is_template(tmpl
)) ||
1683 ctinfo
== IP_CT_UNTRACKED
) {
1684 NF_CT_STAT_INC_ATOMIC(state
->net
, ignore
);
1690 /* rcu_read_lock()ed by nf_hook_thresh */
1691 dataoff
= get_l4proto(skb
, skb_network_offset(skb
), state
->pf
, &protonum
);
1693 pr_debug("not prepared to track yet or error occurred\n");
1694 NF_CT_STAT_INC_ATOMIC(state
->net
, error
);
1695 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1700 if (protonum
== IPPROTO_ICMP
|| protonum
== IPPROTO_ICMPV6
) {
1701 ret
= nf_conntrack_handle_icmp(tmpl
, skb
, dataoff
,
1707 /* ICMP[v6] protocol trackers may assign one conntrack. */
1712 ret
= resolve_normal_ct(tmpl
, skb
, dataoff
,
1715 /* Too stressed to deal. */
1716 NF_CT_STAT_INC_ATOMIC(state
->net
, drop
);
1721 ct
= nf_ct_get(skb
, &ctinfo
);
1723 /* Not valid part of a connection */
1724 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1729 ret
= nf_conntrack_handle_packet(ct
, skb
, dataoff
, ctinfo
, state
);
1731 /* Invalid: inverse of the return code tells
1732 * the netfilter core what to do */
1733 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1734 nf_conntrack_put(&ct
->ct_general
);
1736 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1737 if (ret
== -NF_DROP
)
1738 NF_CT_STAT_INC_ATOMIC(state
->net
, drop
);
1739 /* Special case: TCP tracker reports an attempt to reopen a
1740 * closed/aborted connection. We have to go back and create a
1743 if (ret
== -NF_REPEAT
)
1749 if (ctinfo
== IP_CT_ESTABLISHED_REPLY
&&
1750 !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1751 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1758 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1760 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1761 implicitly racy: see __nf_conntrack_confirm */
1762 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1763 const struct nf_conntrack_tuple
*newreply
)
1765 struct nf_conn_help
*help
= nfct_help(ct
);
1767 /* Should be unconfirmed, so not in hash table yet */
1768 WARN_ON(nf_ct_is_confirmed(ct
));
1770 pr_debug("Altering reply tuple of %p to ", ct
);
1771 nf_ct_dump_tuple(newreply
);
1773 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1774 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1778 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1781 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1783 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1784 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1785 enum ip_conntrack_info ctinfo
,
1786 const struct sk_buff
*skb
,
1790 /* Only update if this is not a fixed timeout */
1791 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1794 /* If not in hash table, timer will not be active yet */
1795 if (nf_ct_is_confirmed(ct
))
1796 extra_jiffies
+= nfct_time_stamp
;
1798 if (ct
->timeout
!= extra_jiffies
)
1799 ct
->timeout
= extra_jiffies
;
1802 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1804 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1806 bool nf_ct_kill_acct(struct nf_conn
*ct
,
1807 enum ip_conntrack_info ctinfo
,
1808 const struct sk_buff
*skb
)
1810 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1812 return nf_ct_delete(ct
, 0, 0);
1814 EXPORT_SYMBOL_GPL(nf_ct_kill_acct
);
1816 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1818 #include <linux/netfilter/nfnetlink.h>
1819 #include <linux/netfilter/nfnetlink_conntrack.h>
1820 #include <linux/mutex.h>
1822 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1823 * in ip_conntrack_core, since we don't want the protocols to autoload
1824 * or depend on ctnetlink */
1825 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1826 const struct nf_conntrack_tuple
*tuple
)
1828 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1829 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1830 goto nla_put_failure
;
1836 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1838 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1839 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1840 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1842 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1844 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1845 struct nf_conntrack_tuple
*t
)
1847 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1850 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1851 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1855 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1857 unsigned int nf_ct_port_nlattr_tuple_size(void)
1859 static unsigned int size __read_mostly
;
1862 size
= nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1866 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1869 /* Used by ipt_REJECT and ip6t_REJECT. */
