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 DEFINE_SPINLOCK(nf_conntrack_locks_all_lock
);
77 static __read_mostly
bool nf_conntrack_locks_all
;
79 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
80 #define GC_MAX_BUCKETS_DIV 128u
81 /* upper bound of full table scan */
82 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
83 /* desired ratio of entries found to be expired */
84 #define GC_EVICT_RATIO 50u
86 static struct conntrack_gc_work conntrack_gc_work
;
88 void nf_conntrack_lock(spinlock_t
*lock
) __acquires(lock
)
90 /* 1) Acquire the lock */
93 /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
94 * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
96 if (likely(smp_load_acquire(&nf_conntrack_locks_all
) == false))
99 /* fast path failed, unlock */
102 /* Slow path 1) get global lock */
103 spin_lock(&nf_conntrack_locks_all_lock
);
105 /* Slow path 2) get the lock we want */
108 /* Slow path 3) release the global lock */
109 spin_unlock(&nf_conntrack_locks_all_lock
);
111 EXPORT_SYMBOL_GPL(nf_conntrack_lock
);
113 static void nf_conntrack_double_unlock(unsigned int h1
, unsigned int h2
)
115 h1
%= CONNTRACK_LOCKS
;
116 h2
%= CONNTRACK_LOCKS
;
117 spin_unlock(&nf_conntrack_locks
[h1
]);
119 spin_unlock(&nf_conntrack_locks
[h2
]);
122 /* return true if we need to recompute hashes (in case hash table was resized) */
123 static bool nf_conntrack_double_lock(struct net
*net
, unsigned int h1
,
124 unsigned int h2
, unsigned int sequence
)
126 h1
%= CONNTRACK_LOCKS
;
127 h2
%= CONNTRACK_LOCKS
;
129 nf_conntrack_lock(&nf_conntrack_locks
[h1
]);
131 spin_lock_nested(&nf_conntrack_locks
[h2
],
132 SINGLE_DEPTH_NESTING
);
134 nf_conntrack_lock(&nf_conntrack_locks
[h2
]);
135 spin_lock_nested(&nf_conntrack_locks
[h1
],
136 SINGLE_DEPTH_NESTING
);
138 if (read_seqcount_retry(&nf_conntrack_generation
, sequence
)) {
139 nf_conntrack_double_unlock(h1
, h2
);
145 static void nf_conntrack_all_lock(void)
146 __acquires(&nf_conntrack_locks_all_lock
)
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)
166 __releases(&nf_conntrack_locks_all_lock
)
168 /* All prior stores must be complete before we clear
169 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
170 * might observe the false value but not the entire
172 * It pairs with the smp_load_acquire() in nf_conntrack_lock()
174 smp_store_release(&nf_conntrack_locks_all
, false);
175 spin_unlock(&nf_conntrack_locks_all_lock
);
178 unsigned int nf_conntrack_htable_size __read_mostly
;
179 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size
);
181 unsigned int nf_conntrack_max __read_mostly
;
182 EXPORT_SYMBOL_GPL(nf_conntrack_max
);
183 seqcount_t nf_conntrack_generation __read_mostly
;
184 static unsigned int nf_conntrack_hash_rnd __read_mostly
;
186 static u32
hash_conntrack_raw(const struct nf_conntrack_tuple
*tuple
,
187 const struct net
*net
)
192 get_random_once(&nf_conntrack_hash_rnd
, sizeof(nf_conntrack_hash_rnd
));
194 /* The direction must be ignored, so we hash everything up to the
195 * destination ports (which is a multiple of 4) and treat the last
196 * three bytes manually.
198 seed
= nf_conntrack_hash_rnd
^ net_hash_mix(net
);
199 n
= (sizeof(tuple
->src
) + sizeof(tuple
->dst
.u3
)) / sizeof(u32
);
200 return jhash2((u32
*)tuple
, n
, seed
^
201 (((__force __u16
)tuple
->dst
.u
.all
<< 16) |
202 tuple
->dst
.protonum
));
205 static u32
scale_hash(u32 hash
)
207 return reciprocal_scale(hash
, nf_conntrack_htable_size
);
210 static u32
__hash_conntrack(const struct net
*net
,
211 const struct nf_conntrack_tuple
*tuple
,
214 return reciprocal_scale(hash_conntrack_raw(tuple
, net
), size
);
217 static u32
hash_conntrack(const struct net
*net
,
218 const struct nf_conntrack_tuple
*tuple
)
220 return scale_hash(hash_conntrack_raw(tuple
, net
));
223 static bool nf_ct_get_tuple_ports(const struct sk_buff
*skb
,
224 unsigned int dataoff
,
225 struct nf_conntrack_tuple
*tuple
)
229 } _inet_hdr
, *inet_hdr
;
231 /* Actually only need first 4 bytes to get ports. */
232 inet_hdr
= skb_header_pointer(skb
, dataoff
, sizeof(_inet_hdr
), &_inet_hdr
);
236 tuple
->src
.u
.udp
.port
= inet_hdr
->sport
;
237 tuple
->dst
.u
.udp
.port
= inet_hdr
->dport
;
242 nf_ct_get_tuple(const struct sk_buff
*skb
,
244 unsigned int dataoff
,
248 struct nf_conntrack_tuple
*tuple
)
254 memset(tuple
, 0, sizeof(*tuple
));
256 tuple
->src
.l3num
= l3num
;
259 nhoff
+= offsetof(struct iphdr
, saddr
);
260 size
= 2 * sizeof(__be32
);
263 nhoff
+= offsetof(struct ipv6hdr
, saddr
);
264 size
= sizeof(_addrs
);
270 ap
= skb_header_pointer(skb
, nhoff
, size
, _addrs
);
276 tuple
->src
.u3
.ip
= ap
[0];
277 tuple
->dst
.u3
.ip
= ap
[1];
280 memcpy(tuple
->src
.u3
.ip6
, ap
, sizeof(tuple
->src
.u3
.ip6
));
281 memcpy(tuple
->dst
.u3
.ip6
, ap
+ 4, sizeof(tuple
->dst
.u3
.ip6
));
285 tuple
->dst
.protonum
= protonum
;
286 tuple
->dst
.dir
= IP_CT_DIR_ORIGINAL
;
289 #if IS_ENABLED(CONFIG_IPV6)
291 return icmpv6_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
294 return icmp_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
295 #ifdef CONFIG_NF_CT_PROTO_GRE
297 return gre_pkt_to_tuple(skb
, dataoff
, net
, tuple
);
300 case IPPROTO_UDP
: /* fallthrough */
301 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
302 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
303 case IPPROTO_UDPLITE
:
304 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
306 #ifdef CONFIG_NF_CT_PROTO_SCTP
308 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
310 #ifdef CONFIG_NF_CT_PROTO_DCCP
312 return nf_ct_get_tuple_ports(skb
, dataoff
, tuple
);
321 static int ipv4_get_l4proto(const struct sk_buff
*skb
, unsigned int nhoff
,
325 const struct iphdr
*iph
;
328 iph
= skb_header_pointer(skb
, nhoff
, sizeof(_iph
), &_iph
);
332 /* Conntrack defragments packets, we might still see fragments
333 * inside ICMP packets though.
335 if (iph
->frag_off
& htons(IP_OFFSET
))
338 dataoff
= nhoff
+ (iph
->ihl
<< 2);
339 *protonum
= iph
->protocol
;
341 /* Check bogus IP headers */
342 if (dataoff
> skb
->len
) {
343 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
344 nhoff
, iph
->ihl
<< 2, skb
->len
);
350 #if IS_ENABLED(CONFIG_IPV6)
351 static int ipv6_get_l4proto(const struct sk_buff
*skb
, unsigned int nhoff
,
355 unsigned int extoff
= nhoff
+ sizeof(struct ipv6hdr
);
359 if (skb_copy_bits(skb
, nhoff
+ offsetof(struct ipv6hdr
, nexthdr
),
360 &nexthdr
, sizeof(nexthdr
)) != 0) {
361 pr_debug("can't get nexthdr\n");
364 protoff
= ipv6_skip_exthdr(skb
, extoff
, &nexthdr
, &frag_off
);
366 * (protoff == skb->len) means the packet has not data, just
367 * IPv6 and possibly extensions headers, but it is tracked anyway
369 if (protoff
< 0 || (frag_off
& htons(~0x7)) != 0) {
370 pr_debug("can't find proto in pkt\n");
379 static int get_l4proto(const struct sk_buff
*skb
,
380 unsigned int nhoff
, u8 pf
, u8
*l4num
)
384 return ipv4_get_l4proto(skb
, nhoff
, l4num
);
385 #if IS_ENABLED(CONFIG_IPV6)
387 return ipv6_get_l4proto(skb
, nhoff
, l4num
);
396 bool nf_ct_get_tuplepr(const struct sk_buff
*skb
, unsigned int nhoff
,
398 struct net
*net
, struct nf_conntrack_tuple
*tuple
)
403 protoff
= get_l4proto(skb
, nhoff
, l3num
, &protonum
);
407 return nf_ct_get_tuple(skb
, nhoff
, protoff
, l3num
, protonum
, net
, tuple
);
409 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr
);
412 nf_ct_invert_tuple(struct nf_conntrack_tuple
*inverse
,
413 const struct nf_conntrack_tuple
*orig
)
415 memset(inverse
, 0, sizeof(*inverse
));
417 inverse
->src
.l3num
= orig
->src
.l3num
;
419 switch (orig
->src
.l3num
) {
421 inverse
->src
.u3
.ip
= orig
->dst
.u3
.ip
;
422 inverse
->dst
.u3
.ip
= orig
->src
.u3
.ip
;
425 inverse
->src
.u3
.in6
= orig
->dst
.u3
.in6
;
426 inverse
->dst
.u3
.in6
= orig
->src
.u3
.in6
;
432 inverse
->dst
.dir
= !orig
->dst
.dir
;
434 inverse
->dst
.protonum
= orig
->dst
.protonum
;
436 switch (orig
->dst
.protonum
) {
438 return nf_conntrack_invert_icmp_tuple(inverse
, orig
);
439 #if IS_ENABLED(CONFIG_IPV6)
441 return nf_conntrack_invert_icmpv6_tuple(inverse
, orig
);
445 inverse
->src
.u
.all
= orig
->dst
.u
.all
;
446 inverse
->dst
.u
.all
= orig
->src
.u
.all
;
449 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple
);
451 /* Generate a almost-unique pseudo-id for a given conntrack.
