1 /* Connection state tracking for netfilter. This is separated from,
2 but required by, the NAT layer; it can also be used by an iptables
5 /* (C) 1999-2001 Paul `Rusty' Russell
6 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
7 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
8 * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/types.h>
18 #include <linux/netfilter.h>
19 #include <linux/module.h>
20 #include <linux/sched.h>
21 #include <linux/skbuff.h>
22 #include <linux/proc_fs.h>
23 #include <linux/vmalloc.h>
24 #include <linux/stddef.h>
25 #include <linux/slab.h>
26 #include <linux/random.h>
27 #include <linux/jhash.h>
28 #include <linux/err.h>
29 #include <linux/percpu.h>
30 #include <linux/moduleparam.h>
31 #include <linux/notifier.h>
32 #include <linux/kernel.h>
33 #include <linux/netdevice.h>
34 #include <linux/socket.h>
36 #include <linux/nsproxy.h>
37 #include <linux/rculist_nulls.h>
39 #include <net/netfilter/nf_conntrack.h>
40 #include <net/netfilter/nf_conntrack_l4proto.h>
41 #include <net/netfilter/nf_conntrack_expect.h>
42 #include <net/netfilter/nf_conntrack_helper.h>
43 #include <net/netfilter/nf_conntrack_seqadj.h>
44 #include <net/netfilter/nf_conntrack_core.h>
45 #include <net/netfilter/nf_conntrack_extend.h>
46 #include <net/netfilter/nf_conntrack_acct.h>
47 #include <net/netfilter/nf_conntrack_ecache.h>
48 #include <net/netfilter/nf_conntrack_zones.h>
49 #include <net/netfilter/nf_conntrack_timestamp.h>
50 #include <net/netfilter/nf_conntrack_timeout.h>
51 #include <net/netfilter/nf_conntrack_labels.h>
52 #include <net/netfilter/nf_conntrack_synproxy.h>
53 #include <net/netfilter/nf_nat.h>
54 #include <net/netfilter/nf_nat_core.h>
55 #include <net/netfilter/nf_nat_helper.h>
56 #include <net/netns/hash.h>
59 #include "nf_internals.h"
61 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
62 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
64 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
65 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
67 struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
68 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
70 struct conntrack_gc_work {
71 struct delayed_work dwork;
78 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
79 static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
80 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
81 static __read_mostly bool nf_conntrack_locks_all;
83 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
84 #define GC_MAX_BUCKETS_DIV 128u
85 /* upper bound of full table scan */
86 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
87 /* desired ratio of entries found to be expired */
88 #define GC_EVICT_RATIO 50u
90 static struct conntrack_gc_work conntrack_gc_work;
92 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
94 /* 1) Acquire the lock */
97 /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
98 * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
100 if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
103 /* fast path failed, unlock */
106 /* Slow path 1) get global lock */
107 spin_lock(&nf_conntrack_locks_all_lock);
109 /* Slow path 2) get the lock we want */
112 /* Slow path 3) release the global lock */
113 spin_unlock(&nf_conntrack_locks_all_lock);
115 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
117 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
119 h1 %= CONNTRACK_LOCKS;
120 h2 %= CONNTRACK_LOCKS;
121 spin_unlock(&nf_conntrack_locks[h1]);
123 spin_unlock(&nf_conntrack_locks[h2]);
126 /* return true if we need to recompute hashes (in case hash table was resized) */
127 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
128 unsigned int h2, unsigned int sequence)
130 h1 %= CONNTRACK_LOCKS;
131 h2 %= CONNTRACK_LOCKS;
133 nf_conntrack_lock(&nf_conntrack_locks[h1]);
135 spin_lock_nested(&nf_conntrack_locks[h2],
136 SINGLE_DEPTH_NESTING);
138 nf_conntrack_lock(&nf_conntrack_locks[h2]);
139 spin_lock_nested(&nf_conntrack_locks[h1],
140 SINGLE_DEPTH_NESTING);
142 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
143 nf_conntrack_double_unlock(h1, h2);
149 static void nf_conntrack_all_lock(void)
153 spin_lock(&nf_conntrack_locks_all_lock);
155 nf_conntrack_locks_all = true;
157 for (i = 0; i < CONNTRACK_LOCKS; i++) {
158 spin_lock(&nf_conntrack_locks[i]);
160 /* This spin_unlock provides the "release" to ensure that
161 * nf_conntrack_locks_all==true is visible to everyone that
162 * acquired spin_lock(&nf_conntrack_locks[]).
164 spin_unlock(&nf_conntrack_locks[i]);
168 static void nf_conntrack_all_unlock(void)
170 /* All prior stores must be complete before we clear
171 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
172 * might observe the false value but not the entire
174 * It pairs with the smp_load_acquire() in nf_conntrack_lock()
176 smp_store_release(&nf_conntrack_locks_all, false);
177 spin_unlock(&nf_conntrack_locks_all_lock);
180 unsigned int nf_conntrack_htable_size __read_mostly;
181 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
183 unsigned int nf_conntrack_max __read_mostly;
184 EXPORT_SYMBOL_GPL(nf_conntrack_max);
185 seqcount_t nf_conntrack_generation __read_mostly;
186 static unsigned int nf_conntrack_hash_rnd __read_mostly;
188 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
189 const struct net *net)
194 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
196 /* The direction must be ignored, so we hash everything up to the
197 * destination ports (which is a multiple of 4) and treat the last
198 * three bytes manually.