1870 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1873 enum ip_conntrack_info ctinfo
;
1875 /* This ICMP is in reverse direction to the packet which caused it */
1876 ct
= nf_ct_get(skb
, &ctinfo
);
1877 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1878 ctinfo
= IP_CT_RELATED_REPLY
;
1880 ctinfo
= IP_CT_RELATED
;
1882 /* Attach to new skbuff, and increment count */
1883 nf_ct_set(nskb
, ct
, ctinfo
);
1884 nf_conntrack_get(skb_nfct(nskb
));
1887 static int nf_conntrack_update(struct net
*net
, struct sk_buff
*skb
)
1889 struct nf_conntrack_tuple_hash
*h
;
1890 struct nf_conntrack_tuple tuple
;
1891 enum ip_conntrack_info ctinfo
;
1892 struct nf_nat_hook
*nat_hook
;
1893 unsigned int status
;
1899 ct
= nf_ct_get(skb
, &ctinfo
);
1900 if (!ct
|| nf_ct_is_confirmed(ct
))
1903 l3num
= nf_ct_l3num(ct
);
1905 dataoff
= get_l4proto(skb
, skb_network_offset(skb
), l3num
, &l4num
);
1909 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
), dataoff
, l3num
,
1910 l4num
, net
, &tuple
))
1913 if (ct
->status
& IPS_SRC_NAT
) {
1914 memcpy(tuple
.src
.u3
.all
,
1915 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.u3
.all
,
1916 sizeof(tuple
.src
.u3
.all
));
1918 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.u
.all
;
1921 if (ct
->status
& IPS_DST_NAT
) {
1922 memcpy(tuple
.dst
.u3
.all
,
1923 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.u3
.all
,
1924 sizeof(tuple
.dst
.u3
.all
));
1926 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.u
.all
;
1929 h
= nf_conntrack_find_get(net
, nf_ct_zone(ct
), &tuple
);
1933 /* Store status bits of the conntrack that is clashing to re-do NAT
1934 * mangling according to what it has been done already to this packet.
1936 status
= ct
->status
;
1939 ct
= nf_ct_tuplehash_to_ctrack(h
);
1940 nf_ct_set(skb
, ct
, ctinfo
);
1942 nat_hook
= rcu_dereference(nf_nat_hook
);
1946 if (status
& IPS_SRC_NAT
&&
1947 nat_hook
->manip_pkt(skb
, ct
, NF_NAT_MANIP_SRC
,
1948 IP_CT_DIR_ORIGINAL
) == NF_DROP
)
1951 if (status
& IPS_DST_NAT
&&
1952 nat_hook
->manip_pkt(skb
, ct
, NF_NAT_MANIP_DST
,
1953 IP_CT_DIR_ORIGINAL
) == NF_DROP
)
1959 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple
*dst_tuple
,
1960 const struct sk_buff
*skb
)
1962 const struct nf_conntrack_tuple
*src_tuple
;
1963 const struct nf_conntrack_tuple_hash
*hash
;
1964 struct nf_conntrack_tuple srctuple
;
1965 enum ip_conntrack_info ctinfo
;
1968 ct
= nf_ct_get(skb
, &ctinfo
);
1970 src_tuple
= nf_ct_tuple(ct
, CTINFO2DIR(ctinfo
));
1971 memcpy(dst_tuple
, src_tuple
, sizeof(*dst_tuple
));
1975 if (!nf_ct_get_tuplepr(skb
, skb_network_offset(skb
),
1976 NFPROTO_IPV4
, dev_net(skb
->dev
),
1980 hash
= nf_conntrack_find_get(dev_net(skb
->dev
),
1986 ct
= nf_ct_tuplehash_to_ctrack(hash
);
1987 src_tuple
= nf_ct_tuple(ct
, !hash
->tuple
.dst
.dir
);
1988 memcpy(dst_tuple
, src_tuple
, sizeof(*dst_tuple
));
1994 /* Bring out ya dead! */
1995 static struct nf_conn
*
1996 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
1997 void *data
, unsigned int *bucket
)
1999 struct nf_conntrack_tuple_hash
*h
;
2001 struct hlist_nulls_node
*n
;
2004 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
2005 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
2007 nf_conntrack_lock(lockp
);
2008 if (*bucket
< nf_conntrack_htable_size
) {
2009 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
2010 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
2012 ct
= nf_ct_tuplehash_to_ctrack(h
);
2024 atomic_inc(&ct
->ct_general
.use
);
2030 static void nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
),
2031 void *data
, u32 portid
, int report
)
2033 unsigned int bucket
= 0, sequence
;
2039 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
2041 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
2042 /* Time to push up daises... */
2044 nf_ct_delete(ct
, portid
, report
);
2049 if (!read_seqcount_retry(&nf_conntrack_generation
, sequence
))
2056 int (*iter
)(struct nf_conn
*i
, void *data
);
2061 static int iter_net_only(struct nf_conn
*i
, void *data
)
2063 struct iter_data
*d
= data
;
2065 if (!net_eq(d
->net
, nf_ct_net(i
)))
2068 return d
->iter(i
, d
->data
);
2072 __nf_ct_unconfirmed_destroy(struct net
*net
)
2076 for_each_possible_cpu(cpu
) {
2077 struct nf_conntrack_tuple_hash
*h
;
2078 struct hlist_nulls_node
*n
;
2079 struct ct_pcpu
*pcpu
;
2081 pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2083 spin_lock_bh(&pcpu
->lock
);
2084 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
2087 ct
= nf_ct_tuplehash_to_ctrack(h
);
2089 /* we cannot call iter() on unconfirmed list, the
2090 * owning cpu can reallocate ct->ext at any time.
