1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Connection state tracking for netfilter. This is separated from,
3 but required by, the NAT layer; it can also be used by an iptables
6 /* (C) 1999-2001 Paul `Rusty' Russell
7 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
8 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
9 * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/module.h>
17 #include <linux/sched.h>
18 #include <linux/skbuff.h>
19 #include <linux/proc_fs.h>
20 #include <linux/vmalloc.h>
21 #include <linux/stddef.h>
22 #include <linux/slab.h>
23 #include <linux/random.h>
24 #include <linux/jhash.h>
25 #include <linux/siphash.h>
26 #include <linux/err.h>
27 #include <linux/percpu.h>
28 #include <linux/moduleparam.h>
29 #include <linux/notifier.h>
30 #include <linux/kernel.h>
31 #include <linux/netdevice.h>
32 #include <linux/socket.h>
34 #include <linux/nsproxy.h>
35 #include <linux/rculist_nulls.h>
37 #include <net/netfilter/nf_conntrack.h>
38 #include <net/netfilter/nf_conntrack_l4proto.h>
39 #include <net/netfilter/nf_conntrack_expect.h>
40 #include <net/netfilter/nf_conntrack_helper.h>
41 #include <net/netfilter/nf_conntrack_seqadj.h>
42 #include <net/netfilter/nf_conntrack_core.h>
43 #include <net/netfilter/nf_conntrack_extend.h>
44 #include <net/netfilter/nf_conntrack_acct.h>
45 #include <net/netfilter/nf_conntrack_ecache.h>
46 #include <net/netfilter/nf_conntrack_zones.h>
47 #include <net/netfilter/nf_conntrack_timestamp.h>
48 #include <net/netfilter/nf_conntrack_timeout.h>
49 #include <net/netfilter/nf_conntrack_labels.h>
50 #include <net/netfilter/nf_conntrack_synproxy.h>
51 #include <net/netfilter/nf_nat.h>
52 #include <net/netfilter/nf_nat_helper.h>
53 #include <net/netns/hash.h>
56 #include "nf_internals.h"
58 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks
[CONNTRACK_LOCKS
];
59 EXPORT_SYMBOL_GPL(nf_conntrack_locks
);
61 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(nf_conntrack_expect_lock
);
62 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock
);
64 struct hlist_nulls_head
*nf_conntrack_hash __read_mostly
;
65 EXPORT_SYMBOL_GPL(nf_conntrack_hash
);
67 struct conntrack_gc_work
{
68 struct delayed_work dwork
;
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
DEFINE_SPINLOCK(nf_conntrack_locks_all_lock
);
78 static __read_mostly
bool nf_conntrack_locks_all
;
80 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
81 #define GC_MAX_BUCKETS_DIV 128u
82 /* upper bound of full table scan */
83 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
84 /* desired ratio of entries found to be expired */
85 #define GC_EVICT_RATIO 50u
87 static struct conntrack_gc_work conntrack_gc_work
;
89 void nf_conntrack_lock(spinlock_t
*lock
) __acquires(lock
)
91 /* 1) Acquire the lock */
94 /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
95 * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
97 if (likely(smp_load_acquire(&nf_conntrack_locks_all
) == false))
100 /* fast path failed, unlock */
103 /* Slow path 1) get global lock */
104 spin_lock(&nf_conntrack_locks_all_lock
);
106 /* Slow path 2) get the lock we want */
109 /* Slow path 3) release the global lock */
110 spin_unlock(&nf_conntrack_locks_all_lock
);
112 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
114 static void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
116 h1
%= CONNTRACK_LOCKS
;
117 h2
%= CONNTRACK_LOCKS
;
118 spin_unlock(&nf_conntrack_locks
[h1
]);
120 spin_unlock(&nf_conntrack_locks
[h2
]);
123 /* return true if we need to recompute hashes (in case hash table was resized) */
124 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
125 unsigned int h2
, unsigned int sequence
)
127 h1
%= CONNTRACK_LOCKS
;
128 h2
%= CONNTRACK_LOCKS
;
130 nf_conntrack_lock(&nf_conntrack_locks
[h1
]);
132 spin_lock_nested(&nf_conntrack_locks
[h2
],
133 SINGLE_DEPTH_NESTING
);
135 nf_conntrack_lock(&nf_conntrack_locks
[h2
]);
136 spin_lock_nested(&nf_conntrack_locks
[h1
],
137 SINGLE_DEPTH_NESTING
);
139 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
140 nf_conntrack_double_unlock(h1
, h2
);
146 static void nf_conntrack_all_lock(void)
150 spin_lock(&nf_conntrack_locks_all_lock
);
152 nf_conntrack_locks_all
= true;
154 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++) {
155 spin_lock(&nf_conntrack_locks
[i
]);
157 /* This spin_unlock provides the "release" to ensure that
158 * nf_conntrack_locks_all==true is visible to everyone that
159 * acquired spin_lock(&nf_conntrack_locks[]).
161 spin_unlock(&nf_conntrack_locks
[i
]);
165 static void nf_conntrack_all_unlock(void)
167 /* All prior stores must be complete before we clear
168 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
169 * might observe the false value but not the entire
171 * It pairs with the smp_load_acquire() in nf_conntrack_lock()
173 smp_store_release(&nf_conntrack_locks_all
, false);
174 spin_unlock(&nf_conntrack_locks_all_lock
);
177 unsigned int nf_conntrack_htable_size __read_mostly
;
178 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
180 unsigned int nf_conntrack_max __read_mostly
;
181 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
182 seqcount_t nf_conntrack_generation __read_mostly
;
183 static unsigned int nf_conntrack_hash_rnd __read_mostly
;
185 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
,
186 const struct net
*net
)
191 get_random_once(&nf_conntrack_hash_rnd
, sizeof(nf_conntrack_hash_rnd
));
193 /* The direction must be ignored, so we hash everything up to the
194 * destination ports (which is a multiple of 4) and treat the last
195 * three bytes manually.
197 seed
= nf_conntrack_hash_rnd
^ net_hash_mix(net
);
198 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
199 return jhash2((u32
*)tuple
, n
, seed
^
200 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
201 tuple
->dst
.protonum
));
204 static u32
scale_hash(u32 hash
)
206 return reciprocal_scale(hash
, nf_conntrack_htable_size
);
209 static u32
__hash_conntrack(const struct net
*net
,
210 const struct nf_conntrack_tuple
*tuple
,
213 return reciprocal_scale(hash_conntrack_raw(tuple
, net
), size
);
216 static u32
hash_conntrack(const struct net
*net
,
217 const struct nf_conntrack_tuple
*tuple
)
219 return scale_hash(hash_conntrack_raw(tuple
, net
));
222 static bool nf_ct_get_tuple_ports(const struct sk_buff
*skb
,
223 unsigned int dataoff
,
224 struct nf_conntrack_tuple
*tuple
)
228 } _inet_hdr
, *inet_hdr
;
230 /* Actually only need first 4 bytes to get ports. */
231 inet_hdr
= skb_header_pointer(skb
, dataoff
, sizeof(_inet_hdr
), &_inet_hdr
);
235 tuple
->src
.u
.udp
.port
= inet_hdr
->sport
;
236 tuple
->dst
.u
.udp
.port
= inet_hdr
->dport
;
241 nf_ct_get_tuple(const struct sk_buff
*skb
,
243 unsigned int dataoff
,
247 struct nf_conntrack_tuple
*tuple
)
253 memset(tuple
, 0, sizeof(*tuple
));
255 tuple
->src
.l3num
= l3num
;
258 nhoff
+= offsetof(struct iphdr
, saddr
);
259 size
= 2 * sizeof(__be32
);
262 nhoff
+= offsetof(struct ipv6hdr
, saddr
);
263 size
= sizeof(_addrs
);
269 ap
= skb_header_pointer(skb
, nhoff
, size
, _addrs
);
275 tuple
->src
.u3
.ip
= ap
[0];
276 tuple
->dst
.u3
.ip
= ap
[1];
279 memcpy(tuple
->src
.u3
.ip6
, ap
, sizeof(tuple
->src
.u3
.ip6
));
280 memcpy(tuple
->dst
.u3
.ip6
, ap
+ 4, sizeof(tuple
->dst
.u3
.ip6
));
284 tuple
->dst
.protonum
= protonum
;
285 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
288 #if IS_ENABLED(CONFIG_IPV6)
290 return icmpv6_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
293 return icmp_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
294 #ifdef CONFIG_NF_CT_PROTO_GRE
296 return gre_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
299 case IPPROTO_UDP
: /* fallthrough */
300 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
301 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
302 case IPPROTO_UDPLITE
:
303 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
305 #ifdef CONFIG_NF_CT_PROTO_SCTP
307 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
309 #ifdef CONFIG_NF_CT_PROTO_DCCP
311 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
320 static int ipv4_get_l4proto(const struct sk_buff
*skb
, unsigned int nhoff
,
324 const struct iphdr
*iph
;
327 iph
= skb_header_pointer(skb
, nhoff
, sizeof(_iph
), &_iph
);
331 /* Conntrack defragments packets, we might still see fragments
332 * inside ICMP packets though.