200 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
201 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
202 return jhash2((u32 *)tuple, n, seed ^
203 (((__force __u16)tuple->dst.u.all << 16) |
204 tuple->dst.protonum));
207 static u32 scale_hash(u32 hash)
209 return reciprocal_scale(hash, nf_conntrack_htable_size);
212 static u32 __hash_conntrack(const struct net *net,
213 const struct nf_conntrack_tuple *tuple,
216 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
219 static u32 hash_conntrack(const struct net *net,
220 const struct nf_conntrack_tuple *tuple)
222 return scale_hash(hash_conntrack_raw(tuple, net));
226 nf_ct_get_tuple(const struct sk_buff *skb,
228 unsigned int dataoff,
232 struct nf_conntrack_tuple *tuple,
233 const struct nf_conntrack_l4proto *l4proto)
241 } _inet_hdr, *inet_hdr;
243 memset(tuple, 0, sizeof(*tuple));
245 tuple->src.l3num = l3num;
248 nhoff += offsetof(struct iphdr, saddr);
249 size = 2 * sizeof(__be32);
252 nhoff += offsetof(struct ipv6hdr, saddr);
253 size = sizeof(_addrs);
259 ap = skb_header_pointer(skb, nhoff, size, _addrs);
265 tuple->src.u3.ip = ap[0];
266 tuple->dst.u3.ip = ap[1];
269 memcpy(tuple->src.u3.ip6, ap, sizeof(tuple->src.u3.ip6));
270 memcpy(tuple->dst.u3.ip6, ap + 4, sizeof(tuple->dst.u3.ip6));
274 tuple->dst.protonum = protonum;
275 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
277 if (unlikely(l4proto->pkt_to_tuple))
278 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
280 /* Actually only need first 4 bytes to get ports. */
281 inet_hdr = skb_header_pointer(skb, dataoff, sizeof(_inet_hdr), &_inet_hdr);
285 tuple->src.u.udp.port = inet_hdr->sport;
286 tuple->dst.u.udp.port = inet_hdr->dport;
290 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
294 const struct iphdr *iph;
297 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
301 /* Conntrack defragments packets, we might still see fragments
302 * inside ICMP packets though.
304 if (iph->frag_off & htons(IP_OFFSET))
307 dataoff = nhoff + (iph->ihl << 2);
308 *protonum = iph->protocol;
310 /* Check bogus IP headers */
311 if (dataoff > skb->len) {
312 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
313 nhoff, iph->ihl << 2, skb->len);
319 #if IS_ENABLED(CONFIG_IPV6)
320 static int ipv6_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
324 unsigned int extoff = nhoff + sizeof(struct ipv6hdr);
328 if (skb_copy_bits(skb, nhoff + offsetof(struct ipv6hdr, nexthdr),
329 &nexthdr, sizeof(nexthdr)) != 0) {
330 pr_debug("can't get nexthdr\n");
333 protoff = ipv6_skip_exthdr(skb, extoff, &nexthdr, &frag_off);
335 * (protoff == skb->len) means the packet has not data, just
336 * IPv6 and possibly extensions headers, but it is tracked anyway
338 if (protoff < 0 || (frag_off & htons(~0x7)) != 0) {
339 pr_debug("can't find proto in pkt\n");
348 static int get_l4proto(const struct sk_buff *skb,
349 unsigned int nhoff, u8 pf, u8 *l4num)
353 return ipv4_get_l4proto(skb, nhoff, l4num);
354 #if IS_ENABLED(CONFIG_IPV6)
356 return ipv6_get_l4proto(skb, nhoff, l4num);
365 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
367 struct net *net, struct nf_conntrack_tuple *tuple)
369 const struct nf_conntrack_l4proto *l4proto;
376 protoff = get_l4proto(skb, nhoff, l3num, &protonum);
382 l4proto = __nf_ct_l4proto_find(protonum);
384 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
390 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
393 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
394 const struct nf_conntrack_tuple *orig,
395 const struct nf_conntrack_l4proto *l4proto)
397 memset(inverse, 0, sizeof(*inverse));
399 inverse->src.l3num = orig->src.l3num;
401 switch (orig->src.l3num) {
403 inverse->src.u3.ip = orig->dst.u3.ip;
404 inverse->dst.u3.ip = orig->src.u3.ip;
407 inverse->src.u3.in6 = orig->dst.u3.in6;
408 inverse->dst.u3.in6 = orig->src.u3.in6;
414 inverse->dst.dir = !orig->dst.dir;
416 inverse->dst.protonum = orig->dst.protonum;
418 if (unlikely(l4proto->invert_tuple))
419 return l4proto->invert_tuple(inverse, orig);
421 inverse->src.u.all = orig->dst.u.all;
422 inverse->dst.u.all = orig->src.u.all;
425 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
428 clean_from_lists(struct nf_conn *ct)
430 pr_debug("clean_from_lists(%p)\n", ct);
431 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
432 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
434 /* Destroy all pending expectations */
435 nf_ct_remove_expectations(ct);
438 /* must be called with local_bh_disable */
439 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
441 struct ct_pcpu *pcpu;
443 /* add this conntrack to the (per cpu) dying list */
444 ct->cpu = smp_processor_id();
445 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
447 spin_lock(&pcpu->lock);
448 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
450 spin_unlock(&pcpu->lock);
453 /* must be called with local_bh_disable */
454 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
456 struct ct_pcpu *pcpu;
458 /* add this conntrack to the (per cpu) unconfirmed list */
459 ct->cpu = smp_processor_id();
460 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
462 spin_lock(&pcpu->lock);
463 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
465 spin_unlock(&pcpu->lock);
468 /* must be called with local_bh_disable */
469 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
471 struct ct_pcpu *pcpu;