453 * intentionally doesn't re-use any of the seeds used for hash
454 * table location, we assume id gets exposed to userspace.
456 * Following nf_conn items do not change throughout lifetime
460 * 2. nf_conn->master address (normally NULL)
461 * 3. the associated net namespace
462 * 4. the original direction tuple
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
;
473 c
= (unsigned long)nf_ct_net(ct
);
474 d
= (unsigned long)siphash(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
475 sizeof(ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
),
478 return siphash_4u64((u64
)a
, (u64
)b
, (u64
)c
, (u64
)d
, &ct_id_seed
);
480 return siphash_4u32((u32
)a
, (u32
)b
, (u32
)c
, (u32
)d
, &ct_id_seed
);
483 EXPORT_SYMBOL_GPL(nf_ct_get_id
);
486 clean_from_lists(struct nf_conn
*ct
)
488 pr_debug("clean_from_lists(%p)\n", ct
);
489 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
490 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
);
492 /* Destroy all pending expectations */
493 nf_ct_remove_expectations(ct
);
496 /* must be called with local_bh_disable */
497 static void nf_ct_add_to_dying_list(struct nf_conn
*ct
)
499 struct ct_pcpu
*pcpu
;
501 /* add this conntrack to the (per cpu) dying list */
502 ct
->cpu
= smp_processor_id();
503 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
505 spin_lock(&pcpu
->lock
);
506 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
508 spin_unlock(&pcpu
->lock
);
511 /* must be called with local_bh_disable */
512 static void nf_ct_add_to_unconfirmed_list(struct nf_conn
*ct
)
514 struct ct_pcpu
*pcpu
;
516 /* add this conntrack to the (per cpu) unconfirmed list */
517 ct
->cpu
= smp_processor_id();
518 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
520 spin_lock(&pcpu
->lock
);
521 hlist_nulls_add_head(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
523 spin_unlock(&pcpu
->lock
);
526 /* must be called with local_bh_disable */
527 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn
*ct
)
529 struct ct_pcpu
*pcpu
;
531 /* We overload first tuple to link into unconfirmed or dying list.*/
532 pcpu
= per_cpu_ptr(nf_ct_net(ct
)->ct
.pcpu_lists
, ct
->cpu
);
534 spin_lock(&pcpu
->lock
);
535 BUG_ON(hlist_nulls_unhashed(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
));
536 hlist_nulls_del_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
537 spin_unlock(&pcpu
->lock
);
540 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
542 /* Released via destroy_conntrack() */
543 struct nf_conn
*nf_ct_tmpl_alloc(struct net
*net
,
544 const struct nf_conntrack_zone
*zone
,
547 struct nf_conn
*tmpl
, *p
;
549 if (ARCH_KMALLOC_MINALIGN
<= NFCT_INFOMASK
) {
550 tmpl
= kzalloc(sizeof(*tmpl
) + NFCT_INFOMASK
, flags
);
555 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
557 tmpl
= (struct nf_conn
*)NFCT_ALIGN((unsigned long)p
);
558 tmpl
->proto
.tmpl_padto
= (char *)tmpl
- (char *)p
;
561 tmpl
= kzalloc(sizeof(*tmpl
), flags
);
566 tmpl
->status
= IPS_TEMPLATE
;
567 write_pnet(&tmpl
->ct_net
, net
);
568 nf_ct_zone_add(tmpl
, zone
);
569 atomic_set(&tmpl
->ct_general
.use
, 0);
573 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc
);
575 void nf_ct_tmpl_free(struct nf_conn
*tmpl
)
577 nf_ct_ext_destroy(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_key_equal(h
, tuple
, zone
, net
))
756 * if the nulls value we got at the end of this lookup is
757 * not the expected one, we must restart lookup.
758 * We probably met an item that was moved to another chain.
760 if (get_nulls_value(n
) != bucket
) {
761 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
768 /* Find a connection corresponding to a tuple. */
769 static struct nf_conntrack_tuple_hash
*
770 __nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
771 const struct nf_conntrack_tuple
*tuple
, u32 hash
)
773 struct nf_conntrack_tuple_hash
*h
;
778 h
= ____nf_conntrack_find(net
, zone
, tuple
, hash
);
780 /* We have a candidate that matches the tuple we're interested
781 * in, try to obtain a reference and re-check tuple
783 ct
= nf_ct_tuplehash_to_ctrack(h
);
784 if (likely(atomic_inc_not_zero(&ct
->ct_general
.use
))) {
785 if (likely(nf_ct_key_equal(h
, tuple
, zone
, net
)))
788 /* TYPESAFE_BY_RCU recycled the candidate */
800 struct nf_conntrack_tuple_hash
*
801 nf_conntrack_find_get(struct net
*net
, const struct nf_conntrack_zone
*zone
,
802 const struct nf_conntrack_tuple
*tuple
)
804 return __nf_conntrack_find_get(net
, zone
, tuple
,
805 hash_conntrack_raw(tuple
, net
));
807 EXPORT_SYMBOL_GPL(nf_conntrack_find_get
);
809 static void __nf_conntrack_hash_insert(struct nf_conn
*ct
,
811 unsigned int reply_hash
)
813 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
,
814 &nf_conntrack_hash
[hash
]);
815 hlist_nulls_add_head_rcu(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
816 &nf_conntrack_hash
[reply_hash
]);
820 nf_conntrack_hash_check_insert(struct nf_conn
*ct
)
822 const struct nf_conntrack_zone
*zone
;
823 struct net
*net
= nf_ct_net(ct
);
824 unsigned int hash
, reply_hash
;
825 struct nf_conntrack_tuple_hash
*h
;
826 struct hlist_nulls_node
*n
;
827 unsigned int sequence
;
829 zone
= nf_ct_zone(ct
);
833 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
834 hash
= hash_conntrack(net
,
835 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
836 reply_hash
= hash_conntrack(net
,
837 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
838 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
840 /* See if there's one in the list already, including reverse */
841 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
842 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
846 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
847 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
852 /* The caller holds a reference to this object */
853 atomic_set(&ct
->ct_general
.use
, 2);
854 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
855 nf_conntrack_double_unlock(hash
, reply_hash
);
856 NF_CT_STAT_INC(net
, insert
);
861 nf_conntrack_double_unlock(hash
, reply_hash
);
862 NF_CT_STAT_INC(net
, insert_failed
);
866 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert
);
868 void nf_ct_acct_add(struct nf_conn
*ct
, u32 dir
, unsigned int packets
,
871 struct nf_conn_acct
*acct
;
873 acct
= nf_conn_acct_find(ct
);
875 struct nf_conn_counter
*counter
= acct
->counter
;
877 atomic64_add(packets
, &counter
[dir
].packets
);
878 atomic64_add(bytes
, &counter
[dir
].bytes
);
881 EXPORT_SYMBOL_GPL(nf_ct_acct_add
);
883 static void nf_ct_acct_merge(struct nf_conn
*ct
, enum ip_conntrack_info ctinfo
,
884 const struct nf_conn
*loser_ct
)
886 struct nf_conn_acct
*acct
;
888 acct
= nf_conn_acct_find(loser_ct
);
890 struct nf_conn_counter
*counter
= acct
->counter
;
893 /* u32 should be fine since we must have seen one packet. */
894 bytes
= atomic64_read(&counter
[CTINFO2DIR(ctinfo
)].bytes
);
895 nf_ct_acct_update(ct
, CTINFO2DIR(ctinfo
), bytes
);
899 static void __nf_conntrack_insert_prepare(struct nf_conn
*ct
)
901 struct nf_conn_tstamp
*tstamp
;
903 atomic_inc(&ct
->ct_general
.use
);
904 ct
->status
|= IPS_CONFIRMED
;
906 /* set conntrack timestamp, if enabled. */
907 tstamp
= nf_conn_tstamp_find(ct
);
909 tstamp
->start
= ktime_get_real_ns();
912 static int __nf_ct_resolve_clash(struct sk_buff
*skb
,
913 struct nf_conntrack_tuple_hash
*h
)
915 /* This is the conntrack entry already in hashes that won race. */
916 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
917 enum ip_conntrack_info ctinfo
;
918 struct nf_conn
*loser_ct
;
920 loser_ct
= nf_ct_get(skb
, &ctinfo
);
922 if (nf_ct_is_dying(ct
))
925 if (!atomic_inc_not_zero(&ct
->ct_general
.use
))
928 if (((ct
->status
& IPS_NAT_DONE_MASK
) == 0) ||
929 nf_ct_match(ct
, loser_ct
)) {
930 struct net
*net
= nf_ct_net(ct
);
932 nf_ct_acct_merge(ct
, ctinfo
, loser_ct
);
933 nf_ct_add_to_dying_list(loser_ct
);
934 nf_conntrack_put(&loser_ct
->ct_general
);
935 nf_ct_set(skb
, ct
, ctinfo
);
937 NF_CT_STAT_INC(net
, insert_failed
);
946 * nf_ct_resolve_clash_harder - attempt to insert clashing conntrack entry
948 * @skb: skb that causes the collision
949 * @repl_idx: hash slot for reply direction
951 * Called when origin or reply direction had a clash.