2092 set_bit(IPS_DYING_BIT
, &ct
->status
);
2094 spin_unlock_bh(&pcpu
->lock
);
2099 void nf_ct_unconfirmed_destroy(struct net
*net
)
2103 if (atomic_read(&net
->ct
.count
) > 0) {
2104 __nf_ct_unconfirmed_destroy(net
);
2105 nf_queue_nf_hook_drop(net
);
2109 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy
);
2111 void nf_ct_iterate_cleanup_net(struct net
*net
,
2112 int (*iter
)(struct nf_conn
*i
, void *data
),
2113 void *data
, u32 portid
, int report
)
2119 if (atomic_read(&net
->ct
.count
) == 0)
2126 nf_ct_iterate_cleanup(iter_net_only
, &d
, portid
, report
);
2128 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net
);
2131 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2132 * @iter: callback to invoke for each conntrack
2133 * @data: data to pass to @iter
2135 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2136 * unconfirmed list as dying (so they will not be inserted into
2139 * Can only be called in module exit path.
2142 nf_ct_iterate_destroy(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
2146 down_read(&net_rwsem
);
2148 if (atomic_read(&net
->ct
.count
) == 0)
2150 __nf_ct_unconfirmed_destroy(net
);
2151 nf_queue_nf_hook_drop(net
);
2153 up_read(&net_rwsem
);
2155 /* Need to wait for netns cleanup worker to finish, if its
2156 * running -- it might have deleted a net namespace from
2157 * the global list, so our __nf_ct_unconfirmed_destroy() might
2158 * not have affected all namespaces.
2162 /* a conntrack could have been unlinked from unconfirmed list
2163 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2164 * This makes sure its inserted into conntrack table.
2168 nf_ct_iterate_cleanup(iter
, data
, 0, 0);
2170 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy
);
2172 static int kill_all(struct nf_conn
*i
, void *data
)
2174 return net_eq(nf_ct_net(i
), data
);
2177 void nf_conntrack_cleanup_start(void)
2179 conntrack_gc_work
.exiting
= true;
2180 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
2183 void nf_conntrack_cleanup_end(void)
2185 RCU_INIT_POINTER(nf_ct_hook
, NULL
);
2186 cancel_delayed_work_sync(&conntrack_gc_work
.dwork
);
2187 kvfree(nf_conntrack_hash
);
2189 nf_conntrack_proto_fini();
2190 nf_conntrack_seqadj_fini();
2191 nf_conntrack_labels_fini();
2192 nf_conntrack_helper_fini();
2193 nf_conntrack_timeout_fini();
2194 nf_conntrack_ecache_fini();
2195 nf_conntrack_tstamp_fini();
2196 nf_conntrack_acct_fini();
2197 nf_conntrack_expect_fini();
2199 kmem_cache_destroy(nf_conntrack_cachep
);
2203 * Mishearing the voices in his head, our hero wonders how he's
2204 * supposed to kill the mall.