334 if (iph
->frag_off
& htons(IP_OFFSET
))
337 dataoff
= nhoff
+ (iph
->ihl
<< 2);
338 *protonum
= iph
->protocol
;
340 /* Check bogus IP headers */
341 if (dataoff
> skb
->len
) {
342 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
343 nhoff
, iph
->ihl
<< 2, skb
->len
);
349 #if IS_ENABLED(CONFIG_IPV6)
350 static int ipv6_get_l4proto(const struct sk_buff
*skb
, unsigned int nhoff
,
354 unsigned int extoff
= nhoff
+ sizeof(struct ipv6hdr
);
358 if (skb_copy_bits(skb
, nhoff
+ offsetof(struct ipv6hdr
, nexthdr
),
359 &nexthdr
, sizeof(nexthdr
)) != 0) {
360 pr_debug("can't get nexthdr\n");
363 protoff
= ipv6_skip_exthdr(skb
, extoff
, &nexthdr
, &frag_off
);
365 * (protoff == skb->len) means the packet has not data, just
366 * IPv6 and possibly extensions headers, but it is tracked anyway
368 if (protoff
< 0 || (frag_off
& htons(~0x7)) != 0) {
369 pr_debug("can't find proto in pkt\n");
378 static int get_l4proto(const struct sk_buff
*skb
,
379 unsigned int nhoff
, u8 pf
, u8
*l4num
)
383 return ipv4_get_l4proto(skb
, nhoff
, l4num
);
384 #if IS_ENABLED(CONFIG_IPV6)
386 return ipv6_get_l4proto(skb
, nhoff
, l4num
);
395 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
397 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
402 protoff
= get_l4proto(skb
, nhoff
, l3num
, &protonum
);
406 return nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
);
408 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
411 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
412 const struct nf_conntrack_tuple
*orig
)
414 memset(inverse
, 0, sizeof(*inverse
));
416 inverse
->src
.l3num
= orig
->src
.l3num
;
418 switch (orig
->src
.l3num
) {
420 inverse
->src
.u3
.ip
= orig
->dst
.u3
.ip
;
421 inverse
->dst
.u3
.ip
= orig
->src
.u3
.ip
;
424 inverse
->src
.u3
.in6
= orig
->dst
.u3
.in6
;
425 inverse
->dst
.u3
.in6
= orig
->src
.u3
.in6
;
431 inverse
->dst
.dir
= !orig
->dst
.dir
;
433 inverse
->dst
.protonum
= orig
->dst
.protonum
;
435 switch (orig
->dst
.protonum
) {
437 return nf_conntrack_invert_icmp_tuple(inverse
, orig
);
438 #if IS_ENABLED(CONFIG_IPV6)
440 return nf_conntrack_invert_icmpv6_tuple(inverse
, orig
);
444 inverse
->src
.u
.all
= orig
->dst
.u
.all
;
445 inverse
->dst
.u
.all
= orig
->src
.u
.all
;
448 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
450 /* Generate a almost-unique pseudo-id for a given conntrack.
452 * intentionally doesn't re-use any of the seeds used for hash
453 * table location, we assume id gets exposed to userspace.
455 * Following nf_conn items do not change throughout lifetime
456 * of the nf_conn after it has been committed to main hash table:
459 * 2. nf_conn->ext address
460 * 3. nf_conn->master address (normally NULL)
462 * 5. the associated net namespace
464 u32
nf_ct_get_id(const struct nf_conn
*ct
)
466 static __read_mostly siphash_key_t ct_id_seed
;
467 unsigned long a
, b
, c
, d
;
469 net_get_random_once(&ct_id_seed
, sizeof(ct_id_seed
));
471 a
= (unsigned long)ct
;
472 b
= (unsigned long)ct
->master
^ net_hash_mix(nf_ct_net(ct
));
473 c
= (unsigned long)ct
->ext
;
474 d
= (unsigned long)siphash(&ct
->tuplehash
, sizeof(ct
->tuplehash
),
477 return siphash_4u64((u64
)a
, (u64
)b
, (u64
)c
, (u64
)d
, &ct_id_seed
);
479 return siphash_4u32((u32
)a
, (u32
)b
, (u32
)c
, (u32
)d
, &ct_id_seed
);
482 EXPORT_SYMBOL_GPL(nf_ct_get_id
);
485 clean_from_lists(struct nf_conn
*ct
)
487 pr_debug("clean_from_lists(%p)\n", ct
);
488 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
489 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
491 /* Destroy all pending expectations */
492 nf_ct_remove_expectations(ct
);
495 /* must be called with local_bh_disable */
496 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
498 struct ct_pcpu
*pcpu
;
500 /* add this conntrack to the (per cpu) dying list */
501 ct
->cpu
= smp_processor_id();
502 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
504 spin_lock(&pcpu
->lock
);
505 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
507 spin_unlock(&pcpu
->lock
);
510 /* must be called with local_bh_disable */
511 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
513 struct ct_pcpu
*pcpu
;
515 /* add this conntrack to the (per cpu) unconfirmed list */
516 ct
->cpu
= smp_processor_id();
517 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
519 spin_lock(&pcpu
->lock
);
520 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
522 spin_unlock(&pcpu
->lock
);
525 /* must be called with local_bh_disable */
526 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
528 struct ct_pcpu
*pcpu
;
530 /* We overload first tuple to link into unconfirmed or dying list.*/
531 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
533 spin_lock(&pcpu
->lock
);
534 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
535 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
536 spin_unlock(&pcpu
->lock
);
539 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
541 /* Released via destroy_conntrack() */
542 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
543 const struct nf_conntrack_zone
*zone
,
546 struct nf_conn
*tmpl
, *p
;
548 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
) {
549 tmpl
= kzalloc(sizeof(*tmpl
) + NFCT_INFOMASK
, flags
);
554 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
556 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
557 tmpl
->proto
.tmpl_padto
= (char *)tmpl
- (char *)p
;
560 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
565 tmpl
->status
= IPS_TEMPLATE
;
566 write_pnet(&tmpl
->ct_net
, net
);
567 nf_ct_zone_add(tmpl
, zone
);
568 atomic_set(&tmpl
->ct_general
.use
, 0);
572 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
574 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
576 nf_ct_ext_destroy(tmpl
);
577 nf_ct_ext_free(tmpl
);
579 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
)
580 kfree((char *)tmpl
- tmpl
->proto
.tmpl_padto
);
584 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free
);
586 static void destroy_gre_conntrack(struct nf_conn
*ct
)
588 #ifdef CONFIG_NF_CT_PROTO_GRE
589 struct nf_conn
*master
= ct
->master
;
592 nf_ct_gre_keymap_destroy(master
);
597 destroy_conntrack(struct nf_conntrack
*nfct
)
599 struct nf_conn
*ct
= (struct nf_conn
*)nfct
;
601 pr_debug("destroy_conntrack(%p)\n", ct
);
602 WARN_ON(atomic_read(&nfct
->use
) != 0);
604 if (unlikely(nf_ct_is_template(ct
))) {
609 if (unlikely(nf_ct_protonum(ct
) == IPPROTO_GRE
))
610 destroy_gre_conntrack(ct
);
613 /* Expectations will have been removed in clean_from_lists,
614 * except TFTP can create an expectation on the first packet,
615 * before connection is in the list, so we need to clean here,
618 nf_ct_remove_expectations(ct
);
620 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
625 nf_ct_put(ct
->master
);
627 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct
);
628 nf_conntrack_free(ct
);
631 static void nf_ct_delete_from_lists(struct nf_conn
*ct
)
633 struct net
*net
= nf_ct_net(ct
);
634 unsigned int hash
, reply_hash
;
635 unsigned int sequence
;
637 nf_ct_helper_destroy(ct
);
641 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
642 hash
= hash_conntrack(net
,
643 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
644 reply_hash
= hash_conntrack(net
,
645 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
646 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
648 clean_from_lists(ct
);
649 nf_conntrack_double_unlock(hash
, reply_hash
);
651 nf_ct_add_to_dying_list(ct
);
656 bool nf_ct_delete(struct nf_conn
*ct
, u32 portid
, int report
)
658 struct nf_conn_tstamp
*tstamp
;
660 if (test_and_set_bit(IPS_DYING_BIT
, &ct
->status
))
663 tstamp
= nf_conn_tstamp_find(ct
);
664 if (tstamp
&& tstamp
->stop
== 0)
665 tstamp
->stop
= ktime_get_real_ns();
667 if (nf_conntrack_event_report(IPCT_DESTROY
, ct
,
668 portid
, report
) < 0) {
669 /* destroy event was not delivered. nf_ct_put will
670 * be done by event cache worker on redelivery.