952 * The skb can be handled without packet drop provided the reply direction
953 * is unique or there the existing entry has the identical tuple in both
956 * Caller must hold conntrack table locks to prevent concurrent updates.
958 * Returns NF_DROP if the clash could not be handled.
960 static int nf_ct_resolve_clash_harder(struct sk_buff
*skb
, u32 repl_idx
)
962 struct nf_conn
*loser_ct
= (struct nf_conn
*)skb_nfct(skb
);
963 const struct nf_conntrack_zone
*zone
;
964 struct nf_conntrack_tuple_hash
*h
;
965 struct hlist_nulls_node
*n
;
968 zone
= nf_ct_zone(loser_ct
);
969 net
= nf_ct_net(loser_ct
);
971 /* Reply direction must never result in a clash, unless both origin
972 * and reply tuples are identical.
974 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[repl_idx
], hnnode
) {
975 if (nf_ct_key_equal(h
,
976 &loser_ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
978 return __nf_ct_resolve_clash(skb
, h
);
981 /* We want the clashing entry to go away real soon: 1 second timeout. */
982 loser_ct
->timeout
= nfct_time_stamp
+ HZ
;
984 /* IPS_NAT_CLASH removes the entry automatically on the first
985 * reply. Also prevents UDP tracker from moving the entry to
986 * ASSURED state, i.e. the entry can always be evicted under
989 loser_ct
->status
|= IPS_FIXED_TIMEOUT
| IPS_NAT_CLASH
;
991 __nf_conntrack_insert_prepare(loser_ct
);
993 /* fake add for ORIGINAL dir: we want lookups to only find the entry
994 * already in the table. This also hides the clashing entry from
995 * ctnetlink iteration, i.e. conntrack -L won't show them.
997 hlist_nulls_add_fake(&loser_ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
);
999 hlist_nulls_add_head_rcu(&loser_ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
,
1000 &nf_conntrack_hash
[repl_idx
]);
1005 * nf_ct_resolve_clash - attempt to handle clash without packet drop
1007 * @skb: skb that causes the clash
1008 * @h: tuplehash of the clashing entry already in table
1009 * @hash_reply: hash slot for reply direction
1011 * A conntrack entry can be inserted to the connection tracking table
1012 * if there is no existing entry with an identical tuple.
1014 * If there is one, @skb (and the assocated, unconfirmed conntrack) has
1015 * to be dropped. In case @skb is retransmitted, next conntrack lookup
1016 * will find the already-existing entry.
1018 * The major problem with such packet drop is the extra delay added by
1019 * the packet loss -- it will take some time for a retransmit to occur
1020 * (or the sender to time out when waiting for a reply).
1022 * This function attempts to handle the situation without packet drop.
1024 * If @skb has no NAT transformation or if the colliding entries are
1025 * exactly the same, only the to-be-confirmed conntrack entry is discarded
1026 * and @skb is associated with the conntrack entry already in the table.
1028 * Failing that, the new, unconfirmed conntrack is still added to the table
1029 * provided that the collision only occurs in the ORIGINAL direction.
1030 * The new entry will be added after the existing one in the hash list,
1031 * so packets in the ORIGINAL direction will continue to match the existing
1032 * entry. The new entry will also have a fixed timeout so it expires --
1033 * due to the collision, it will not see bidirectional traffic.
1035 * Returns NF_DROP if the clash could not be resolved.
1037 static __cold noinline
int
1038 nf_ct_resolve_clash(struct sk_buff
*skb
, struct nf_conntrack_tuple_hash
*h
,
1041 /* This is the conntrack entry already in hashes that won race. */
1042 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
1043 const struct nf_conntrack_l4proto
*l4proto
;
1044 enum ip_conntrack_info ctinfo
;
1045 struct nf_conn
*loser_ct
;
1049 loser_ct
= nf_ct_get(skb
, &ctinfo
);
1050 net
= nf_ct_net(loser_ct
);
1052 l4proto
= nf_ct_l4proto_find(nf_ct_protonum(ct
));
1053 if (!l4proto
->allow_clash
)
1056 ret
= __nf_ct_resolve_clash(skb
, h
);
1057 if (ret
== NF_ACCEPT
)
1060 ret
= nf_ct_resolve_clash_harder(skb
, reply_hash
);
1061 if (ret
== NF_ACCEPT
)
1065 nf_ct_add_to_dying_list(loser_ct
);
1066 NF_CT_STAT_INC(net
, drop
);
1067 NF_CT_STAT_INC(net
, insert_failed
);
1071 /* Confirm a connection given skb; places it in hash table */
1073 __nf_conntrack_confirm(struct sk_buff
*skb
)
1075 const struct nf_conntrack_zone
*zone
;
1076 unsigned int hash
, reply_hash
;
1077 struct nf_conntrack_tuple_hash
*h
;
1079 struct nf_conn_help
*help
;
1080 struct hlist_nulls_node
*n
;
1081 enum ip_conntrack_info ctinfo
;
1083 unsigned int sequence
;
1086 ct
= nf_ct_get(skb
, &ctinfo
);
1087 net
= nf_ct_net(ct
);
1089 /* ipt_REJECT uses nf_conntrack_attach to attach related
1090 ICMP/TCP RST packets in other direction. Actual packet
1091 which created connection will be IP_CT_NEW or for an
1092 expected connection, IP_CT_RELATED. */
1093 if (CTINFO2DIR(ctinfo
) != IP_CT_DIR_ORIGINAL
)
1096 zone
= nf_ct_zone(ct
);
1100 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
1101 /* reuse the hash saved before */
1102 hash
= *(unsigned long *)&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
;
1103 hash
= scale_hash(hash
);
1104 reply_hash
= hash_conntrack(net
,
1105 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
1107 } while (nf_conntrack_double_lock(net
, hash
, reply_hash
, sequence
));
1109 /* We're not in hash table, and we refuse to set up related
1110 * connections for unconfirmed conns. But packet copies and
1111 * REJECT will give spurious warnings here.
1114 /* Another skb with the same unconfirmed conntrack may
1115 * win the race. This may happen for bridge(br_flood)
1116 * or broadcast/multicast packets do skb_clone with
1117 * unconfirmed conntrack.
1119 if (unlikely(nf_ct_is_confirmed(ct
))) {
1121 nf_conntrack_double_unlock(hash
, reply_hash
);
1126 pr_debug("Confirming conntrack %p\n", ct
);
1127 /* We have to check the DYING flag after unlink to prevent
1128 * a race against nf_ct_get_next_corpse() possibly called from
1129 * user context, else we insert an already 'dead' hash, blocking
1130 * further use of that particular connection -JM.
1132 nf_ct_del_from_dying_or_unconfirmed_list(ct
);
1134 if (unlikely(nf_ct_is_dying(ct
))) {
1135 nf_ct_add_to_dying_list(ct
);
1136 NF_CT_STAT_INC(net
, insert_failed
);
1140 /* See if there's one in the list already, including reverse:
1141 NAT could have grabbed it without realizing, since we're
1142 not in the hash. If there is, we lost race. */
1143 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[hash
], hnnode
)
1144 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
1148 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[reply_hash
], hnnode
)
1149 if (nf_ct_key_equal(h
, &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
,
1153 /* Timer relative to confirmation time, not original
1154 setting time, otherwise we'd get timer wrap in
1155 weird delay cases. */
1156 ct
->timeout
+= nfct_time_stamp
;
1158 __nf_conntrack_insert_prepare(ct
);
1160 /* Since the lookup is lockless, hash insertion must be done after
1161 * starting the timer and setting the CONFIRMED bit. The RCU barriers
1162 * guarantee that no other CPU can find the conntrack before the above
1163 * stores are visible.
1165 __nf_conntrack_hash_insert(ct
, hash
, reply_hash
);
1166 nf_conntrack_double_unlock(hash
, reply_hash
);
1169 help
= nfct_help(ct
);
1170 if (help
&& help
->helper
)
1171 nf_conntrack_event_cache(IPCT_HELPER
, ct
);
1173 nf_conntrack_event_cache(master_ct(ct
) ?