473 /* We overload first tuple to link into unconfirmed or dying list.*/
474 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
476 spin_lock(&pcpu->lock);
477 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
478 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
479 spin_unlock(&pcpu->lock);
482 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
484 /* Released via destroy_conntrack() */
485 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
486 const struct nf_conntrack_zone *zone,
489 struct nf_conn *tmpl, *p;
491 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
492 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
497 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
499 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
500 tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
503 tmpl = kzalloc(sizeof(*tmpl), flags);
508 tmpl->status = IPS_TEMPLATE;
509 write_pnet(&tmpl->ct_net, net);
510 nf_ct_zone_add(tmpl, zone);
511 atomic_set(&tmpl->ct_general.use, 0);
515 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
517 void nf_ct_tmpl_free(struct nf_conn *tmpl)
519 nf_ct_ext_destroy(tmpl);
520 nf_ct_ext_free(tmpl);
522 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
523 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
527 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
530 destroy_conntrack(struct nf_conntrack *nfct)
532 struct nf_conn *ct = (struct nf_conn *)nfct;
533 const struct nf_conntrack_l4proto *l4proto;
535 pr_debug("destroy_conntrack(%p)\n", ct);
536 WARN_ON(atomic_read(&nfct->use) != 0);
538 if (unlikely(nf_ct_is_template(ct))) {
542 l4proto = __nf_ct_l4proto_find(nf_ct_protonum(ct));
543 if (l4proto->destroy)
544 l4proto->destroy(ct);
547 /* Expectations will have been removed in clean_from_lists,
548 * except TFTP can create an expectation on the first packet,
549 * before connection is in the list, so we need to clean here,
552 nf_ct_remove_expectations(ct);
554 nf_ct_del_from_dying_or_unconfirmed_list(ct);
559 nf_ct_put(ct->master);
561 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
562 nf_conntrack_free(ct);
565 static void nf_ct_delete_from_lists(struct nf_conn *ct)
567 struct net *net = nf_ct_net(ct);
568 unsigned int hash, reply_hash;
569 unsigned int sequence;
571 nf_ct_helper_destroy(ct);
575 sequence = read_seqcount_begin(&nf_conntrack_generation);
576 hash = hash_conntrack(net,
577 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
578 reply_hash = hash_conntrack(net,
579 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
580 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
582 clean_from_lists(ct);
583 nf_conntrack_double_unlock(hash, reply_hash);
585 nf_ct_add_to_dying_list(ct);
590 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
592 struct nf_conn_tstamp *tstamp;
594 if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
597 tstamp = nf_conn_tstamp_find(ct);
598 if (tstamp && tstamp->stop == 0)
599 tstamp->stop = ktime_get_real_ns();
601 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
602 portid, report) < 0) {
603 /* destroy event was not delivered. nf_ct_put will
604 * be done by event cache worker on redelivery.
606 nf_ct_delete_from_lists(ct);
607 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
611 nf_conntrack_ecache_work(nf_ct_net(ct));
612 nf_ct_delete_from_lists(ct);
616 EXPORT_SYMBOL_GPL(nf_ct_delete);
619 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
620 const struct nf_conntrack_tuple *tuple,
621 const struct nf_conntrack_zone *zone,
622 const struct net *net)
624 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
626 /* A conntrack can be recreated with the equal tuple,
627 * so we need to check that the conntrack is confirmed
629 return nf_ct_tuple_equal(tuple, &h->tuple) &&
630 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
631 nf_ct_is_confirmed(ct) &&
632 net_eq(net, nf_ct_net(ct));
636 nf_ct_match(const struct nf_conn *ct1, const struct nf_conn *ct2)
638 return nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
639 &ct2->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
640 nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_REPLY].tuple,
641 &ct2->tuplehash[IP_CT_DIR_REPLY].tuple) &&
642 nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_ORIGINAL) &&
643 nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_REPLY) &&
644 net_eq(nf_ct_net(ct1), nf_ct_net(ct2));
647 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
648 static void nf_ct_gc_expired(struct nf_conn *ct)
650 if (!atomic_inc_not_zero(&ct->ct_general.use))
653 if (nf_ct_should_gc(ct))
661 * - Caller must take a reference on returned object
662 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
664 static struct nf_conntrack_tuple_hash *
665 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
666 const struct nf_conntrack_tuple *tuple, u32 hash)
668 struct nf_conntrack_tuple_hash *h;
669 struct hlist_nulls_head *ct_hash;
670 struct hlist_nulls_node *n;
671 unsigned int bucket, hsize;
674 nf_conntrack_get_ht(&ct_hash, &hsize);
675 bucket = reciprocal_scale(hash, hsize);
677 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
680 ct = nf_ct_tuplehash_to_ctrack(h);
681 if (nf_ct_is_expired(ct)) {
682 nf_ct_gc_expired(ct);
686 if (nf_ct_is_dying(ct))
689 if (nf_ct_key_equal(h, tuple, zone, net))
693 * if the nulls value we got at the end of this lookup is
694 * not the expected one, we must restart lookup.