672 nf_ct_delete_from_lists(ct
);
673 nf_conntrack_ecache_delayed_work(nf_ct_net(ct
));
677 nf_conntrack_ecache_work(nf_ct_net(ct
));
678 nf_ct_delete_from_lists(ct
);
682 EXPORT_SYMBOL_GPL(nf_ct_delete
);
685 nf_ct_key_equal(struct nf_conntrack_tuple_hash
*h
,
686 const struct nf_conntrack_tuple
*tuple
,
687 const struct nf_conntrack_zone
*zone
,
688 const struct net
*net
)
690 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
692 /* A conntrack can be recreated with the equal tuple,
693 * so we need to check that the conntrack is confirmed
695 return nf_ct_tuple_equal(tuple
, &h
->tuple
) &&
696 nf_ct_zone_equal(ct
, zone
, NF_CT_DIRECTION(h
)) &&
697 nf_ct_is_confirmed(ct
) &&
698 net_eq(net
, nf_ct_net(ct
));
702 nf_ct_match(const struct nf_conn
*ct1
, const struct nf_conn
*ct2
)
704 return nf_ct_tuple_equal(&ct1
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
705 &ct2
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
) &&
706 nf_ct_tuple_equal(&ct1
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
707 &ct2
->tuplehash
[IP_CT_DIR_REPLY
].tuple
) &&
708 nf_ct_zone_equal(ct1
, nf_ct_zone(ct2
), IP_CT_DIR_ORIGINAL
) &&
709 nf_ct_zone_equal(ct1
, nf_ct_zone(ct2
), IP_CT_DIR_REPLY
) &&
710 net_eq(nf_ct_net(ct1
), nf_ct_net(ct2
));
713 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
714 static void nf_ct_gc_expired(struct nf_conn
*ct
)
716 if (!atomic_inc_not_zero(&ct
->ct_general
.use
))
719 if (nf_ct_should_gc(ct
))
727 * - Caller must take a reference on returned object
728 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
730 static struct nf_conntrack_tuple_hash
*
731 ____nf_conntrack_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
732 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
734 struct nf_conntrack_tuple_hash
*h
;
735 struct hlist_nulls_head
*ct_hash
;
736 struct hlist_nulls_node
*n
;
737 unsigned int bucket
, hsize
;
740 nf_conntrack_get_ht(&ct_hash
, &hsize
);
741 bucket
= reciprocal_scale(hash
, hsize
);
743 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[bucket
], hnnode
) {
746 ct
= nf_ct_tuplehash_to_ctrack(h
);
747 if (nf_ct_is_expired(ct
)) {
748 nf_ct_gc_expired(ct
);
752 if (nf_ct_is_dying(ct
))
755 if (nf_ct_key_equal(h
, tuple
, zone
, net
))
759 * if the nulls value we got at the end of this lookup is
760 * not the expected one, we must restart lookup.
761 * We probably met an item that was moved to another chain.
763 if (get_nulls_value(n
) != bucket
) {
764 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
771 /* Find a connection corresponding to a tuple. */
772 static struct nf_conntrack_tuple_hash
*
773 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
774 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
776 struct nf_conntrack_tuple_hash
*h
;
781 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
783 ct
= nf_ct_tuplehash_to_ctrack(h
);
784 if (unlikely(nf_ct_is_dying(ct
) ||
785 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
788 if (unlikely(!nf_ct_key_equal(h
, tuple
, zone
, net
))) {
799 struct nf_conntrack_tuple_hash
*
800 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
801 const struct nf_conntrack_tuple
*tuple
)
803 return __nf_conntrack_find_get(net
, zone
, tuple
,
804 hash_conntrack_raw(tuple
, net
));
806 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
808 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
810 unsigned int reply_hash
)
812 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
813 &nf_conntrack_hash
[hash
]);
814 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
815 &nf_conntrack_hash
[reply_hash
]);
819 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
821 const struct nf_conntrack_zone
*zone
;
822 struct net
*net
= nf_ct_net(ct
);
823 unsigned int hash
, reply_hash
;
824 struct nf_conntrack_tuple_hash
*h
;
825 struct hlist_nulls_node
*n
;
826 unsigned int sequence
;
828 zone
= nf_ct_zone(ct
);
832 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
833 hash
= hash_conntrack(net
,
834 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
835 reply_hash
= hash_conntrack(net
,
836 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
837 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
839 /* See if there's one in the list already, including reverse */
840 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
841 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
845 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
846 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
851 /* The caller holds a reference to this object */
852 atomic_set(&ct
->ct_general
.use
, 2);
853 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
854 nf_conntrack_double_unlock(hash
, reply_hash
);
855 NF_CT_STAT_INC(net
, insert
);
860 nf_conntrack_double_unlock(hash
, reply_hash
);
861 NF_CT_STAT_INC(net
, insert_failed
);
865 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
867 static inline void nf_ct_acct_update(struct nf_conn
*ct
,
868 enum ip_conntrack_info ctinfo
,
871 struct nf_conn_acct
*acct
;
873 acct
= nf_conn_acct_find(ct
);
875 struct nf_conn_counter
*counter
= acct
->counter
;
877 atomic64_inc(&counter
[CTINFO2DIR(ctinfo
)].packets
);
878 atomic64_add(len
, &counter
[CTINFO2DIR(ctinfo
)].bytes
);
882 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
883 const struct nf_conn
*loser_ct
)
885 struct nf_conn_acct
*acct
;
887 acct
= nf_conn_acct_find(loser_ct
);
889 struct nf_conn_counter
*counter
= acct
->counter
;
892 /* u32 should be fine since we must have seen one packet. */
893 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
894 nf_ct_acct_update(ct
, ctinfo
, bytes
);
898 /* Resolve race on insertion if this protocol allows this. */
899 static int nf_ct_resolve_clash(struct net
*net
, struct sk_buff
*skb
,
900 enum ip_conntrack_info ctinfo
,
901 struct nf_conntrack_tuple_hash
*h
)
903 /* This is the conntrack entry already in hashes that won race. */
904 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
905 const struct nf_conntrack_l4proto
*l4proto
;
906 enum ip_conntrack_info oldinfo
;
907 struct nf_conn
*loser_ct
= nf_ct_get(skb
, &oldinfo
);
909 l4proto
= nf_ct_l4proto_find(nf_ct_protonum(ct
));
910 if (l4proto
->allow_clash
&&
911 !nf_ct_is_dying(ct
) &&
912 atomic_inc_not_zero(&ct
->ct_general
.use
)) {
913 if (((ct
->status
& IPS_NAT_DONE_MASK
) == 0) ||
914 nf_ct_match(ct
, loser_ct
)) {
915 nf_ct_acct_merge(ct
, ctinfo
, loser_ct
);
916 nf_conntrack_put(&loser_ct
->ct_general
);
917 nf_ct_set(skb
, ct
, oldinfo
);
922 NF_CT_STAT_INC(net
, drop
);
926 /* Confirm a connection given skb; places it in hash table */
928 __nf_conntrack_confirm(struct sk_buff
*skb
)
930 const struct nf_conntrack_zone
*zone
;
931 unsigned int hash
, reply_hash
;
932 struct nf_conntrack_tuple_hash
*h
;
934 struct nf_conn_help
*help
;
935 struct nf_conn_tstamp
*tstamp
;
936 struct hlist_nulls_node
*n
;
937 enum ip_conntrack_info ctinfo
;
939 unsigned int sequence
;
942 ct
= nf_ct_get(skb
, &ctinfo
);
945 /* ipt_REJECT uses nf_conntrack_attach to attach related
946 ICMP/TCP RST packets in other direction. Actual packet
947 which created connection will be IP_CT_NEW or for an
948 expected connection, IP_CT_RELATED. */
949 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
952 zone
= nf_ct_zone(ct
);
956 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
957 /* reuse the hash saved before */
958 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
959 hash
= scale_hash(hash
);
960 reply_hash
= hash_conntrack(net
,
961 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
963 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
965 /* We're not in hash table, and we refuse to set up related
966 * connections for unconfirmed conns. But packet copies and
967 * REJECT will give spurious warnings here.