1174 IPCT_RELATED
: IPCT_NEW
, ct
);
1178 ret
= nf_ct_resolve_clash(skb
, h
, reply_hash
);
1180 nf_conntrack_double_unlock(hash
, reply_hash
);
1184 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm
);
1186 /* Returns true if a connection correspondings to the tuple (required
1189 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple
*tuple
,
1190 const struct nf_conn
*ignored_conntrack
)
1192 struct net
*net
= nf_ct_net(ignored_conntrack
);
1193 const struct nf_conntrack_zone
*zone
;
1194 struct nf_conntrack_tuple_hash
*h
;
1195 struct hlist_nulls_head
*ct_hash
;
1196 unsigned int hash
, hsize
;
1197 struct hlist_nulls_node
*n
;
1200 zone
= nf_ct_zone(ignored_conntrack
);
1204 nf_conntrack_get_ht(&ct_hash
, &hsize
);
1205 hash
= __hash_conntrack(net
, tuple
, hsize
);
1207 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[hash
], hnnode
) {
1208 ct
= nf_ct_tuplehash_to_ctrack(h
);
1210 if (ct
== ignored_conntrack
)
1213 if (nf_ct_is_expired(ct
)) {
1214 nf_ct_gc_expired(ct
);
1218 if (nf_ct_key_equal(h
, tuple
, zone
, net
)) {
1219 /* Tuple is taken already, so caller will need to find
1220 * a new source port to use.
1223 * If the *original tuples* are identical, then both
1224 * conntracks refer to the same flow.
1225 * This is a rare situation, it can occur e.g. when
1226 * more than one UDP packet is sent from same socket
1227 * in different threads.
1229 * Let nf_ct_resolve_clash() deal with this later.
1231 if (nf_ct_tuple_equal(&ignored_conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
,
1232 &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
))
1235 NF_CT_STAT_INC_ATOMIC(net
, found
);
1241 if (get_nulls_value(n
) != hash
) {
1242 NF_CT_STAT_INC_ATOMIC(net
, search_restart
);
1250 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken
);
1252 #define NF_CT_EVICTION_RANGE 8
1254 /* There's a small race here where we may free a just-assured
1255 connection. Too bad: we're in trouble anyway. */
1256 static unsigned int early_drop_list(struct net
*net
,
1257 struct hlist_nulls_head
*head
)
1259 struct nf_conntrack_tuple_hash
*h
;
1260 struct hlist_nulls_node
*n
;
1261 unsigned int drops
= 0;
1262 struct nf_conn
*tmp
;
1264 hlist_nulls_for_each_entry_rcu(h
, n
, head
, hnnode
) {
1265 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1267 if (test_bit(IPS_OFFLOAD_BIT
, &tmp
->status
))
1270 if (nf_ct_is_expired(tmp
)) {
1271 nf_ct_gc_expired(tmp
);
1275 if (test_bit(IPS_ASSURED_BIT
, &tmp
->status
) ||
1276 !net_eq(nf_ct_net(tmp
), net
) ||
1277 nf_ct_is_dying(tmp
))
1280 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1283 /* kill only if still in same netns -- might have moved due to
1284 * SLAB_TYPESAFE_BY_RCU rules.
1286 * We steal the timer reference. If that fails timer has
1287 * already fired or someone else deleted it. Just drop ref
1288 * and move to next entry.
1290 if (net_eq(nf_ct_net(tmp
), net
) &&
1291 nf_ct_is_confirmed(tmp
) &&
1292 nf_ct_delete(tmp
, 0, 0))
1301 static noinline
int early_drop(struct net
*net
, unsigned int hash
)
1303 unsigned int i
, bucket
;
1305 for (i
= 0; i
< NF_CT_EVICTION_RANGE
; i
++) {
1306 struct hlist_nulls_head
*ct_hash
;
1307 unsigned int hsize
, drops
;
1310 nf_conntrack_get_ht(&ct_hash
, &hsize
);
1312 bucket
= reciprocal_scale(hash
, hsize
);
1314 bucket
= (bucket
+ 1) % hsize
;
1316 drops
= early_drop_list(net
, &ct_hash
[bucket
]);
1320 NF_CT_STAT_ADD_ATOMIC(net
, early_drop
, drops
);
1328 static bool gc_worker_skip_ct(const struct nf_conn
*ct
)
1330 return !nf_ct_is_confirmed(ct
) || nf_ct_is_dying(ct
);
1333 static bool gc_worker_can_early_drop(const struct nf_conn
*ct
)
1335 const struct nf_conntrack_l4proto
*l4proto
;
1337 if (!test_bit(IPS_ASSURED_BIT
, &ct
->status
))
1340 l4proto
= nf_ct_l4proto_find(nf_ct_protonum(ct
));
1341 if (l4proto
->can_early_drop
&& l4proto
->can_early_drop(ct
))
1347 #define DAY (86400 * HZ)
1349 /* Set an arbitrary timeout large enough not to ever expire, this save
1350 * us a check for the IPS_OFFLOAD_BIT from the packet path via
1351 * nf_ct_is_expired().
1353 static void nf_ct_offload_timeout(struct nf_conn
*ct
)
1355 if (nf_ct_expires(ct
) < DAY
/ 2)
1356 ct
->timeout
= nfct_time_stamp
+ DAY
;
1359 static void gc_worker(struct work_struct
*work
)
1361 unsigned int min_interval
= max(HZ
/ GC_MAX_BUCKETS_DIV
, 1u);
1362 unsigned int i
, goal
, buckets
= 0, expired_count
= 0;
1363 unsigned int nf_conntrack_max95
= 0;
1364 struct conntrack_gc_work
*gc_work
;
1365 unsigned int ratio
, scanned
= 0;
1366 unsigned long next_run
;
1368 gc_work
= container_of(work
, struct conntrack_gc_work
, dwork
.work
);
1370 goal
= nf_conntrack_htable_size
/ GC_MAX_BUCKETS_DIV
;
1371 i
= gc_work
->last_bucket
;
1372 if (gc_work
->early_drop
)
1373 nf_conntrack_max95
= nf_conntrack_max
/ 100u * 95u;
1376 struct nf_conntrack_tuple_hash
*h
;
1377 struct hlist_nulls_head
*ct_hash
;
1378 struct hlist_nulls_node
*n
;
1379 unsigned int hashsz
;
1380 struct nf_conn
*tmp
;
1385 nf_conntrack_get_ht(&ct_hash
, &hashsz
);
1389 hlist_nulls_for_each_entry_rcu(h
, n
, &ct_hash
[i
], hnnode
) {
1392 tmp
= nf_ct_tuplehash_to_ctrack(h
);
1395 if (test_bit(IPS_OFFLOAD_BIT
, &tmp
->status
)) {
1396 nf_ct_offload_timeout(tmp
);
1400 if (nf_ct_is_expired(tmp
)) {
1401 nf_ct_gc_expired(tmp
);
1406 if (nf_conntrack_max95
== 0 || gc_worker_skip_ct(tmp
))
1409 net
= nf_ct_net(tmp
);
1410 if (atomic_read(&net
->ct
.count
) < nf_conntrack_max95
)
1413 /* need to take reference to avoid possible races */
1414 if (!atomic_inc_not_zero(&tmp
->ct_general
.use
))
1417 if (gc_worker_skip_ct(tmp
)) {
1422 if (gc_worker_can_early_drop(tmp
))
1428 /* could check get_nulls_value() here and restart if ct
1429 * was moved to another chain. But given gc is best-effort
1430 * we will just continue with next hash slot.
1434 } while (++buckets
< goal
);
1436 if (gc_work
->exiting
)
1440 * Eviction will normally happen from the packet path, and not
1441 * from this gc worker.
1443 * This worker is only here to reap expired entries when system went
1444 * idle after a busy period.
1446 * The heuristics below are supposed to balance conflicting goals:
1448 * 1. Minimize time until we notice a stale entry
1449 * 2. Maximize scan intervals to not waste cycles
1451 * Normally, expire ratio will be close to 0.
1453 * As soon as a sizeable fraction of the entries have expired
1454 * increase scan frequency.
1456 ratio
= scanned
? expired_count
* 100 / scanned
: 0;
1457 if (ratio
> GC_EVICT_RATIO
) {
1458 gc_work
->next_gc_run
= min_interval
;
1460 unsigned int max
= GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
;
1462 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES
/ GC_MAX_BUCKETS_DIV
) == 0);
1464 gc_work
->next_gc_run
+= min_interval
;
1465 if (gc_work
->next_gc_run
> max
)
1466 gc_work
->next_gc_run
= max
;
1469 next_run
= gc_work
->next_gc_run
;
1470 gc_work
->last_bucket
= i
;
1471 gc_work
->early_drop
= false;
1472 queue_delayed_work(system_power_efficient_wq
, &gc_work
->dwork
, next_run
);
1475 static void conntrack_gc_work_init(struct conntrack_gc_work
*gc_work
)
1477 INIT_DEFERRABLE_WORK(&gc_work
->dwork
, gc_worker
);
1478 gc_work
->next_gc_run
= HZ
;
1479 gc_work
->exiting
= false;
1482 static struct nf_conn
*
1483 __nf_conntrack_alloc(struct net
*net
,
1484 const struct nf_conntrack_zone
*zone
,
1485 const struct nf_conntrack_tuple
*orig
,
1486 const struct nf_conntrack_tuple
*repl
,
1487 gfp_t gfp
, u32 hash
)
1491 /* We don't want any race condition at early drop stage */
1492 atomic_inc(&net
->ct
.count
);
1494 if (nf_conntrack_max
&&
1495 unlikely(atomic_read(&net
->ct
.count
) > nf_conntrack_max
)) {
1496 if (!early_drop(net
, hash
)) {
1497 if (!conntrack_gc_work
.early_drop
)
1498 conntrack_gc_work
.early_drop
= true;
1499 atomic_dec(&net
->ct
.count
);
1500 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1501 return ERR_PTR(-ENOMEM
);
1506 * Do not use kmem_cache_zalloc(), as this cache uses
1507 * SLAB_TYPESAFE_BY_RCU.