695 * We probably met an item that was moved to another chain.
697 if (get_nulls_value(n) != bucket) {
698 NF_CT_STAT_INC_ATOMIC(net, search_restart);
705 /* Find a connection corresponding to a tuple. */
706 static struct nf_conntrack_tuple_hash *
707 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
708 const struct nf_conntrack_tuple *tuple, u32 hash)
710 struct nf_conntrack_tuple_hash *h;
715 h = ____nf_conntrack_find(net, zone, tuple, hash);
717 ct = nf_ct_tuplehash_to_ctrack(h);
718 if (unlikely(nf_ct_is_dying(ct) ||
719 !atomic_inc_not_zero(&ct->ct_general.use)))
722 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
733 struct nf_conntrack_tuple_hash *
734 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
735 const struct nf_conntrack_tuple *tuple)
737 return __nf_conntrack_find_get(net, zone, tuple,
738 hash_conntrack_raw(tuple, net));
740 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
742 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
744 unsigned int reply_hash)
746 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
747 &nf_conntrack_hash[hash]);
748 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
749 &nf_conntrack_hash[reply_hash]);
753 nf_conntrack_hash_check_insert(struct nf_conn *ct)
755 const struct nf_conntrack_zone *zone;
756 struct net *net = nf_ct_net(ct);
757 unsigned int hash, reply_hash;
758 struct nf_conntrack_tuple_hash *h;
759 struct hlist_nulls_node *n;
760 unsigned int sequence;
762 zone = nf_ct_zone(ct);
766 sequence = read_seqcount_begin(&nf_conntrack_generation);
767 hash = hash_conntrack(net,
768 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
769 reply_hash = hash_conntrack(net,
770 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
771 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
773 /* See if there's one in the list already, including reverse */
774 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
775 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
779 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
780 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
785 /* The caller holds a reference to this object */
786 atomic_set(&ct->ct_general.use, 2);
787 __nf_conntrack_hash_insert(ct, hash, reply_hash);
788 nf_conntrack_double_unlock(hash, reply_hash);
789 NF_CT_STAT_INC(net, insert);
794 nf_conntrack_double_unlock(hash, reply_hash);
795 NF_CT_STAT_INC(net, insert_failed);
799 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
801 static inline void nf_ct_acct_update(struct nf_conn *ct,
802 enum ip_conntrack_info ctinfo,
805 struct nf_conn_acct *acct;
807 acct = nf_conn_acct_find(ct);
809 struct nf_conn_counter *counter = acct->counter;
811 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
812 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
816 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
817 const struct nf_conn *loser_ct)
819 struct nf_conn_acct *acct;
821 acct = nf_conn_acct_find(loser_ct);
823 struct nf_conn_counter *counter = acct->counter;
826 /* u32 should be fine since we must have seen one packet. */
827 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
828 nf_ct_acct_update(ct, ctinfo, bytes);
832 /* Resolve race on insertion if this protocol allows this. */
833 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
834 enum ip_conntrack_info ctinfo,
835 struct nf_conntrack_tuple_hash *h)
837 /* This is the conntrack entry already in hashes that won race. */
838 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
839 const struct nf_conntrack_l4proto *l4proto;
840 enum ip_conntrack_info oldinfo;
841 struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo);
843 l4proto = __nf_ct_l4proto_find(nf_ct_protonum(ct));
844 if (l4proto->allow_clash &&
845 !nf_ct_is_dying(ct) &&
846 atomic_inc_not_zero(&ct->ct_general.use)) {
847 if (((ct->status & IPS_NAT_DONE_MASK) == 0) ||
848 nf_ct_match(ct, loser_ct)) {
849 nf_ct_acct_merge(ct, ctinfo, loser_ct);
850 nf_conntrack_put(&loser_ct->ct_general);
851 nf_ct_set(skb, ct, oldinfo);
856 NF_CT_STAT_INC(net, drop);
860 /* Confirm a connection given skb; places it in hash table */
862 __nf_conntrack_confirm(struct sk_buff *skb)
864 const struct nf_conntrack_zone *zone;
865 unsigned int hash, reply_hash;
866 struct nf_conntrack_tuple_hash *h;
868 struct nf_conn_help *help;
869 struct nf_conn_tstamp *tstamp;
870 struct hlist_nulls_node *n;
871 enum ip_conntrack_info ctinfo;
873 unsigned int sequence;
876 ct = nf_ct_get(skb, &ctinfo);
879 /* ipt_REJECT uses nf_conntrack_attach to attach related
880 ICMP/TCP RST packets in other direction. Actual packet
881 which created connection will be IP_CT_NEW or for an
882 expected connection, IP_CT_RELATED. */
883 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
886 zone = nf_ct_zone(ct);
890 sequence = read_seqcount_begin(&nf_conntrack_generation);
891 /* reuse the hash saved before */
892 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
893 hash = scale_hash(hash);
894 reply_hash = hash_conntrack(net,
895 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
897 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
899 /* We're not in hash table, and we refuse to set up related
900 * connections for unconfirmed conns. But packet copies and
901 * REJECT will give spurious warnings here.