970 /* Another skb with the same unconfirmed conntrack may
971 * win the race. This may happen for bridge(br_flood)
972 * or broadcast/multicast packets do skb_clone with
973 * unconfirmed conntrack.
975 if (unlikely(nf_ct_is_confirmed(ct
))) {
977 nf_conntrack_double_unlock(hash
, reply_hash
);
982 pr_debug("Confirming conntrack %p\n", ct
);
983 /* We have to check the DYING flag after unlink to prevent
984 * a race against nf_ct_get_next_corpse() possibly called from
985 * user context, else we insert an already 'dead' hash, blocking
986 * further use of that particular connection -JM.
988 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
990 if (unlikely(nf_ct_is_dying(ct
))) {
991 nf_ct_add_to_dying_list(ct
);
995 /* See if there's one in the list already, including reverse:
996 NAT could have grabbed it without realizing, since we're
997 not in the hash. If there is, we lost race. */
998 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
999 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
1003 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
1004 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
1008 /* Timer relative to confirmation time, not original
1009 setting time, otherwise we'd get timer wrap in
1010 weird delay cases. */
1011 ct
->timeout
+= nfct_time_stamp
;
1012 atomic_inc(&ct
->ct_general
.use
);
1013 ct
->status
|= IPS_CONFIRMED
;
1015 /* set conntrack timestamp, if enabled. */
1016 tstamp
= nf_conn_tstamp_find(ct
);
1018 tstamp
->start
= ktime_get_real_ns();
1020 /* Since the lookup is lockless, hash insertion must be done after
1021 * starting the timer and setting the CONFIRMED bit. The RCU barriers
1022 * guarantee that no other CPU can find the conntrack before the above
1023 * stores are visible.
1025 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
1026 nf_conntrack_double_unlock(hash
, reply_hash
);
1029 help
= nfct_help(ct
);
1030 if (help
&& help
->helper
)
1031 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
1033 nf_conntrack_event_cache(master_ct(ct
) ?
1034 IPCT_RELATED
: IPCT_NEW
, ct
);
1038 nf_ct_add_to_dying_list(ct
);
1039 ret
= nf_ct_resolve_clash(net
, skb
, ctinfo
, h
);
1041 nf_conntrack_double_unlock(hash
, reply_hash
);
1042 NF_CT_STAT_INC(net
, insert_failed
);
1046 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
1048 /* Returns true if a connection correspondings to the tuple (required
1051 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
1052 const struct nf_conn
*ignored_conntrack
)
1054 struct net
*net
= nf_ct_net(ignored_conntrack
);
1055 const struct nf_conntrack_zone
*zone
;
1056 struct nf_conntrack_tuple_hash
*h
;
1057 struct hlist_nulls_head
*ct_hash
;
1058 unsigned int hash
, hsize
;
1059 struct hlist_nulls_node
*n
;
1062 zone
= nf_ct_zone(ignored_conntrack
);
1066 nf_conntrack_get_ht(&ct_hash
, &hsize
);
1067 hash
= __hash_conntrack(net
, tuple
, hsize
);
1069 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
1070 ct
= nf_ct_tuplehash_to_ctrack(h
);
1072 if (ct
== ignored_conntrack
)
1075 if (nf_ct_is_expired(ct
)) {
1076 nf_ct_gc_expired(ct
);
1080 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
1081 /* Tuple is taken already, so caller will need to find
1082 * a new source port to use.
1085 * If the *original tuples* are identical, then both
1086 * conntracks refer to the same flow.
1087 * This is a rare situation, it can occur e.g. when
1088 * more than one UDP packet is sent from same socket
1089 * in different threads.
1091 * Let nf_ct_resolve_clash() deal with this later.
1093 if (nf_ct_tuple_equal(&ignored_conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
1094 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
))
1097 NF_CT_STAT_INC_ATOMIC(net
, found
);
1103 if (get_nulls_value(n
) != hash
) {
1104 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
1112 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
1114 #define NF_CT_EVICTION_RANGE 8
1116 /* There's a small race here where we may free a just-assured
1117 connection. Too bad: we're in trouble anyway. */
1118 static unsigned int early_drop_list(struct net
*net
,
1119 struct hlist_nulls_head
*head
)
1121 struct nf_conntrack_tuple_hash
*h
;
1122 struct hlist_nulls_node
*n
;
1123 unsigned int drops
= 0;
1124 struct nf_conn
*tmp
;
1126 hlist_nulls_for_each_entry_rcu(h
, n
, head
, hnnode
) {
1127 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1129 if (test_bit(IPS_OFFLOAD_BIT
, &tmp
->status
))
1132 if (nf_ct_is_expired(tmp
)) {
1133 nf_ct_gc_expired(tmp
);
1137 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
1138 !net_eq(nf_ct_net(tmp
), net
) ||
1139 nf_ct_is_dying(tmp
))
1142 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1145 /* kill only if still in same netns -- might have moved due to
1146 * SLAB_TYPESAFE_BY_RCU rules.
1148 * We steal the timer reference. If that fails timer has
1149 * already fired or someone else deleted it. Just drop ref
1150 * and move to next entry.
1152 if (net_eq(nf_ct_net(tmp
), net
) &&
1153 nf_ct_is_confirmed(tmp
) &&
1154 nf_ct_delete(tmp
, 0, 0))
1163 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
1165 unsigned int i
, bucket
;
1167 for (i
= 0; i
< NF_CT_EVICTION_RANGE
; i
++) {
1168 struct hlist_nulls_head
*ct_hash
;
1169 unsigned int hsize
, drops
;
1172 nf_conntrack_get_ht(&ct_hash
, &hsize
);
1174 bucket
= reciprocal_scale(hash
, hsize
);
1176 bucket
= (bucket
+ 1) % hsize
;
1178 drops
= early_drop_list(net
, &ct_hash
[bucket
]);
1182 NF_CT_STAT_ADD_ATOMIC(net
, early_drop
, drops
);
1190 static bool gc_worker_skip_ct(const struct nf_conn
*ct
)
1192 return !nf_ct_is_confirmed(ct
) || nf_ct_is_dying(ct
);
1195 static bool gc_worker_can_early_drop(const struct nf_conn
*ct
)
1197 const struct nf_conntrack_l4proto
*l4proto
;
1199 if (!test_bit(IPS_ASSURED_BIT
, &ct
->status
))
1202 l4proto
= nf_ct_l4proto_find(nf_ct_protonum(ct
));
1203 if (l4proto
->can_early_drop
&& l4proto
->can_early_drop(ct
))
1209 #define DAY (86400 * HZ)
1211 /* Set an arbitrary timeout large enough not to ever expire, this save
1212 * us a check for the IPS_OFFLOAD_BIT from the packet path via
1213 * nf_ct_is_expired().
1215 static void nf_ct_offload_timeout(struct nf_conn
*ct
)
1217 if (nf_ct_expires(ct
) < DAY
/ 2)
1218 ct
->timeout
= nfct_time_stamp
+ DAY
;
1221 static void gc_worker(struct work_struct
*work
)
1223 unsigned int min_interval
= max(HZ
/ GC_MAX_BUCKETS_DIV
, 1u);
1224 unsigned int i
, goal
, buckets
= 0, expired_count
= 0;
1225 unsigned int nf_conntrack_max95
= 0;
1226 struct conntrack_gc_work
*gc_work
;
1227 unsigned int ratio
, scanned
= 0;
1228 unsigned long next_run
;
1230 gc_work
= container_of(work
, struct conntrack_gc_work
, dwork
.work
);
1232 goal
= nf_conntrack_htable_size
/ GC_MAX_BUCKETS_DIV
;
1233 i
= gc_work
->last_bucket
;
1234 if (gc_work
->early_drop
)
1235 nf_conntrack_max95
= nf_conntrack_max
/ 100u * 95u;
1238 struct nf_conntrack_tuple_hash
*h
;
1239 struct hlist_nulls_head
*ct_hash
;
1240 struct hlist_nulls_node
*n
;
1241 unsigned int hashsz
;
1242 struct nf_conn
*tmp
;
1247 nf_conntrack_get_ht(&ct_hash
, &hashsz
);
1251 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[i
], hnnode
) {
1254 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1257 if (test_bit(IPS_OFFLOAD_BIT
, &tmp
->status
)) {
1258 nf_ct_offload_timeout(tmp
);
1262 if (nf_ct_is_expired(tmp
)) {
1263 nf_ct_gc_expired(tmp
);
1268 if (nf_conntrack_max95
== 0 || gc_worker_skip_ct(tmp
))
1271 net
= nf_ct_net(tmp
);
1272 if (atomic_read(&net
->ct
.count
) < nf_conntrack_max95
)
1275 /* need to take reference to avoid possible races */
1276 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1279 if (gc_worker_skip_ct(tmp
)) {
1284 if (gc_worker_can_early_drop(tmp
))
1290 /* could check get_nulls_value() here and restart if ct
1291 * was moved to another chain. But given gc is best-effort
1292 * we will just continue with next hash slot.