1509 ct
= kmem_cache_alloc(nf_conntrack_cachep
, gfp
);
1513 spin_lock_init(&ct
->lock
);
1514 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
= *orig
;
1515 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].hnnode
.pprev
= NULL
;
1516 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *repl
;
1517 /* save hash for reusing when confirming */
1518 *(unsigned long *)(&ct
->tuplehash
[IP_CT_DIR_REPLY
].hnnode
.pprev
) = hash
;
1521 write_pnet(&ct
->ct_net
, net
);
1522 memset(&ct
->__nfct_init_offset
[0], 0,
1523 offsetof(struct nf_conn
, proto
) -
1524 offsetof(struct nf_conn
, __nfct_init_offset
[0]));
1526 nf_ct_zone_add(ct
, zone
);
1528 /* Because we use RCU lookups, we set ct_general.use to zero before
1529 * this is inserted in any list.
1531 atomic_set(&ct
->ct_general
.use
, 0);
1534 atomic_dec(&net
->ct
.count
);
1535 return ERR_PTR(-ENOMEM
);
1538 struct nf_conn
*nf_conntrack_alloc(struct net
*net
,
1539 const struct nf_conntrack_zone
*zone
,
1540 const struct nf_conntrack_tuple
*orig
,
1541 const struct nf_conntrack_tuple
*repl
,
1544 return __nf_conntrack_alloc(net
, zone
, orig
, repl
, gfp
, 0);
1546 EXPORT_SYMBOL_GPL(nf_conntrack_alloc
);
1548 void nf_conntrack_free(struct nf_conn
*ct
)
1550 struct net
*net
= nf_ct_net(ct
);
1552 /* A freed object has refcnt == 0, that's
1553 * the golden rule for SLAB_TYPESAFE_BY_RCU
1555 WARN_ON(atomic_read(&ct
->ct_general
.use
) != 0);
1557 nf_ct_ext_destroy(ct
);
1558 kmem_cache_free(nf_conntrack_cachep
, ct
);
1559 smp_mb__before_atomic();
1560 atomic_dec(&net
->ct
.count
);
1562 EXPORT_SYMBOL_GPL(nf_conntrack_free
);
1565 /* Allocate a new conntrack: we return -ENOMEM if classification
1566 failed due to stress. Otherwise it really is unclassifiable. */
1567 static noinline
struct nf_conntrack_tuple_hash
*
1568 init_conntrack(struct net
*net
, struct nf_conn
*tmpl
,
1569 const struct nf_conntrack_tuple
*tuple
,
1570 struct sk_buff
*skb
,
1571 unsigned int dataoff
, u32 hash
)
1574 struct nf_conn_help
*help
;
1575 struct nf_conntrack_tuple repl_tuple
;
1576 struct nf_conntrack_ecache
*ecache
;
1577 struct nf_conntrack_expect
*exp
= NULL
;
1578 const struct nf_conntrack_zone
*zone
;
1579 struct nf_conn_timeout
*timeout_ext
;
1580 struct nf_conntrack_zone tmp
;
1582 if (!nf_ct_invert_tuple(&repl_tuple
, tuple
)) {
1583 pr_debug("Can't invert tuple.\n");
1587 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1588 ct
= __nf_conntrack_alloc(net
, zone
, tuple
, &repl_tuple
, GFP_ATOMIC
,
1591 return (struct nf_conntrack_tuple_hash
*)ct
;
1593 if (!nf_ct_add_synproxy(ct
, tmpl
)) {
1594 nf_conntrack_free(ct
);
1595 return ERR_PTR(-ENOMEM
);
1598 timeout_ext
= tmpl
? nf_ct_timeout_find(tmpl
) : NULL
;
1601 nf_ct_timeout_ext_add(ct
, rcu_dereference(timeout_ext
->timeout
),
1604 nf_ct_acct_ext_add(ct
, GFP_ATOMIC
);
1605 nf_ct_tstamp_ext_add(ct
, GFP_ATOMIC
);
1606 nf_ct_labels_ext_add(ct
);
1608 ecache
= tmpl
? nf_ct_ecache_find(tmpl
) : NULL
;
1609 nf_ct_ecache_ext_add(ct
, ecache
? ecache
->ctmask
: 0,
1610 ecache
? ecache
->expmask
: 0,
1614 if (net
->ct
.expect_count
) {
1615 spin_lock(&nf_conntrack_expect_lock
);
1616 exp
= nf_ct_find_expectation(net
, zone
, tuple
);
1618 pr_debug("expectation arrives ct=%p exp=%p\n",
1620 /* Welcome, Mr. Bond. We've been expecting you... */
1621 __set_bit(IPS_EXPECTED_BIT
, &ct
->status
);
1622 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1623 ct
->master
= exp
->master
;
1625 help
= nf_ct_helper_ext_add(ct
, GFP_ATOMIC
);
1627 rcu_assign_pointer(help
->helper
, exp
->helper
);
1630 #ifdef CONFIG_NF_CONNTRACK_MARK
1631 ct
->mark
= exp
->master
->mark
;
1633 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1634 ct
->secmark
= exp
->master
->secmark
;
1636 NF_CT_STAT_INC(net
, expect_new
);
1638 spin_unlock(&nf_conntrack_expect_lock
);
1641 __nf_ct_try_assign_helper(ct
, tmpl
, GFP_ATOMIC
);
1643 /* Now it is inserted into the unconfirmed list, bump refcount */
1644 nf_conntrack_get(&ct
->ct_general
);
1645 nf_ct_add_to_unconfirmed_list(ct
);
1651 exp
->expectfn(ct
, exp
);
1652 nf_ct_expect_put(exp
);
1655 return &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
];
1658 /* On success, returns 0, sets skb->_nfct | ctinfo */
1660 resolve_normal_ct(struct nf_conn
*tmpl
,
1661 struct sk_buff
*skb
,
1662 unsigned int dataoff
,
1664 const struct nf_hook_state
*state
)
1666 const struct nf_conntrack_zone
*zone
;
1667 struct nf_conntrack_tuple tuple
;
1668 struct nf_conntrack_tuple_hash
*h
;
1669 enum ip_conntrack_info ctinfo
;
1670 struct nf_conntrack_zone tmp
;
1674 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
),
1675 dataoff
, state
->pf
, protonum
, state
->net
,
1677 pr_debug("Can't get tuple\n");
1681 /* look for tuple match */
1682 zone
= nf_ct_zone_tmpl(tmpl
, skb
, &tmp
);
1683 hash
= hash_conntrack_raw(&tuple
, state
->net
);
1684 h
= __nf_conntrack_find_get(state
->net
, zone
, &tuple
, hash
);
1686 h
= init_conntrack(state
->net
, tmpl
, &tuple
,
1687 skb
, dataoff
, hash
);
1693 ct
= nf_ct_tuplehash_to_ctrack(h
);
1695 /* It exists; we have (non-exclusive) reference. */
1696 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
) {
1697 ctinfo
= IP_CT_ESTABLISHED_REPLY
;
1699 /* Once we've had two way comms, always ESTABLISHED. */
1700 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
)) {
1701 pr_debug("normal packet for %p\n", ct
);
1702 ctinfo
= IP_CT_ESTABLISHED
;
1703 } else if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
)) {
1704 pr_debug("related packet for %p\n", ct
);
1705 ctinfo
= IP_CT_RELATED
;
1707 pr_debug("new packet for %p\n", ct
);
1711 nf_ct_set(skb
, ct
, ctinfo
);
1716 * icmp packets need special treatment to handle error messages that are
1717 * related to a connection.
1719 * Callers need to check if skb has a conntrack assigned when this
1720 * helper returns; in such case skb belongs to an already known connection.