904 /* No external references means no one else could have
907 WARN_ON(nf_ct_is_confirmed(ct));
908 pr_debug("Confirming conntrack %p\n", ct);
909 /* We have to check the DYING flag after unlink to prevent
910 * a race against nf_ct_get_next_corpse() possibly called from
911 * user context, else we insert an already 'dead' hash, blocking
912 * further use of that particular connection -JM.
914 nf_ct_del_from_dying_or_unconfirmed_list(ct);
916 if (unlikely(nf_ct_is_dying(ct))) {
917 nf_ct_add_to_dying_list(ct);
921 /* See if there's one in the list already, including reverse:
922 NAT could have grabbed it without realizing, since we're
923 not in the hash. If there is, we lost race. */
924 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
925 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
929 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
930 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
934 /* Timer relative to confirmation time, not original
935 setting time, otherwise we'd get timer wrap in
936 weird delay cases. */
937 ct->timeout += nfct_time_stamp;
938 atomic_inc(&ct->ct_general.use);
939 ct->status |= IPS_CONFIRMED;
941 /* set conntrack timestamp, if enabled. */
942 tstamp = nf_conn_tstamp_find(ct);
944 if (skb->tstamp == 0)
945 __net_timestamp(skb);
947 tstamp->start = ktime_to_ns(skb->tstamp);
949 /* Since the lookup is lockless, hash insertion must be done after
950 * starting the timer and setting the CONFIRMED bit. The RCU barriers
951 * guarantee that no other CPU can find the conntrack before the above
952 * stores are visible.
954 __nf_conntrack_hash_insert(ct, hash, reply_hash);
955 nf_conntrack_double_unlock(hash, reply_hash);
958 help = nfct_help(ct);
959 if (help && help->helper)
960 nf_conntrack_event_cache(IPCT_HELPER, ct);
962 nf_conntrack_event_cache(master_ct(ct) ?
963 IPCT_RELATED : IPCT_NEW, ct);
967 nf_ct_add_to_dying_list(ct);
968 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
970 nf_conntrack_double_unlock(hash, reply_hash);
971 NF_CT_STAT_INC(net, insert_failed);
975 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
977 /* Returns true if a connection correspondings to the tuple (required
980 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
981 const struct nf_conn *ignored_conntrack)
983 struct net *net = nf_ct_net(ignored_conntrack);
984 const struct nf_conntrack_zone *zone;
985 struct nf_conntrack_tuple_hash *h;
986 struct hlist_nulls_head *ct_hash;
987 unsigned int hash, hsize;
988 struct hlist_nulls_node *n;
991 zone = nf_ct_zone(ignored_conntrack);
995 nf_conntrack_get_ht(&ct_hash, &hsize);
996 hash = __hash_conntrack(net, tuple, hsize);
998 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
999 ct = nf_ct_tuplehash_to_ctrack(h);
1001 if (ct == ignored_conntrack)
1004 if (nf_ct_is_expired(ct)) {
1005 nf_ct_gc_expired(ct);
1009 if (nf_ct_key_equal(h, tuple, zone, net)) {
1010 /* Tuple is taken already, so caller will need to find
1011 * a new source port to use.
1014 * If the *original tuples* are identical, then both
1015 * conntracks refer to the same flow.
1016 * This is a rare situation, it can occur e.g. when
1017 * more than one UDP packet is sent from same socket
1018 * in different threads.
1020 * Let nf_ct_resolve_clash() deal with this later.
1022 if (nf_ct_tuple_equal(&ignored_conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1023 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple))
1026 NF_CT_STAT_INC_ATOMIC(net, found);
1032 if (get_nulls_value(n) != hash) {
1033 NF_CT_STAT_INC_ATOMIC(net, search_restart);
1041 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
1043 #define NF_CT_EVICTION_RANGE 8
1045 /* There's a small race here where we may free a just-assured
1046 connection. Too bad: we're in trouble anyway. */
1047 static unsigned int early_drop_list(struct net *net,
1048 struct hlist_nulls_head *head)
1050 struct nf_conntrack_tuple_hash *h;
1051 struct hlist_nulls_node *n;
1052 unsigned int drops = 0;
1053 struct nf_conn *tmp;
1055 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
1056 tmp = nf_ct_tuplehash_to_ctrack(h);
1058 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status))
1061 if (nf_ct_is_expired(tmp)) {
1062 nf_ct_gc_expired(tmp);
1066 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
1067 !net_eq(nf_ct_net(tmp), net) ||
1068 nf_ct_is_dying(tmp))
1071 if (!atomic_inc_not_zero(&tmp->ct_general.use))
1074 /* kill only if still in same netns -- might have moved due to
1075 * SLAB_TYPESAFE_BY_RCU rules.