2206 void nf_conntrack_cleanup_net(struct net
*net
)
2210 list_add(&net
->exit_list
, &single
);
2211 nf_conntrack_cleanup_net_list(&single
);
2214 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
2220 * This makes sure all current packets have passed through
2221 * netfilter framework. Roll on, two-stage module
2227 list_for_each_entry(net
, net_exit_list
, exit_list
) {
2228 nf_ct_iterate_cleanup(kill_all
, net
, 0, 0);
2229 if (atomic_read(&net
->ct
.count
) != 0)
2234 goto i_see_dead_people
;
2237 list_for_each_entry(net
, net_exit_list
, exit_list
) {
2238 nf_conntrack_proto_pernet_fini(net
);
2239 nf_conntrack_ecache_pernet_fini(net
);
2240 nf_conntrack_expect_pernet_fini(net
);
2241 free_percpu(net
->ct
.stat
);
2242 free_percpu(net
->ct
.pcpu_lists
);
2246 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
2248 struct hlist_nulls_head
*hash
;
2249 unsigned int nr_slots
, i
;
2251 if (*sizep
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
2254 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
2255 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
2257 hash
= kvmalloc_array(nr_slots
, sizeof(struct hlist_nulls_head
),
2258 GFP_KERNEL
| __GFP_ZERO
);
2261 for (i
= 0; i
< nr_slots
; i
++)
2262 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
2266 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
2268 int nf_conntrack_hash_resize(unsigned int hashsize
)
2271 unsigned int old_size
;
2272 struct hlist_nulls_head
*hash
, *old_hash
;
2273 struct nf_conntrack_tuple_hash
*h
;
2279 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
2283 old_size
= nf_conntrack_htable_size
;
2284 if (old_size
== hashsize
) {
2290 nf_conntrack_all_lock();
2291 write_seqcount_begin(&nf_conntrack_generation
);
2293 /* Lookups in the old hash might happen in parallel, which means we
2294 * might get false negatives during connection lookup. New connections
2295 * created because of a false negative won't make it into the hash
2296 * though since that required taking the locks.
2299 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
2300 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
2301 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
2302 struct nf_conntrack_tuple_hash
, hnnode
);
2303 ct
= nf_ct_tuplehash_to_ctrack(h
);
2304 hlist_nulls_del_rcu(&h
->hnnode
);
2305 bucket
= __hash_conntrack(nf_ct_net(ct
),
2306 &h
->tuple
, hashsize
);
2307 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
2310 old_size
= nf_conntrack_htable_size
;
2311 old_hash
= nf_conntrack_hash
;
2313 nf_conntrack_hash
= hash
;
2314 nf_conntrack_htable_size
= hashsize
;
2316 write_seqcount_end(&nf_conntrack_generation
);
2317 nf_conntrack_all_unlock();
2325 int nf_conntrack_set_hashsize(const char *val
, const struct kernel_param
*kp
)
2327 unsigned int hashsize
;
2330 if (current
->nsproxy
->net_ns
!= &init_net
)
2333 /* On boot, we can set this without any fancy locking. */
2334 if (!nf_conntrack_hash
)
2335 return param_set_uint(val
, kp
);
2337 rc
= kstrtouint(val
, 0, &hashsize
);
2341 return nf_conntrack_hash_resize(hashsize
);
2343 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
2345 static __always_inline
unsigned int total_extension_size(void)
2347 /* remember to add new extensions below */
2348 BUILD_BUG_ON(NF_CT_EXT_NUM
> 9);
2350 return sizeof(struct nf_ct_ext
) +
2351 sizeof(struct nf_conn_help
)
2352 #if IS_ENABLED(CONFIG_NF_NAT)
2353 + sizeof(struct nf_conn_nat
)
2355 + sizeof(struct nf_conn_seqadj
)
2356 + sizeof(struct nf_conn_acct
)
2357 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2358 + sizeof(struct nf_conntrack_ecache
)
2360 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2361 + sizeof(struct nf_conn_tstamp
)
2363 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2364 + sizeof(struct nf_conn_timeout
)
2366 #ifdef CONFIG_NF_CONNTRACK_LABELS
2367 + sizeof(struct nf_conn_labels
)
2369 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2370 + sizeof(struct nf_conn_synproxy
)
2375 int nf_conntrack_init_start(void)
2377 unsigned long nr_pages
= totalram_pages();
2382 /* struct nf_ct_ext uses u8 to store offsets/size */
2383 BUILD_BUG_ON(total_extension_size() > 255u);
2385 seqcount_init(&nf_conntrack_generation
);
2387 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
2388 spin_lock_init(&nf_conntrack_locks
[i
]);
2390 if (!nf_conntrack_htable_size
) {
2391 /* Idea from tcp.c: use 1/16384 of memory.