1722 static unsigned int __cold
1723 nf_conntrack_handle_icmp(struct nf_conn
*tmpl
,
1724 struct sk_buff
*skb
,
1725 unsigned int dataoff
,
1727 const struct nf_hook_state
*state
)
1731 if (state
->pf
== NFPROTO_IPV4
&& protonum
== IPPROTO_ICMP
)
1732 ret
= nf_conntrack_icmpv4_error(tmpl
, skb
, dataoff
, state
);
1733 #if IS_ENABLED(CONFIG_IPV6)
1734 else if (state
->pf
== NFPROTO_IPV6
&& protonum
== IPPROTO_ICMPV6
)
1735 ret
= nf_conntrack_icmpv6_error(tmpl
, skb
, dataoff
, state
);
1741 NF_CT_STAT_INC_ATOMIC(state
->net
, error
);
1742 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1748 static int generic_packet(struct nf_conn
*ct
, struct sk_buff
*skb
,
1749 enum ip_conntrack_info ctinfo
)
1751 const unsigned int *timeout
= nf_ct_timeout_lookup(ct
);
1754 timeout
= &nf_generic_pernet(nf_ct_net(ct
))->timeout
;
1756 nf_ct_refresh_acct(ct
, ctinfo
, skb
, *timeout
);
1760 /* Returns verdict for packet, or -1 for invalid. */
1761 static int nf_conntrack_handle_packet(struct nf_conn
*ct
,
1762 struct sk_buff
*skb
,
1763 unsigned int dataoff
,
1764 enum ip_conntrack_info ctinfo
,
1765 const struct nf_hook_state
*state
)
1767 switch (nf_ct_protonum(ct
)) {
1769 return nf_conntrack_tcp_packet(ct
, skb
, dataoff
,
1772 return nf_conntrack_udp_packet(ct
, skb
, dataoff
,
1775 return nf_conntrack_icmp_packet(ct
, skb
, ctinfo
, state
);
1776 #if IS_ENABLED(CONFIG_IPV6)
1777 case IPPROTO_ICMPV6
:
1778 return nf_conntrack_icmpv6_packet(ct
, skb
, ctinfo
, state
);
1780 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
1781 case IPPROTO_UDPLITE
:
1782 return nf_conntrack_udplite_packet(ct
, skb
, dataoff
,
1785 #ifdef CONFIG_NF_CT_PROTO_SCTP
1787 return nf_conntrack_sctp_packet(ct
, skb
, dataoff
,
1790 #ifdef CONFIG_NF_CT_PROTO_DCCP
1792 return nf_conntrack_dccp_packet(ct
, skb
, dataoff
,
1795 #ifdef CONFIG_NF_CT_PROTO_GRE
1797 return nf_conntrack_gre_packet(ct
, skb
, dataoff
,
1802 return generic_packet(ct
, skb
, ctinfo
);
1806 nf_conntrack_in(struct sk_buff
*skb
, const struct nf_hook_state
*state
)
1808 enum ip_conntrack_info ctinfo
;
1809 struct nf_conn
*ct
, *tmpl
;
1813 tmpl
= nf_ct_get(skb
, &ctinfo
);
1814 if (tmpl
|| ctinfo
== IP_CT_UNTRACKED
) {
1815 /* Previously seen (loopback or untracked)? Ignore. */
1816 if ((tmpl
&& !nf_ct_is_template(tmpl
)) ||
1817 ctinfo
== IP_CT_UNTRACKED
) {
1818 NF_CT_STAT_INC_ATOMIC(state
->net
, ignore
);
1824 /* rcu_read_lock()ed by nf_hook_thresh */
1825 dataoff
= get_l4proto(skb
, skb_network_offset(skb
), state
->pf
, &protonum
);
1827 pr_debug("not prepared to track yet or error occurred\n");
1828 NF_CT_STAT_INC_ATOMIC(state
->net
, error
);
1829 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1834 if (protonum
== IPPROTO_ICMP
|| protonum
== IPPROTO_ICMPV6
) {
1835 ret
= nf_conntrack_handle_icmp(tmpl
, skb
, dataoff
,
1841 /* ICMP[v6] protocol trackers may assign one conntrack. */
1846 ret
= resolve_normal_ct(tmpl
, skb
, dataoff
,
1849 /* Too stressed to deal. */
1850 NF_CT_STAT_INC_ATOMIC(state
->net
, drop
);
1855 ct
= nf_ct_get(skb
, &ctinfo
);
1857 /* Not valid part of a connection */
1858 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1863 ret
= nf_conntrack_handle_packet(ct
, skb
, dataoff
, ctinfo
, state
);
1865 /* Invalid: inverse of the return code tells
1866 * the netfilter core what to do */
1867 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1868 nf_conntrack_put(&ct
->ct_general
);
1870 NF_CT_STAT_INC_ATOMIC(state
->net
, invalid
);
1871 if (ret
== -NF_DROP
)
1872 NF_CT_STAT_INC_ATOMIC(state
->net
, drop
);
1873 /* Special case: TCP tracker reports an attempt to reopen a
1874 * closed/aborted connection. We have to go back and create a
1877 if (ret
== -NF_REPEAT
)
1883 if (ctinfo
== IP_CT_ESTABLISHED_REPLY
&&
1884 !test_and_set_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
1885 nf_conntrack_event_cache(IPCT_REPLY
, ct
);
1892 EXPORT_SYMBOL_GPL(nf_conntrack_in
);
1894 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1895 implicitly racy: see __nf_conntrack_confirm */
1896 void nf_conntrack_alter_reply(struct nf_conn
*ct
,
1897 const struct nf_conntrack_tuple
*newreply
)
1899 struct nf_conn_help
*help
= nfct_help(ct
);
1901 /* Should be unconfirmed, so not in hash table yet */
1902 WARN_ON(nf_ct_is_confirmed(ct
));
1904 pr_debug("Altering reply tuple of %p to ", ct
);
1905 nf_ct_dump_tuple(newreply
);
1907 ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
= *newreply
;
1908 if (ct
->master
|| (help
&& !hlist_empty(&help
->expectations
)))
1912 __nf_ct_try_assign_helper(ct
, NULL
, GFP_ATOMIC
);
1915 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply
);
1917 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1918 void __nf_ct_refresh_acct(struct nf_conn
*ct
,
1919 enum ip_conntrack_info ctinfo
,
1920 const struct sk_buff
*skb
,
1924 /* Only update if this is not a fixed timeout */
1925 if (test_bit(IPS_FIXED_TIMEOUT_BIT
, &ct
->status
))
1928 /* If not in hash table, timer will not be active yet */
1929 if (nf_ct_is_confirmed(ct
))
1930 extra_jiffies
+= nfct_time_stamp
;
1932 if (READ_ONCE(ct
->timeout
) != extra_jiffies
)
1933 WRITE_ONCE(ct
->timeout
, extra_jiffies
);
1936 nf_ct_acct_update(ct
, CTINFO2DIR(ctinfo
), skb
->len
);
1938 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct
);
1940 bool nf_ct_kill_acct(struct nf_conn
*ct
,
1941 enum ip_conntrack_info ctinfo
,
1942 const struct sk_buff
*skb
)
1944 nf_ct_acct_update(ct
, CTINFO2DIR(ctinfo
), skb
->len
);
1946 return nf_ct_delete(ct
, 0, 0);
1948 EXPORT_SYMBOL_GPL(nf_ct_kill_acct
);
1950 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1952 #include <linux/netfilter/nfnetlink.h>
1953 #include <linux/netfilter/nfnetlink_conntrack.h>
1954 #include <linux/mutex.h>
1956 /* Generic function for tcp/udp/sctp/dccp and alike. */
1957 int nf_ct_port_tuple_to_nlattr(struct sk_buff
*skb
,
1958 const struct nf_conntrack_tuple
*tuple
)
1960 if (nla_put_be16(skb
, CTA_PROTO_SRC_PORT
, tuple
->src
.u
.tcp
.port
) ||
1961 nla_put_be16(skb
, CTA_PROTO_DST_PORT
, tuple
->dst
.u
.tcp
.port
))
1962 goto nla_put_failure
;
1968 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr
);
1970 const struct nla_policy nf_ct_port_nla_policy
[CTA_PROTO_MAX
+1] = {
1971 [CTA_PROTO_SRC_PORT
] = { .type
= NLA_U16
},
1972 [CTA_PROTO_DST_PORT
] = { .type
= NLA_U16
},
1974 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy
);
1976 int nf_ct_port_nlattr_to_tuple(struct nlattr
*tb
[],
1977 struct nf_conntrack_tuple
*t
)
1979 if (!tb
[CTA_PROTO_SRC_PORT
] || !tb
[CTA_PROTO_DST_PORT
])
1982 t
->src
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_SRC_PORT
]);
1983 t
->dst
.u
.tcp
.port
= nla_get_be16(tb
[CTA_PROTO_DST_PORT
]);
1987 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple
);
1989 unsigned int nf_ct_port_nlattr_tuple_size(void)
1991 static unsigned int size __read_mostly
;
1994 size
= nla_policy_len(nf_ct_port_nla_policy
, CTA_PROTO_MAX
+ 1);
1998 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size
);
2001 /* Used by ipt_REJECT and ip6t_REJECT. */
2002 static void nf_conntrack_attach(struct sk_buff
*nskb
, const struct sk_buff
*skb
)
2005 enum ip_conntrack_info ctinfo
;
2007 /* This ICMP is in reverse direction to the packet which caused it */
2008 ct
= nf_ct_get(skb
, &ctinfo
);
2009 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_ORIGINAL
)
2010 ctinfo
= IP_CT_RELATED_REPLY
;
2012 ctinfo
= IP_CT_RELATED
;
2014 /* Attach to new skbuff, and increment count */
2015 nf_ct_set(nskb
, ct
, ctinfo
);
2016 nf_conntrack_get(skb_nfct(nskb
));
2019 static int nf_conntrack_update(struct net
*net
, struct sk_buff
*skb
)
2021 struct nf_conntrack_tuple_hash
*h
;
2022 struct nf_conntrack_tuple tuple
;
2023 enum ip_conntrack_info ctinfo
;
2024 struct nf_nat_hook
*nat_hook
;
2025 unsigned int status
;
2031 ct
= nf_ct_get(skb
, &ctinfo
);
2032 if (!ct
|| nf_ct_is_confirmed(ct
))
2035 l3num
= nf_ct_l3num(ct
);
2037 dataoff
= get_l4proto(skb
, skb_network_offset(skb
), l3num
, &l4num
);
2041 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
), dataoff
, l3num
,
2042 l4num
, net
, &tuple
))
2045 if (ct
->status
& IPS_SRC_NAT
) {
2046 memcpy(tuple
.src
.u3
.all
,
2047 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.u3
.all
,
2048 sizeof(tuple
.src
.u3
.all
));
2050 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.u
.all
;
2053 if (ct
->status
& IPS_DST_NAT
) {
2054 memcpy(tuple
.dst
.u3
.all
,
2055 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.u3
.all
,
2056 sizeof(tuple
.dst
.u3
.all
));
2058 ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.u
.all
;
2061 h
= nf_conntrack_find_get(net
, nf_ct_zone(ct
), &tuple
);
2065 /* Store status bits of the conntrack that is clashing to re-do NAT
2066 * mangling according to what it has been done already to this packet.