1077 * We steal the timer reference. If that fails timer has
1078 * already fired or someone else deleted it. Just drop ref
1079 * and move to next entry.
1081 if (net_eq(nf_ct_net(tmp), net) &&
1082 nf_ct_is_confirmed(tmp) &&
1083 nf_ct_delete(tmp, 0, 0))
1092 static noinline int early_drop(struct net *net, unsigned int hash)
1094 unsigned int i, bucket;
1096 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
1097 struct hlist_nulls_head *ct_hash;
1098 unsigned int hsize, drops;
1101 nf_conntrack_get_ht(&ct_hash, &hsize);
1103 bucket = reciprocal_scale(hash, hsize);
1105 bucket = (bucket + 1) % hsize;
1107 drops = early_drop_list(net, &ct_hash[bucket]);
1111 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
1119 static bool gc_worker_skip_ct(const struct nf_conn *ct)
1121 return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct);
1124 static bool gc_worker_can_early_drop(const struct nf_conn *ct)
1126 const struct nf_conntrack_l4proto *l4proto;
1128 if (!test_bit(IPS_ASSURED_BIT, &ct->status))
1131 l4proto = __nf_ct_l4proto_find(nf_ct_protonum(ct));
1132 if (l4proto->can_early_drop && l4proto->can_early_drop(ct))
1138 #define DAY (86400 * HZ)
1140 /* Set an arbitrary timeout large enough not to ever expire, this save
1141 * us a check for the IPS_OFFLOAD_BIT from the packet path via
1142 * nf_ct_is_expired().
1144 static void nf_ct_offload_timeout(struct nf_conn *ct)
1146 if (nf_ct_expires(ct) < DAY / 2)
1147 ct->timeout = nfct_time_stamp + DAY;
1150 static void gc_worker(struct work_struct *work)
1152 unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
1153 unsigned int i, goal, buckets = 0, expired_count = 0;
1154 unsigned int nf_conntrack_max95 = 0;
1155 struct conntrack_gc_work *gc_work;
1156 unsigned int ratio, scanned = 0;
1157 unsigned long next_run;
1159 gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
1161 goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
1162 i = gc_work->last_bucket;
1163 if (gc_work->early_drop)
1164 nf_conntrack_max95 = nf_conntrack_max / 100u * 95u;
1167 struct nf_conntrack_tuple_hash *h;
1168 struct hlist_nulls_head *ct_hash;
1169 struct hlist_nulls_node *n;
1170 unsigned int hashsz;
1171 struct nf_conn *tmp;
1176 nf_conntrack_get_ht(&ct_hash, &hashsz);
1180 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
1183 tmp = nf_ct_tuplehash_to_ctrack(h);
1186 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
1187 nf_ct_offload_timeout(tmp);
1191 if (nf_ct_is_expired(tmp)) {
1192 nf_ct_gc_expired(tmp);
1197 if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp))
1200 net = nf_ct_net(tmp);
1201 if (atomic_read(&net->ct.count) < nf_conntrack_max95)
1204 /* need to take reference to avoid possible races */
1205 if (!atomic_inc_not_zero(&tmp->ct_general.use))
1208 if (gc_worker_skip_ct(tmp)) {
1213 if (gc_worker_can_early_drop(tmp))
1219 /* could check get_nulls_value() here and restart if ct
1220 * was moved to another chain. But given gc is best-effort
1221 * we will just continue with next hash slot.
1225 } while (++buckets < goal);
1227 if (gc_work->exiting)
1231 * Eviction will normally happen from the packet path, and not
1232 * from this gc worker.
1234 * This worker is only here to reap expired entries when system went
1235 * idle after a busy period.
1237 * The heuristics below are supposed to balance conflicting goals:
1239 * 1. Minimize time until we notice a stale entry
1240 * 2. Maximize scan intervals to not waste cycles
1242 * Normally, expire ratio will be close to 0.
1244 * As soon as a sizeable fraction of the entries have expired
1245 * increase scan frequency.