2392 * On i386: 32MB machine has 512 buckets.
2393 * >= 1GB machines have 16384 buckets.
2394 * >= 4GB machines have 65536 buckets.
2396 nf_conntrack_htable_size
2397 = (((nr_pages
<< PAGE_SHIFT
) / 16384)
2398 / sizeof(struct hlist_head
));
2399 if (nr_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
2400 nf_conntrack_htable_size
= 65536;
2401 else if (nr_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
2402 nf_conntrack_htable_size
= 16384;
2403 if (nf_conntrack_htable_size
< 32)
2404 nf_conntrack_htable_size
= 32;
2406 /* Use a max. factor of four by default to get the same max as
2407 * with the old struct list_heads. When a table size is given
2408 * we use the old value of 8 to avoid reducing the max.
2413 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
2414 if (!nf_conntrack_hash
)
2417 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
2419 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
2420 sizeof(struct nf_conn
),
2422 SLAB_TYPESAFE_BY_RCU
| SLAB_HWCACHE_ALIGN
, NULL
);
2423 if (!nf_conntrack_cachep
)
2426 ret
= nf_conntrack_expect_init();
2430 ret
= nf_conntrack_acct_init();
2434 ret
= nf_conntrack_tstamp_init();
2438 ret
= nf_conntrack_ecache_init();
2442 ret
= nf_conntrack_timeout_init();
2446 ret
= nf_conntrack_helper_init();
2450 ret
= nf_conntrack_labels_init();
2454 ret
= nf_conntrack_seqadj_init();
2458 ret
= nf_conntrack_proto_init();
2462 conntrack_gc_work_init(&conntrack_gc_work
);
2463 queue_delayed_work(system_power_efficient_wq
, &conntrack_gc_work
.dwork
, HZ
);
2468 nf_conntrack_seqadj_fini();
2470 nf_conntrack_labels_fini();
2472 nf_conntrack_helper_fini();
2474 nf_conntrack_timeout_fini();
2476 nf_conntrack_ecache_fini();
2478 nf_conntrack_tstamp_fini();
2480 nf_conntrack_acct_fini();
2482 nf_conntrack_expect_fini();
2484 kmem_cache_destroy(nf_conntrack_cachep
);
2486 kvfree(nf_conntrack_hash
);
2490 static struct nf_ct_hook nf_conntrack_hook
= {
2491 .update
= nf_conntrack_update
,
2492 .destroy
= destroy_conntrack
,
2493 .get_tuple_skb
= nf_conntrack_get_tuple_skb
,
2496 void nf_conntrack_init_end(void)
2498 /* For use by REJECT target */
2499 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
2500 RCU_INIT_POINTER(nf_ct_hook
, &nf_conntrack_hook
);
2504 * We need to use special "null" values, not used in hash table
2506 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2507 #define DYING_NULLS_VAL ((1<<30)+1)
2508 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
2510 int nf_conntrack_init_net(struct net
*net
)
2515 BUILD_BUG_ON(IP_CT_UNTRACKED
== IP_CT_NUMBER
);
2516 BUILD_BUG_ON_NOT_POWER_OF_2(CONNTRACK_LOCKS
);
2517 atomic_set(&net
->ct
.count
, 0);
2519 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
2520 if (!net
->ct
.pcpu_lists
)
2523 for_each_possible_cpu(cpu
) {
2524 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2526 spin_lock_init(&pcpu
->lock
);
2527 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
2528 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
2531 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
2533 goto err_pcpu_lists
;
2535 ret
= nf_conntrack_expect_pernet_init(net
);
2539 nf_conntrack_acct_pernet_init(net
);
2540 nf_conntrack_tstamp_pernet_init(net
);
2541 nf_conntrack_ecache_pernet_init(net
);
2542 nf_conntrack_helper_pernet_init(net
);
2543 nf_conntrack_proto_pernet_init(net
);
2548 free_percpu(net
->ct
.stat
);
2550 free_percpu(net
->ct
.pcpu_lists
);