1296 } while (++buckets
< goal
);
1298 if (gc_work
->exiting
)
1302 * Eviction will normally happen from the packet path, and not
1303 * from this gc worker.
1305 * This worker is only here to reap expired entries when system went
1306 * idle after a busy period.
1308 * The heuristics below are supposed to balance conflicting goals:
1310 * 1. Minimize time until we notice a stale entry
1311 * 2. Maximize scan intervals to not waste cycles
1313 * Normally, expire ratio will be close to 0.
1315 * As soon as a sizeable fraction of the entries have expired
1316 * increase scan frequency.
1318 ratio
= scanned
? expired_count
* 100 / scanned
: 0;
1319 if (ratio
> GC_EVICT_RATIO
) {
1320 gc_work
->next_gc_run
= min_interval
;
1322 unsigned int max
= GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
;
1324 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
) == 0);
1326 gc_work
->next_gc_run
+= min_interval
;
1327 if (gc_work
->next_gc_run
> max
)
1328 gc_work
->next_gc_run
= max
;
1331 next_run
= gc_work
->next_gc_run
;
1332 gc_work
->last_bucket
= i
;
1333 gc_work
->early_drop
= false;
1334 queue_delayed_work(system_power_efficient_wq
, &gc_work
->dwork
, next_run
);
1337 static void conntrack_gc_work_init(struct conntrack_gc_work
*gc_work
)
1339 INIT_DEFERRABLE_WORK(&gc_work
->dwork
, gc_worker
);
1340 gc_work
->next_gc_run
= HZ
;
1341 gc_work
->exiting
= false;
1344 static struct nf_conn
*
1345 __nf_conntrack_alloc(struct net
*net
,
1346 const struct nf_conntrack_zone
*zone
,
1347 const struct nf_conntrack_tuple
*orig
,
1348 const struct nf_conntrack_tuple
*repl
,
1349 gfp_t gfp
, u32 hash
)
1353 /* We don't want any race condition at early drop stage */
1354 atomic_inc(&net
->ct
.count
);
1356 if (nf_conntrack_max
&&
1357 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
1358 if (!early_drop(net
, hash
)) {
1359 if (!conntrack_gc_work
.early_drop
)
1360 conntrack_gc_work
.early_drop
= true;
1361 atomic_dec(&net
->ct
.count
);
1362 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1363 return ERR_PTR(-ENOMEM
);
1368 * Do not use kmem_cache_zalloc(), as this cache uses
1369 * SLAB_TYPESAFE_BY_RCU.
1371 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
1375 spin_lock_init(&ct
->lock
);
1376 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
1377 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
1378 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
1379 /* save hash for reusing when confirming */
1380 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
1383 write_pnet(&ct
->ct_net
, net
);
1384 memset(&ct
->__nfct_init_offset
[0], 0,
1385 offsetof(struct nf_conn
, proto
) -
1386 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
1388 nf_ct_zone_add(ct
, zone
);
1390 /* Because we use RCU lookups, we set ct_general.use to zero before
1391 * this is inserted in any list.
1393 atomic_set(&ct
->ct_general
.use
, 0);
1396 atomic_dec(&net
->ct
.count
);
1397 return ERR_PTR(-ENOMEM
);
1400 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
1401 const struct nf_conntrack_zone
*zone
,
1402 const struct nf_conntrack_tuple
*orig
,
1403 const struct nf_conntrack_tuple
*repl
,
1406 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
1408 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
1410 void nf_conntrack_free(struct nf_conn
*ct
)
1412 struct net
*net
= nf_ct_net(ct
);
1414 /* A freed object has refcnt == 0, that's
1415 * the golden rule for SLAB_TYPESAFE_BY_RCU
1417 WARN_ON(atomic_read(&ct
->ct_general
.use
) != 0);
1419 nf_ct_ext_destroy(ct
);
1421 kmem_cache_free(nf_conntrack_cachep
, ct
);
1422 smp_mb__before_atomic();
1423 atomic_dec(&net
->ct
.count
);
1425 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
1428 /* Allocate a new conntrack: we return -ENOMEM if classification
1429 failed due to stress. Otherwise it really is unclassifiable. */
1430 static noinline
struct nf_conntrack_tuple_hash
*
1431 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
1432 const struct nf_conntrack_tuple
*tuple
,
1433 struct sk_buff
*skb
,
1434 unsigned int dataoff
, u32 hash
)
1437 struct nf_conn_help
*help
;
1438 struct nf_conntrack_tuple repl_tuple
;
1439 struct nf_conntrack_ecache
*ecache
;
1440 struct nf_conntrack_expect
*exp
= NULL
;
1441 const struct nf_conntrack_zone
*zone
;
1442 struct nf_conn_timeout
*timeout_ext
;
1443 struct nf_conntrack_zone tmp
;
1445 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
)) {
1446 pr_debug("Can't invert tuple.\n");
1450 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1451 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
1454 return (struct nf_conntrack_tuple_hash
*)ct
;
1456 if (!nf_ct_add_synproxy(ct
, tmpl
)) {
1457 nf_conntrack_free(ct
);
1458 return ERR_PTR(-ENOMEM
);
1461 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1464 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1467 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1468 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1469 nf_ct_labels_ext_add(ct
);
1471 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1472 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1473 ecache
? ecache
->expmask
: 0,
1477 if (net
->ct
.expect_count
) {
1478 spin_lock(&nf_conntrack_expect_lock
);
1479 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1481 pr_debug("expectation arrives ct=%p exp=%p\n",
1483 /* Welcome, Mr. Bond. We've been expecting you... */
1484 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1485 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1486 ct
->master
= exp
->master
;
1488 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
1490 rcu_assign_pointer(help
->helper
, exp
->helper
);
1493 #ifdef CONFIG_NF_CONNTRACK_MARK
1494 ct
->mark
= exp
->master
->mark
;
1496 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1497 ct
->secmark
= exp
->master
->secmark
;
1499 NF_CT_STAT_INC(net
, expect_new
);
1501 spin_unlock(&nf_conntrack_expect_lock
);
1504 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1506 /* Now it is inserted into the unconfirmed list, bump refcount */
1507 nf_conntrack_get(&ct
->ct_general
);
1508 nf_ct_add_to_unconfirmed_list(ct
);
1514 exp
->expectfn(ct
, exp
);
1515 nf_ct_expect_put(exp
);
1518 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1521 /* On success, returns 0, sets skb->_nfct | ctinfo */
1523 resolve_normal_ct(struct nf_conn
*tmpl
,
1524 struct sk_buff
*skb
,
1525 unsigned int dataoff
,
1527 const struct nf_hook_state
*state
)
1529 const struct nf_conntrack_zone
*zone
;
1530 struct nf_conntrack_tuple tuple
;
1531 struct nf_conntrack_tuple_hash
*h
;
1532 enum ip_conntrack_info ctinfo
;
1533 struct nf_conntrack_zone tmp
;
1537 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1538 dataoff
, state
->pf
, protonum
, state
->net
,
1540 pr_debug("Can't get tuple\n");
1544 /* look for tuple match */
1545 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1546 hash
= hash_conntrack_raw(&tuple
, state
->net
);
1547 h
= __nf_conntrack_find_get(state
->net
, zone
, &tuple
, hash
);
1549 h
= init_conntrack(state
->net
, tmpl
, &tuple
,
1550 skb
, dataoff
, hash
);
1556 ct
= nf_ct_tuplehash_to_ctrack(h
);
1558 /* It exists; we have (non-exclusive) reference. */
1559 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1560 ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1562 /* Once we've had two way comms, always ESTABLISHED. */
1563 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1564 pr_debug("normal packet for %p\n", ct
);
1565 ctinfo
= IP_CT_ESTABLISHED
;
1566 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1567 pr_debug("related packet for %p\n", ct
);
1568 ctinfo
= IP_CT_RELATED
;
1570 pr_debug("new packet for %p\n", ct
);
1574 nf_ct_set(skb
, ct
, ctinfo
);
1579 * icmp packets need special treatment to handle error messages that are
1580 * related to a connection.