1247 ratio = scanned ? expired_count * 100 / scanned : 0;
1248 if (ratio > GC_EVICT_RATIO) {
1249 gc_work->next_gc_run = min_interval;
1251 unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV;
1253 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0);
1255 gc_work->next_gc_run += min_interval;
1256 if (gc_work->next_gc_run > max)
1257 gc_work->next_gc_run = max;
1260 next_run = gc_work->next_gc_run;
1261 gc_work->last_bucket = i;
1262 gc_work->early_drop = false;
1263 queue_delayed_work(system_power_efficient_wq, &gc_work->dwork, next_run);
1266 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1268 INIT_DEFERRABLE_WORK(&gc_work->dwork, gc_worker);
1269 gc_work->next_gc_run = HZ;
1270 gc_work->exiting = false;
1273 static struct nf_conn *
1274 __nf_conntrack_alloc(struct net *net,
1275 const struct nf_conntrack_zone *zone,
1276 const struct nf_conntrack_tuple *orig,
1277 const struct nf_conntrack_tuple *repl,
1278 gfp_t gfp, u32 hash)
1282 /* We don't want any race condition at early drop stage */
1283 atomic_inc(&net->ct.count);
1285 if (nf_conntrack_max &&
1286 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1287 if (!early_drop(net, hash)) {
1288 if (!conntrack_gc_work.early_drop)
1289 conntrack_gc_work.early_drop = true;
1290 atomic_dec(&net->ct.count);
1291 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1292 return ERR_PTR(-ENOMEM);
1297 * Do not use kmem_cache_zalloc(), as this cache uses
1298 * SLAB_TYPESAFE_BY_RCU.
1300 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1304 spin_lock_init(&ct->lock);
1305 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1306 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1307 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1308 /* save hash for reusing when confirming */
1309 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1311 write_pnet(&ct->ct_net, net);
1312 memset(&ct->__nfct_init_offset[0], 0,
1313 offsetof(struct nf_conn, proto) -
1314 offsetof(struct nf_conn, __nfct_init_offset[0]));
1316 nf_ct_zone_add(ct, zone);
1318 /* Because we use RCU lookups, we set ct_general.use to zero before
1319 * this is inserted in any list.
1321 atomic_set(&ct->ct_general.use, 0);
1324 atomic_dec(&net->ct.count);
1325 return ERR_PTR(-ENOMEM);
1328 struct nf_conn *nf_conntrack_alloc(struct net *net,
1329 const struct nf_conntrack_zone *zone,
1330 const struct nf_conntrack_tuple *orig,
1331 const struct nf_conntrack_tuple *repl,
1334 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1336 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1338 void nf_conntrack_free(struct nf_conn *ct)
1340 struct net *net = nf_ct_net(ct);
1342 /* A freed object has refcnt == 0, that's
1343 * the golden rule for SLAB_TYPESAFE_BY_RCU
1345 WARN_ON(atomic_read(&ct->ct_general.use) != 0);
1347 nf_ct_ext_destroy(ct);
1349 kmem_cache_free(nf_conntrack_cachep, ct);
1350 smp_mb__before_atomic();
1351 atomic_dec(&net->ct.count);
1353 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1356 /* Allocate a new conntrack: we return -ENOMEM if classification
1357 failed due to stress. Otherwise it really is unclassifiable. */
1358 static noinline struct nf_conntrack_tuple_hash *
1359 init_conntrack(struct net *net, struct nf_conn *tmpl,
1360 const struct nf_conntrack_tuple *tuple,
1361 const struct nf_conntrack_l4proto *l4proto,
1362 struct sk_buff *skb,
1363 unsigned int dataoff, u32 hash)
1366 struct nf_conn_help *help;
1367 struct nf_conntrack_tuple repl_tuple;
1368 struct nf_conntrack_ecache *ecache;
1369 struct nf_conntrack_expect *exp = NULL;
1370 const struct nf_conntrack_zone *zone;
1371 struct nf_conn_timeout *timeout_ext;
1372 struct nf_conntrack_zone tmp;
1374 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l4proto)) {
1375 pr_debug("Can't invert tuple.\n");
1379 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1380 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1383 return (struct nf_conntrack_tuple_hash *)ct;
1385 if (!nf_ct_add_synproxy(ct, tmpl)) {
1386 nf_conntrack_free(ct);
1387 return ERR_PTR(-ENOMEM);
1390 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1393 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1396 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1397 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1398 nf_ct_labels_ext_add(ct);
1400 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1401 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1402 ecache ? ecache->expmask : 0,
1406 if (net->ct.expect_count) {
1407 spin_lock(&nf_conntrack_expect_lock);
1408 exp = nf_ct_find_expectation(net, zone, tuple);
1410 pr_debug("expectation arrives ct=%p exp=%p\n",
1412 /* Welcome, Mr. Bond. We've been expecting you... */
1413 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1414 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1415 ct->master = exp->master;
1417 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
1419 rcu_assign_pointer(help->helper, exp->helper);
1422 #ifdef CONFIG_NF_CONNTRACK_MARK
1423 ct->mark = exp->master->mark;
1425 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1426 ct->secmark = exp->master->secmark;