2068 status
= ct
->status
;
2071 ct
= nf_ct_tuplehash_to_ctrack(h
);
2072 nf_ct_set(skb
, ct
, ctinfo
);
2074 nat_hook
= rcu_dereference(nf_nat_hook
);
2078 if (status
& IPS_SRC_NAT
&&
2079 nat_hook
->manip_pkt(skb
, ct
, NF_NAT_MANIP_SRC
,
2080 IP_CT_DIR_ORIGINAL
) == NF_DROP
)
2083 if (status
& IPS_DST_NAT
&&
2084 nat_hook
->manip_pkt(skb
, ct
, NF_NAT_MANIP_DST
,
2085 IP_CT_DIR_ORIGINAL
) == NF_DROP
)
2091 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple
*dst_tuple
,
2092 const struct sk_buff
*skb
)
2094 const struct nf_conntrack_tuple
*src_tuple
;
2095 const struct nf_conntrack_tuple_hash
*hash
;
2096 struct nf_conntrack_tuple srctuple
;
2097 enum ip_conntrack_info ctinfo
;
2100 ct
= nf_ct_get(skb
, &ctinfo
);
2102 src_tuple
= nf_ct_tuple(ct
, CTINFO2DIR(ctinfo
));
2103 memcpy(dst_tuple
, src_tuple
, sizeof(*dst_tuple
));
2107 if (!nf_ct_get_tuplepr(skb
, skb_network_offset(skb
),
2108 NFPROTO_IPV4
, dev_net(skb
->dev
),
2112 hash
= nf_conntrack_find_get(dev_net(skb
->dev
),
2118 ct
= nf_ct_tuplehash_to_ctrack(hash
);
2119 src_tuple
= nf_ct_tuple(ct
, !hash
->tuple
.dst
.dir
);
2120 memcpy(dst_tuple
, src_tuple
, sizeof(*dst_tuple
));
2126 /* Bring out ya dead! */
2127 static struct nf_conn
*
2128 get_next_corpse(int (*iter
)(struct nf_conn
*i
, void *data
),
2129 void *data
, unsigned int *bucket
)
2131 struct nf_conntrack_tuple_hash
*h
;
2133 struct hlist_nulls_node
*n
;
2136 for (; *bucket
< nf_conntrack_htable_size
; (*bucket
)++) {
2137 lockp
= &nf_conntrack_locks
[*bucket
% CONNTRACK_LOCKS
];
2139 nf_conntrack_lock(lockp
);
2140 if (*bucket
< nf_conntrack_htable_size
) {
2141 hlist_nulls_for_each_entry(h
, n
, &nf_conntrack_hash
[*bucket
], hnnode
) {
2142 if (NF_CT_DIRECTION(h
) != IP_CT_DIR_ORIGINAL
)
2144 ct
= nf_ct_tuplehash_to_ctrack(h
);
2156 atomic_inc(&ct
->ct_general
.use
);
2162 static void nf_ct_iterate_cleanup(int (*iter
)(struct nf_conn
*i
, void *data
),
2163 void *data
, u32 portid
, int report
)
2165 unsigned int bucket
= 0, sequence
;
2171 sequence
= read_seqcount_begin(&nf_conntrack_generation
);
2173 while ((ct
= get_next_corpse(iter
, data
, &bucket
)) != NULL
) {
2174 /* Time to push up daises... */
2176 nf_ct_delete(ct
, portid
, report
);
2181 if (!read_seqcount_retry(&nf_conntrack_generation
, sequence
))
2188 int (*iter
)(struct nf_conn
*i
, void *data
);
2193 static int iter_net_only(struct nf_conn
*i
, void *data
)
2195 struct iter_data
*d
= data
;
2197 if (!net_eq(d
->net
, nf_ct_net(i
)))
2200 return d
->iter(i
, d
->data
);
2204 __nf_ct_unconfirmed_destroy(struct net
*net
)
2208 for_each_possible_cpu(cpu
) {
2209 struct nf_conntrack_tuple_hash
*h
;
2210 struct hlist_nulls_node
*n
;
2211 struct ct_pcpu
*pcpu
;
2213 pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2215 spin_lock_bh(&pcpu
->lock
);
2216 hlist_nulls_for_each_entry(h
, n
, &pcpu
->unconfirmed
, hnnode
) {
2219 ct
= nf_ct_tuplehash_to_ctrack(h
);
2221 /* we cannot call iter() on unconfirmed list, the
2222 * owning cpu can reallocate ct->ext at any time.
2224 set_bit(IPS_DYING_BIT
, &ct
->status
);
2226 spin_unlock_bh(&pcpu
->lock
);
2231 void nf_ct_unconfirmed_destroy(struct net
*net
)
2235 if (atomic_read(&net
->ct
.count
) > 0) {
2236 __nf_ct_unconfirmed_destroy(net
);
2237 nf_queue_nf_hook_drop(net
);
2241 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy
);
2243 void nf_ct_iterate_cleanup_net(struct net
*net
,
2244 int (*iter
)(struct nf_conn
*i
, void *data
),
2245 void *data
, u32 portid
, int report
)
2251 if (atomic_read(&net
->ct
.count
) == 0)
2258 nf_ct_iterate_cleanup(iter_net_only
, &d
, portid
, report
);
2260 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net
);
2263 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2264 * @iter: callback to invoke for each conntrack
2265 * @data: data to pass to @iter
2267 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2268 * unconfirmed list as dying (so they will not be inserted into
2271 * Can only be called in module exit path.
2274 nf_ct_iterate_destroy(int (*iter
)(struct nf_conn
*i
, void *data
), void *data
)
2278 down_read(&net_rwsem
);
2280 if (atomic_read(&net
->ct
.count
) == 0)
2282 __nf_ct_unconfirmed_destroy(net
);
2283 nf_queue_nf_hook_drop(net
);
2285 up_read(&net_rwsem
);
2287 /* Need to wait for netns cleanup worker to finish, if its
2288 * running -- it might have deleted a net namespace from
2289 * the global list, so our __nf_ct_unconfirmed_destroy() might
2290 * not have affected all namespaces.
2294 /* a conntrack could have been unlinked from unconfirmed list
2295 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2296 * This makes sure its inserted into conntrack table.
2300 nf_ct_iterate_cleanup(iter
, data
, 0, 0);
2302 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy
);
2304 static int kill_all(struct nf_conn
*i
, void *data
)
2306 return net_eq(nf_ct_net(i
), data
);
2309 void nf_conntrack_cleanup_start(void)
2311 conntrack_gc_work
.exiting
= true;
2312 RCU_INIT_POINTER(ip_ct_attach
, NULL
);
2315 void nf_conntrack_cleanup_end(void)
2317 RCU_INIT_POINTER(nf_ct_hook
, NULL
);
2318 cancel_delayed_work_sync(&conntrack_gc_work
.dwork
);
2319 kvfree(nf_conntrack_hash
);
2321 nf_conntrack_proto_fini();
2322 nf_conntrack_seqadj_fini();
2323 nf_conntrack_labels_fini();
2324 nf_conntrack_helper_fini();
2325 nf_conntrack_timeout_fini();
2326 nf_conntrack_ecache_fini();
2327 nf_conntrack_tstamp_fini();
2328 nf_conntrack_acct_fini();
2329 nf_conntrack_expect_fini();
2331 kmem_cache_destroy(nf_conntrack_cachep
);
2335 * Mishearing the voices in his head, our hero wonders how he's
2336 * supposed to kill the mall.