1582 * Callers need to check if skb has a conntrack assigned when this
1583 * helper returns; in such case skb belongs to an already known connection.
1585 static unsigned int __cold
1586 nf_conntrack_handle_icmp(struct nf_conn
*tmpl
,
1587 struct sk_buff
*skb
,
1588 unsigned int dataoff
,
1590 const struct nf_hook_state
*state
)
1594 if (state
->pf
== NFPROTO_IPV4
&& protonum
== IPPROTO_ICMP
)
1595 ret
= nf_conntrack_icmpv4_error(tmpl
, skb
, dataoff
, state
);
1596 #if IS_ENABLED(CONFIG_IPV6)
1597 else if (state
->pf
== NFPROTO_IPV6
&& protonum
== IPPROTO_ICMPV6
)
1598 ret
= nf_conntrack_icmpv6_error(tmpl
, skb
, dataoff
, state
);
1604 NF_CT_STAT_INC_ATOMIC(state
->net
, error
);
1605 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1611 static int generic_packet(struct nf_conn
*ct
, struct sk_buff
*skb
,
1612 enum ip_conntrack_info ctinfo
)
1614 const unsigned int *timeout
= nf_ct_timeout_lookup(ct
);
1617 timeout
= &nf_generic_pernet(nf_ct_net(ct
))->timeout
;
1619 nf_ct_refresh_acct(ct
, ctinfo
, skb
, *timeout
);
1623 /* Returns verdict for packet, or -1 for invalid. */
1624 static int nf_conntrack_handle_packet(struct nf_conn
*ct
,
1625 struct sk_buff
*skb
,
1626 unsigned int dataoff
,
1627 enum ip_conntrack_info ctinfo
,
1628 const struct nf_hook_state
*state
)
1630 switch (nf_ct_protonum(ct
)) {
1632 return nf_conntrack_tcp_packet(ct
, skb
, dataoff
,
1635 return nf_conntrack_udp_packet(ct
, skb
, dataoff
,
1638 return nf_conntrack_icmp_packet(ct
, skb
, ctinfo
, state
);
1639 #if IS_ENABLED(CONFIG_IPV6)
1640 case IPPROTO_ICMPV6
:
1641 return nf_conntrack_icmpv6_packet(ct
, skb
, ctinfo
, state
);
1643 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
1644 case IPPROTO_UDPLITE
:
1645 return nf_conntrack_udplite_packet(ct
, skb
, dataoff
,
1648 #ifdef CONFIG_NF_CT_PROTO_SCTP
1650 return nf_conntrack_sctp_packet(ct
, skb
, dataoff
,
1653 #ifdef CONFIG_NF_CT_PROTO_DCCP
1655 return nf_conntrack_dccp_packet(ct
, skb
, dataoff
,
1658 #ifdef CONFIG_NF_CT_PROTO_GRE
1660 return nf_conntrack_gre_packet(ct
, skb
, dataoff
,
1665 return generic_packet(ct
, skb
, ctinfo
);
1669 nf_conntrack_in(struct sk_buff
*skb
, const struct nf_hook_state
*state
)
1671 enum ip_conntrack_info ctinfo
;
1672 struct nf_conn
*ct
, *tmpl
;
1676 tmpl
= nf_ct_get(skb
, &ctinfo
);
1677 if (tmpl
|| ctinfo
== IP_CT_UNTRACKED
) {
1678 /* Previously seen (loopback or untracked)? Ignore. */
1679 if ((tmpl
&& !nf_ct_is_template(tmpl
)) ||
1680 ctinfo
== IP_CT_UNTRACKED
) {
1681 NF_CT_STAT_INC_ATOMIC(state
->net
, ignore
);
1687 /* rcu_read_lock()ed by nf_hook_thresh */
1688 dataoff
= get_l4proto(skb
, skb_network_offset(skb
), state
->pf
, &protonum
);
1690 pr_debug("not prepared to track yet or error occurred\n");
1691 NF_CT_STAT_INC_ATOMIC(state
->net
, error
);
1692 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1697 if (protonum
== IPPROTO_ICMP
|| protonum
== IPPROTO_ICMPV6
) {
1698 ret
= nf_conntrack_handle_icmp(tmpl
, skb
, dataoff
,
1704 /* ICMP[v6] protocol trackers may assign one conntrack. */
1709 ret
= resolve_normal_ct(tmpl
, skb
, dataoff
,
1712 /* Too stressed to deal. */
1713 NF_CT_STAT_INC_ATOMIC(state
->net
, drop
);
1718 ct
= nf_ct_get(skb
, &ctinfo
);
1720 /* Not valid part of a connection */
1721 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1726 ret
= nf_conntrack_handle_packet(ct
, skb
, dataoff
, ctinfo
, state
);
1728 /* Invalid: inverse of the return code tells
1729 * the netfilter core what to do */
1730 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1731 nf_conntrack_put(&ct
->ct_general
);
1733 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1734 if (ret
== -NF_DROP
)
1735 NF_CT_STAT_INC_ATOMIC(state
->net
, drop
);
1736 /* Special case: TCP tracker reports an attempt to reopen a
1737 * closed/aborted connection. We have to go back and create a
1740 if (ret
== -NF_REPEAT
)
1746 if (ctinfo
== IP_CT_ESTABLISHED_REPLY
&&
1747 !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1748 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1755 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1757 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1758 implicitly racy: see __nf_conntrack_confirm */
1759 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1760 const struct nf_conntrack_tuple
*newreply
)
1762 struct nf_conn_help
*help
= nfct_help(ct
);
1764 /* Should be unconfirmed, so not in hash table yet */
1765 WARN_ON(nf_ct_is_confirmed(ct
));
1767 pr_debug("Altering reply tuple of %p to ", ct
);
1768 nf_ct_dump_tuple(newreply
);
1770 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1771 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1775 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1778 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1780 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1781 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1782 enum ip_conntrack_info ctinfo
,
1783 const struct sk_buff
*skb
,
1787 /* Only update if this is not a fixed timeout */
1788 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1791 /* If not in hash table, timer will not be active yet */
1792 if (nf_ct_is_confirmed(ct
))
1793 extra_jiffies
+= nfct_time_stamp
;
1795 if (ct
->timeout
!= extra_jiffies
)
1796 ct
->timeout
= extra_jiffies
;
1799 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1801 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1803 bool nf_ct_kill_acct(struct nf_conn
*ct
,
1804 enum ip_conntrack_info ctinfo
,
1805 const struct sk_buff
*skb
)
1807 nf_ct_acct_update(ct
, ctinfo
, skb
->len
);
1809 return nf_ct_delete(ct
, 0, 0);
1811 EXPORT_SYMBOL_GPL(nf_ct_kill_acct
);
1813 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1815 #include <linux/netfilter/nfnetlink.h>
1816 #include <linux/netfilter/nfnetlink_conntrack.h>
1817 #include <linux/mutex.h>
1819 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1820 * in ip_conntrack_core, since we don't want the protocols to autoload
1821 * or depend on ctnetlink */
1822 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1823 const struct nf_conntrack_tuple
*tuple
)
1825 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1826 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1827 goto nla_put_failure
;
1833 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1835 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1836 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1837 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1839 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1841 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1842 struct nf_conntrack_tuple
*t
)
1844 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1847 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1848 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1852 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1854 unsigned int nf_ct_port_nlattr_tuple_size(void)
1856 static unsigned int size __read_mostly
;
1859 size
= nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1863 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
1866 /* Used by ipt_REJECT and ip6t_REJECT. */
1867 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
1870 enum ip_conntrack_info ctinfo
;
1872 /* This ICMP is in reverse direction to the packet which caused it */
1873 ct
= nf_ct_get(skb
, &ctinfo
);
1874 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
1875 ctinfo
= IP_CT_RELATED_REPLY
;
1877 ctinfo
= IP_CT_RELATED
;
1879 /* Attach to new skbuff, and increment count */
1880 nf_ct_set(nskb
, ct
, ctinfo
);
1881 nf_conntrack_get(skb_nfct(nskb
));
1884 static int nf_conntrack_update(struct net
*net
, struct sk_buff
*skb
)
1886 struct nf_conntrack_tuple_hash
*h
;
1887 struct nf_conntrack_tuple tuple
;
1888 enum ip_conntrack_info ctinfo
;
1889 struct nf_nat_hook
*nat_hook
;
1890 unsigned int status
;
1896 ct
= nf_ct_get(skb
, &ctinfo
);
1897 if (!ct
|| nf_ct_is_confirmed(ct
))
1900 l3num
= nf_ct_l3num(ct
);
1902 dataoff
= get_l4proto(skb
, skb_network_offset(skb
), l3num
, &l4num
);
1906 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
), dataoff
, l3num
,
1907 l4num
, net
, &tuple
))
1910 if (ct
->status
& IPS_SRC_NAT
) {
1911 memcpy(tuple
.src
.u3
.all
,
1912 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.u3
.all
,
1913 sizeof(tuple
.src
.u3
.all
));
1915 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.u
.all
;
1918 if (ct
->status
& IPS_DST_NAT
) {
1919 memcpy(tuple
.dst
.u3
.all
,
1920 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.u3
.all
,
1921 sizeof(tuple
.dst
.u3
.all
));
1923 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.u
.all
;
1926 h
= nf_conntrack_find_get(net
, nf_ct_zone(ct
), &tuple
);
1930 /* Store status bits of the conntrack that is clashing to re-do NAT
1931 * mangling according to what it has been done already to this packet.
1933 status
= ct
->status
;
1936 ct
= nf_ct_tuplehash_to_ctrack(h
);
1937 nf_ct_set(skb
, ct
, ctinfo
);
1939 nat_hook
= rcu_dereference(nf_nat_hook
);
1943 if (status
& IPS_SRC_NAT
&&
1944 nat_hook
->manip_pkt(skb
, ct
, NF_NAT_MANIP_SRC
,
1945 IP_CT_DIR_ORIGINAL
) == NF_DROP
)
1948 if (status
& IPS_DST_NAT
&&
1949 nat_hook
->manip_pkt(skb
, ct
, NF_NAT_MANIP_DST
,
1950 IP_CT_DIR_ORIGINAL
) == NF_DROP
)
1956 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple
*dst_tuple
,
1957 const struct sk_buff
*skb
)
1959 const struct nf_conntrack_tuple
*src_tuple
;
1960 const struct nf_conntrack_tuple_hash
*hash
;
1961 struct nf_conntrack_tuple srctuple
;
1962 enum ip_conntrack_info ctinfo
;
1965 ct
= nf_ct_get(skb
, &ctinfo
);
1967 src_tuple
= nf_ct_tuple(ct
, CTINFO2DIR(ctinfo
));
1968 memcpy(dst_tuple
, src_tuple
, sizeof(*dst_tuple
));
1972 if (!nf_ct_get_tuplepr(skb
, skb_network_offset(skb
),
1973 NFPROTO_IPV4
, dev_net(skb
->dev
),
1977 hash
= nf_conntrack_find_get(dev_net(skb
->dev
),
1983 ct
= nf_ct_tuplehash_to_ctrack(hash
);
1984 src_tuple
= nf_ct_tuple(ct
, !hash
->tuple
.dst
.dir
);
1985 memcpy(dst_tuple
, src_tuple
, sizeof(*dst_tuple
));
1991 /* Bring out ya dead! */
1992 static struct nf_conn
*
1993 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
1994 void *data
, unsigned int *bucket
)
1996 struct nf_conntrack_tuple_hash
*h
;
1998 struct hlist_nulls_node
*n
;
2001 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
2002 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
2004 nf_conntrack_lock(lockp
);
2005 if (*bucket
< nf_conntrack_htable_size
) {
2006 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
2007 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
2009 ct
= nf_ct_tuplehash_to_ctrack(h
);
2021 atomic_inc(&ct
->ct_general
.use
);
2027 static void nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
),
2028 void *data
, u32 portid
, int report
)
2030 unsigned int bucket
= 0, sequence
;
2036 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
2038 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
2039 /* Time to push up daises... */
2041 nf_ct_delete(ct
, portid
, report
);
2046 if (!read_seqcount_retry(&nf_conntrack_generation
, sequence
))
2053 int (*iter
)(struct nf_conn
*i
, void *data
);
2058 static int iter_net_only(struct nf_conn
*i
, void *data
)
2060 struct iter_data
*d
= data
;
2062 if (!net_eq(d
->net
, nf_ct_net(i
)))
2065 return d
->iter(i
, d
->data
);
2069 __nf_ct_unconfirmed_destroy(struct net
*net
)
2073 for_each_possible_cpu(cpu
) {
2074 struct nf_conntrack_tuple_hash
*h
;
2075 struct hlist_nulls_node
*n
;
2076 struct ct_pcpu
*pcpu
;
2078 pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2080 spin_lock_bh(&pcpu
->lock
);
2081 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
2084 ct
= nf_ct_tuplehash_to_ctrack(h
);
2086 /* we cannot call iter() on unconfirmed list, the
2087 * owning cpu can reallocate ct->ext at any time.
2089 set_bit(IPS_DYING_BIT
, &ct
->status
);
2091 spin_unlock_bh(&pcpu
->lock
);
2096 void nf_ct_unconfirmed_destroy(struct net
*net
)
2100 if (atomic_read(&net
->ct
.count
) > 0) {
2101 __nf_ct_unconfirmed_destroy(net
);
2102 nf_queue_nf_hook_drop(net
);
2106 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy
);
2108 void nf_ct_iterate_cleanup_net(struct net
*net
,
2109 int (*iter
)(struct nf_conn
*i
, void *data
),
2110 void *data
, u32 portid
, int report
)
2116 if (atomic_read(&net
->ct
.count
) == 0)
2123 nf_ct_iterate_cleanup(iter_net_only
, &d
, portid
, report
);
2125 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net
);
2128 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2129 * @iter: callback to invoke for each conntrack
2130 * @data: data to pass to @iter
2132 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2133 * unconfirmed list as dying (so they will not be inserted into
2136 * Can only be called in module exit path.
2139 nf_ct_iterate_destroy(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
2143 down_read(&net_rwsem
);
2145 if (atomic_read(&net
->ct
.count
) == 0)
2147 __nf_ct_unconfirmed_destroy(net
);
2148 nf_queue_nf_hook_drop(net
);
2150 up_read(&net_rwsem
);
2152 /* Need to wait for netns cleanup worker to finish, if its
2153 * running -- it might have deleted a net namespace from
2154 * the global list, so our __nf_ct_unconfirmed_destroy() might
2155 * not have affected all namespaces.
2159 /* a conntrack could have been unlinked from unconfirmed list
2160 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2161 * This makes sure its inserted into conntrack table.
2165 nf_ct_iterate_cleanup(iter
, data
, 0, 0);
2167 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy
);
2169 static int kill_all(struct nf_conn
*i
, void *data
)
2171 return net_eq(nf_ct_net(i
), data
);
2174 void nf_conntrack_cleanup_start(void)
2176 conntrack_gc_work
.exiting
= true;
2177 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
2180 void nf_conntrack_cleanup_end(void)
2182 RCU_INIT_POINTER(nf_ct_hook
, NULL
);
2183 cancel_delayed_work_sync(&conntrack_gc_work
.dwork
);
2184 kvfree(nf_conntrack_hash
);
2186 nf_conntrack_proto_fini();
2187 nf_conntrack_seqadj_fini();
2188 nf_conntrack_labels_fini();
2189 nf_conntrack_helper_fini();
2190 nf_conntrack_timeout_fini();
2191 nf_conntrack_ecache_fini();
2192 nf_conntrack_tstamp_fini();
2193 nf_conntrack_acct_fini();
2194 nf_conntrack_expect_fini();
2196 kmem_cache_destroy(nf_conntrack_cachep
);
2200 * Mishearing the voices in his head, our hero wonders how he's
2201 * supposed to kill the mall.