2338 void nf_conntrack_cleanup_net(struct net
*net
)
2342 list_add(&net
->exit_list
, &single
);
2343 nf_conntrack_cleanup_net_list(&single
);
2346 void nf_conntrack_cleanup_net_list(struct list_head
*net_exit_list
)
2352 * This makes sure all current packets have passed through
2353 * netfilter framework. Roll on, two-stage module
2359 list_for_each_entry(net
, net_exit_list
, exit_list
) {
2360 nf_ct_iterate_cleanup(kill_all
, net
, 0, 0);
2361 if (atomic_read(&net
->ct
.count
) != 0)
2366 goto i_see_dead_people
;
2369 list_for_each_entry(net
, net_exit_list
, exit_list
) {
2370 nf_conntrack_proto_pernet_fini(net
);
2371 nf_conntrack_ecache_pernet_fini(net
);
2372 nf_conntrack_expect_pernet_fini(net
);
2373 free_percpu(net
->ct
.stat
);
2374 free_percpu(net
->ct
.pcpu_lists
);
2378 void *nf_ct_alloc_hashtable(unsigned int *sizep
, int nulls
)
2380 struct hlist_nulls_head
*hash
;
2381 unsigned int nr_slots
, i
;
2383 if (*sizep
> (UINT_MAX
/ sizeof(struct hlist_nulls_head
)))
2386 BUILD_BUG_ON(sizeof(struct hlist_nulls_head
) != sizeof(struct hlist_head
));
2387 nr_slots
= *sizep
= roundup(*sizep
, PAGE_SIZE
/ sizeof(struct hlist_nulls_head
));
2389 hash
= kvcalloc(nr_slots
, sizeof(struct hlist_nulls_head
), GFP_KERNEL
);
2392 for (i
= 0; i
< nr_slots
; i
++)
2393 INIT_HLIST_NULLS_HEAD(&hash
[i
], i
);
2397 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable
);
2399 int nf_conntrack_hash_resize(unsigned int hashsize
)
2402 unsigned int old_size
;
2403 struct hlist_nulls_head
*hash
, *old_hash
;
2404 struct nf_conntrack_tuple_hash
*h
;
2410 hash
= nf_ct_alloc_hashtable(&hashsize
, 1);
2414 old_size
= nf_conntrack_htable_size
;
2415 if (old_size
== hashsize
) {
2421 nf_conntrack_all_lock();
2422 write_seqcount_begin(&nf_conntrack_generation
);
2424 /* Lookups in the old hash might happen in parallel, which means we
2425 * might get false negatives during connection lookup. New connections
2426 * created because of a false negative won't make it into the hash
2427 * though since that required taking the locks.
2430 for (i
= 0; i
< nf_conntrack_htable_size
; i
++) {
2431 while (!hlist_nulls_empty(&nf_conntrack_hash
[i
])) {
2432 h
= hlist_nulls_entry(nf_conntrack_hash
[i
].first
,
2433 struct nf_conntrack_tuple_hash
, hnnode
);
2434 ct
= nf_ct_tuplehash_to_ctrack(h
);
2435 hlist_nulls_del_rcu(&h
->hnnode
);
2436 bucket
= __hash_conntrack(nf_ct_net(ct
),
2437 &h
->tuple
, hashsize
);
2438 hlist_nulls_add_head_rcu(&h
->hnnode
, &hash
[bucket
]);
2441 old_size
= nf_conntrack_htable_size
;
2442 old_hash
= nf_conntrack_hash
;
2444 nf_conntrack_hash
= hash
;
2445 nf_conntrack_htable_size
= hashsize
;
2447 write_seqcount_end(&nf_conntrack_generation
);
2448 nf_conntrack_all_unlock();
2456 int nf_conntrack_set_hashsize(const char *val
, const struct kernel_param
*kp
)
2458 unsigned int hashsize
;
2461 if (current
->nsproxy
->net_ns
!= &init_net
)
2464 /* On boot, we can set this without any fancy locking. */
2465 if (!nf_conntrack_hash
)
2466 return param_set_uint(val
, kp
);
2468 rc
= kstrtouint(val
, 0, &hashsize
);
2472 return nf_conntrack_hash_resize(hashsize
);
2475 static __always_inline
unsigned int total_extension_size(void)
2477 /* remember to add new extensions below */
2478 BUILD_BUG_ON(NF_CT_EXT_NUM
> 9);
2480 return sizeof(struct nf_ct_ext
) +
2481 sizeof(struct nf_conn_help
)
2482 #if IS_ENABLED(CONFIG_NF_NAT)
2483 + sizeof(struct nf_conn_nat
)
2485 + sizeof(struct nf_conn_seqadj
)
2486 + sizeof(struct nf_conn_acct
)
2487 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2488 + sizeof(struct nf_conntrack_ecache
)
2490 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2491 + sizeof(struct nf_conn_tstamp
)
2493 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2494 + sizeof(struct nf_conn_timeout
)
2496 #ifdef CONFIG_NF_CONNTRACK_LABELS
2497 + sizeof(struct nf_conn_labels
)
2499 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2500 + sizeof(struct nf_conn_synproxy
)
2505 int nf_conntrack_init_start(void)
2507 unsigned long nr_pages
= totalram_pages();
2512 /* struct nf_ct_ext uses u8 to store offsets/size */
2513 BUILD_BUG_ON(total_extension_size() > 255u);
2515 seqcount_init(&nf_conntrack_generation
);
2517 for (i
= 0; i
< CONNTRACK_LOCKS
; i
++)
2518 spin_lock_init(&nf_conntrack_locks
[i
]);
2520 if (!nf_conntrack_htable_size
) {
2521 /* Idea from tcp.c: use 1/16384 of memory.
2522 * On i386: 32MB machine has 512 buckets.
2523 * >= 1GB machines have 16384 buckets.
2524 * >= 4GB machines have 65536 buckets.
2526 nf_conntrack_htable_size
2527 = (((nr_pages
<< PAGE_SHIFT
) / 16384)
2528 / sizeof(struct hlist_head
));
2529 if (nr_pages
> (4 * (1024 * 1024 * 1024 / PAGE_SIZE
)))
2530 nf_conntrack_htable_size
= 65536;
2531 else if (nr_pages
> (1024 * 1024 * 1024 / PAGE_SIZE
))
2532 nf_conntrack_htable_size
= 16384;
2533 if (nf_conntrack_htable_size
< 32)
2534 nf_conntrack_htable_size
= 32;
2536 /* Use a max. factor of four by default to get the same max as
2537 * with the old struct list_heads. When a table size is given
2538 * we use the old value of 8 to avoid reducing the max.
2543 nf_conntrack_hash
= nf_ct_alloc_hashtable(&nf_conntrack_htable_size
, 1);
2544 if (!nf_conntrack_hash
)
2547 nf_conntrack_max
= max_factor
* nf_conntrack_htable_size
;
2549 nf_conntrack_cachep
= kmem_cache_create("nf_conntrack",
2550 sizeof(struct nf_conn
),
2552 SLAB_TYPESAFE_BY_RCU
| SLAB_HWCACHE_ALIGN
, NULL
);
2553 if (!nf_conntrack_cachep
)
2556 ret
= nf_conntrack_expect_init();
2560 ret
= nf_conntrack_acct_init();
2564 ret
= nf_conntrack_tstamp_init();
2568 ret
= nf_conntrack_ecache_init();
2572 ret
= nf_conntrack_timeout_init();
2576 ret
= nf_conntrack_helper_init();
2580 ret
= nf_conntrack_labels_init();
2584 ret
= nf_conntrack_seqadj_init();
2588 ret
= nf_conntrack_proto_init();
2592 conntrack_gc_work_init(&conntrack_gc_work
);
2593 queue_delayed_work(system_power_efficient_wq
, &conntrack_gc_work
.dwork
, HZ
);
2598 nf_conntrack_seqadj_fini();
2600 nf_conntrack_labels_fini();
2602 nf_conntrack_helper_fini();
2604 nf_conntrack_timeout_fini();
2606 nf_conntrack_ecache_fini();
2608 nf_conntrack_tstamp_fini();
2610 nf_conntrack_acct_fini();
2612 nf_conntrack_expect_fini();
2614 kmem_cache_destroy(nf_conntrack_cachep
);
2616 kvfree(nf_conntrack_hash
);
2620 static struct nf_ct_hook nf_conntrack_hook
= {
2621 .update
= nf_conntrack_update
,
2622 .destroy
= destroy_conntrack
,
2623 .get_tuple_skb
= nf_conntrack_get_tuple_skb
,
2626 void nf_conntrack_init_end(void)
2628 /* For use by REJECT target */
2629 RCU_INIT_POINTER(ip_ct_attach
, nf_conntrack_attach
);
2630 RCU_INIT_POINTER(nf_ct_hook
, &nf_conntrack_hook
);
2634 * We need to use special "null" values, not used in hash table
2636 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2637 #define DYING_NULLS_VAL ((1<<30)+1)
2639 int nf_conntrack_init_net(struct net
*net
)
2644 BUILD_BUG_ON(IP_CT_UNTRACKED
== IP_CT_NUMBER
);
2645 BUILD_BUG_ON_NOT_POWER_OF_2(CONNTRACK_LOCKS
);
2646 atomic_set(&net
->ct
.count
, 0);
2648 net
->ct
.pcpu_lists
= alloc_percpu(struct ct_pcpu
);
2649 if (!net
->ct
.pcpu_lists
)
2652 for_each_possible_cpu(cpu
) {
2653 struct ct_pcpu
*pcpu
= per_cpu_ptr(net
->ct
.pcpu_lists
, cpu
);
2655 spin_lock_init(&pcpu
->lock
);
2656 INIT_HLIST_NULLS_HEAD(&pcpu
->unconfirmed
, UNCONFIRMED_NULLS_VAL
);
2657 INIT_HLIST_NULLS_HEAD(&pcpu
->dying
, DYING_NULLS_VAL
);
2660 net
->ct
.stat
= alloc_percpu(struct ip_conntrack_stat
);
2662 goto err_pcpu_lists
;
2664 ret
= nf_conntrack_expect_pernet_init(net
);
2668 nf_conntrack_acct_pernet_init(net
);
2669 nf_conntrack_tstamp_pernet_init(net
);
2670 nf_conntrack_ecache_pernet_init(net
);
2671 nf_conntrack_helper_pernet_init(net
);
2672 nf_conntrack_proto_pernet_init(net
);
2677 free_percpu(net
->ct
.stat
);
2679 free_percpu(net
->ct
.pcpu_lists
);