2203 void nf_conntrack_cleanup_net(struct net
*net
)
2207 list_add(&net
->exit_list
, &single
);
2208 nf_conntrack_cleanup_net_list(&single
);
2211 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
2217 * This makes sure all current packets have passed through
2218 * netfilter framework. Roll on, two-stage module
2224 list_for_each_entry(net
, net_exit_list
, exit_list
) {
2225 nf_ct_iterate_cleanup(kill_all
, net
, 0, 0);
2226 if (atomic_read(&net
->ct
.count
) != 0)
2231 goto i_see_dead_people
;
2234 list_for_each_entry(net
, net_exit_list
, exit_list
) {
2235 nf_conntrack_proto_pernet_fini(net
);
2236 nf_conntrack_ecache_pernet_fini(net
);
2237 nf_conntrack_expect_pernet_fini(net
);
2238 free_percpu(net
->ct
.stat
);
2239 free_percpu(net
->ct
.pcpu_lists
);
2243 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
2245 struct hlist_nulls_head
*hash
;
2246 unsigned int nr_slots
, i
;
2248 if (*sizep
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
2251 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
2252 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
2254 hash
= kvmalloc_array(nr_slots
, sizeof(struct hlist_nulls_head
),
2255 GFP_KERNEL
| __GFP_ZERO
);
2258 for (i
= 0; i
< nr_slots
; i
++)
2259 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
2263 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
2265 int nf_conntrack_hash_resize(unsigned int hashsize
)
2268 unsigned int old_size
;
2269 struct hlist_nulls_head
*hash
, *old_hash
;
2270 struct nf_conntrack_tuple_hash
*h
;
2276 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
2280 old_size
= nf_conntrack_htable_size
;
2281 if (old_size
== hashsize
) {
2287 nf_conntrack_all_lock();
2288 write_seqcount_begin(&nf_conntrack_generation
);
2290 /* Lookups in the old hash might happen in parallel, which means we
2291 * might get false negatives during connection lookup. New connections
2292 * created because of a false negative won't make it into the hash
2293 * though since that required taking the locks.
2296 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
2297 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
2298 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
2299 struct nf_conntrack_tuple_hash
, hnnode
);
2300 ct
= nf_ct_tuplehash_to_ctrack(h
);
2301 hlist_nulls_del_rcu(&h
->hnnode
);
2302 bucket
= __hash_conntrack(nf_ct_net(ct
),
2303 &h
->tuple
, hashsize
);
2304 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
2307 old_size
= nf_conntrack_htable_size
;
2308 old_hash
= nf_conntrack_hash
;
2310 nf_conntrack_hash
= hash
;
2311 nf_conntrack_htable_size
= hashsize
;
2313 write_seqcount_end(&nf_conntrack_generation
);
2314 nf_conntrack_all_unlock();
2322 int nf_conntrack_set_hashsize(const char *val
, const struct kernel_param
*kp
)
2324 unsigned int hashsize
;
2327 if (current
->nsproxy
->net_ns
!= &init_net
)
2330 /* On boot, we can set this without any fancy locking. */
2331 if (!nf_conntrack_hash
)
2332 return param_set_uint(val
, kp
);
2334 rc
= kstrtouint(val
, 0, &hashsize
);
2338 return nf_conntrack_hash_resize(hashsize
);
2340 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize
);
2342 static __always_inline
unsigned int total_extension_size(void)
2344 /* remember to add new extensions below */
2345 BUILD_BUG_ON(NF_CT_EXT_NUM
> 9);
2347 return sizeof(struct nf_ct_ext
) +
2348 sizeof(struct nf_conn_help
)
2349 #if IS_ENABLED(CONFIG_NF_NAT)
2350 + sizeof(struct nf_conn_nat
)
2352 + sizeof(struct nf_conn_seqadj
)
2353 + sizeof(struct nf_conn_acct
)
2354 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2355 + sizeof(struct nf_conntrack_ecache
)
2357 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2358 + sizeof(struct nf_conn_tstamp
)
2360 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2361 + sizeof(struct nf_conn_timeout
)
2363 #ifdef CONFIG_NF_CONNTRACK_LABELS
2364 + sizeof(struct nf_conn_labels
)
2366 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2367 + sizeof(struct nf_conn_synproxy
)
2372 int nf_conntrack_init_start(void)
2374 unsigned long nr_pages
= totalram_pages();
2379 /* struct nf_ct_ext uses u8 to store offsets/size */
2380 BUILD_BUG_ON(total_extension_size() > 255u);
2382 seqcount_init(&nf_conntrack_generation
);
2384 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
2385 spin_lock_init(&nf_conntrack_locks
[i
]);
2387 if (!nf_conntrack_htable_size
) {
2388 /* Idea from tcp.c: use 1/16384 of memory.
2389 * On i386: 32MB machine has 512 buckets.
2390 * >= 1GB machines have 16384 buckets.
2391 * >= 4GB machines have 65536 buckets.
2393 nf_conntrack_htable_size
2394 = (((nr_pages
<< PAGE_SHIFT
) / 16384)
2395 / sizeof(struct hlist_head
));
2396 if (nr_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
2397 nf_conntrack_htable_size
= 65536;
2398 else if (nr_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
2399 nf_conntrack_htable_size
= 16384;
2400 if (nf_conntrack_htable_size
< 32)
2401 nf_conntrack_htable_size
= 32;
2403 /* Use a max. factor of four by default to get the same max as
2404 * with the old struct list_heads. When a table size is given
2405 * we use the old value of 8 to avoid reducing the max.
2410 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
2411 if (!nf_conntrack_hash
)
2414 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
2416 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
2417 sizeof(struct nf_conn
),
2419 SLAB_TYPESAFE_BY_RCU
| SLAB_HWCACHE_ALIGN
, NULL
);
2420 if (!nf_conntrack_cachep
)
2423 ret
= nf_conntrack_expect_init();
2427 ret
= nf_conntrack_acct_init();
2431 ret
= nf_conntrack_tstamp_init();
2435 ret
= nf_conntrack_ecache_init();
2439 ret
= nf_conntrack_timeout_init();
2443 ret
= nf_conntrack_helper_init();
2447 ret
= nf_conntrack_labels_init();
2451 ret
= nf_conntrack_seqadj_init();
2455 ret
= nf_conntrack_proto_init();
2459 conntrack_gc_work_init(&conntrack_gc_work
);
2460 queue_delayed_work(system_power_efficient_wq
, &conntrack_gc_work
.dwork
, HZ
);
2465 nf_conntrack_seqadj_fini();
2467 nf_conntrack_labels_fini();
2469 nf_conntrack_helper_fini();
2471 nf_conntrack_timeout_fini();
2473 nf_conntrack_ecache_fini();
2475 nf_conntrack_tstamp_fini();
2477 nf_conntrack_acct_fini();
2479 nf_conntrack_expect_fini();
2481 kmem_cache_destroy(nf_conntrack_cachep
);
2483 kvfree(nf_conntrack_hash
);
2487 static struct nf_ct_hook nf_conntrack_hook
= {
2488 .update
= nf_conntrack_update
,
2489 .destroy
= destroy_conntrack
,
2490 .get_tuple_skb
= nf_conntrack_get_tuple_skb
,
2493 void nf_conntrack_init_end(void)
2495 /* For use by REJECT target */
2496 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
2497 RCU_INIT_POINTER(nf_ct_hook
, &nf_conntrack_hook
);
2501 * We need to use special "null" values, not used in hash table
2503 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2504 #define DYING_NULLS_VAL ((1<<30)+1)
2505 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
2507 int nf_conntrack_init_net(struct net
*net
)
2512 BUILD_BUG_ON(IP_CT_UNTRACKED
== IP_CT_NUMBER
);
2513 BUILD_BUG_ON_NOT_POWER_OF_2(CONNTRACK_LOCKS
);
2514 atomic_set(&net
->ct
.count
, 0);
2516 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
2517 if (!net
->ct
.pcpu_lists
)
2520 for_each_possible_cpu(cpu
) {
2521 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2523 spin_lock_init(&pcpu
->lock
);
2524 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
2525 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
2528 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
2530 goto err_pcpu_lists
;
2532 ret
= nf_conntrack_expect_pernet_init(net
);
2536 nf_conntrack_acct_pernet_init(net
);
2537 nf_conntrack_tstamp_pernet_init(net
);
2538 nf_conntrack_ecache_pernet_init(net
);
2539 nf_conntrack_helper_pernet_init(net
);
2540 nf_conntrack_proto_pernet_init(net
);
2545 free_percpu(net
->ct
.stat
);
2547 free_percpu(net
->ct
.pcpu_lists
);