1428 NF_CT_STAT_INC(net, expect_new);
1430 spin_unlock(&nf_conntrack_expect_lock);
1433 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1435 /* Now it is inserted into the unconfirmed list, bump refcount */
1436 nf_conntrack_get(&ct->ct_general);
1437 nf_ct_add_to_unconfirmed_list(ct);
1443 exp->expectfn(ct, exp);
1444 nf_ct_expect_put(exp);
1447 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1450 /* On success, returns 0, sets skb->_nfct | ctinfo */
1452 resolve_normal_ct(struct nf_conn *tmpl,
1453 struct sk_buff *skb,
1454 unsigned int dataoff,
1456 const struct nf_conntrack_l4proto *l4proto,
1457 const struct nf_hook_state *state)
1459 const struct nf_conntrack_zone *zone;
1460 struct nf_conntrack_tuple tuple;
1461 struct nf_conntrack_tuple_hash *h;
1462 enum ip_conntrack_info ctinfo;
1463 struct nf_conntrack_zone tmp;
1467 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1468 dataoff, state->pf, protonum, state->net,
1470 pr_debug("Can't get tuple\n");
1474 /* look for tuple match */
1475 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1476 hash = hash_conntrack_raw(&tuple, state->net);
1477 h = __nf_conntrack_find_get(state->net, zone, &tuple, hash);
1479 h = init_conntrack(state->net, tmpl, &tuple, l4proto,
1480 skb, dataoff, hash);
1486 ct = nf_ct_tuplehash_to_ctrack(h);
1488 /* It exists; we have (non-exclusive) reference. */
1489 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1490 ctinfo = IP_CT_ESTABLISHED_REPLY;
1492 /* Once we've had two way comms, always ESTABLISHED. */
1493 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1494 pr_debug("normal packet for %p\n", ct);
1495 ctinfo = IP_CT_ESTABLISHED;
1496 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1497 pr_debug("related packet for %p\n", ct);
1498 ctinfo = IP_CT_RELATED;
1500 pr_debug("new packet for %p\n", ct);
1504 nf_ct_set(skb, ct, ctinfo);
1509 * icmp packets need special treatment to handle error messages that are
1510 * related to a connection.
1512 * Callers need to check if skb has a conntrack assigned when this
1513 * helper returns; in such case skb belongs to an already known connection.
1515 static unsigned int __cold
1516 nf_conntrack_handle_icmp(struct nf_conn *tmpl,
1517 struct sk_buff *skb,
1518 unsigned int dataoff,
1520 const struct nf_hook_state *state)
1524 if (state->pf == NFPROTO_IPV4 && protonum == IPPROTO_ICMP)
1525 ret = nf_conntrack_icmpv4_error(tmpl, skb, dataoff, state);
1526 #if IS_ENABLED(CONFIG_IPV6)
1527 else if (state->pf == NFPROTO_IPV6 && protonum == IPPROTO_ICMPV6)
1528 ret = nf_conntrack_icmpv6_error(tmpl, skb, dataoff, state);
1534 NF_CT_STAT_INC_ATOMIC(state->net, error);
1535 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1542 nf_conntrack_in(struct sk_buff *skb, const struct nf_hook_state *state)
1544 const struct nf_conntrack_l4proto *l4proto;
1545 enum ip_conntrack_info ctinfo;
1546 struct nf_conn *ct, *tmpl;
1550 tmpl = nf_ct_get(skb, &ctinfo);
1551 if (tmpl || ctinfo == IP_CT_UNTRACKED) {
1552 /* Previously seen (loopback or untracked)? Ignore. */
1553 if ((tmpl && !nf_ct_is_template(tmpl)) ||
1554 ctinfo == IP_CT_UNTRACKED) {
1555 NF_CT_STAT_INC_ATOMIC(state->net, ignore);
1561 /* rcu_read_lock()ed by nf_hook_thresh */
1562 dataoff = get_l4proto(skb, skb_network_offset(skb), state->pf, &protonum);
1564 pr_debug("not prepared to track yet or error occurred\n");
1565 NF_CT_STAT_INC_ATOMIC(state->net, error);
1566 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1571 l4proto = __nf_ct_l4proto_find(protonum);
1573 if (protonum == IPPROTO_ICMP || protonum == IPPROTO_ICMPV6) {
1574 ret = nf_conntrack_handle_icmp(tmpl, skb, dataoff,
1580 /* ICMP[v6] protocol trackers may assign one conntrack. */
1585 ret = resolve_normal_ct(tmpl, skb, dataoff,
1586 protonum, l4proto, state);
1588 /* Too stressed to deal. */
1589 NF_CT_STAT_INC_ATOMIC(state->net, drop);
1594 ct = nf_ct_get(skb, &ctinfo);
1596 /* Not valid part of a connection */
1597 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1602 ret = l4proto->packet(ct, skb, dataoff, ctinfo, state);
1604 /* Invalid: inverse of the return code tells
1605 * the netfilter core what to do */
1606 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1607 nf_conntrack_put(&ct->ct_general);
1609 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1610 if (ret == -NF_DROP)
1611 NF_CT_STAT_INC_ATOMIC(state->net, drop);
1612 /* Special case: TCP tracker reports an attempt to reopen a
1613 * closed/aborted connection. We have to go back and create a
1616 if (ret == -NF_REPEAT)
1622 if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
1623 !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1624 nf_conntrack_event_cache(IPCT_REPLY, ct);
1631 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1633 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1634 const struct nf_conntrack_tuple *orig)
1639 ret = nf_ct_invert_tuple(inverse, orig,
1640 __nf_ct_l4proto_find(orig->dst.protonum));
1644 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1646 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1647 implicitly racy: see __nf_conntrack_confirm */
1648 void nf_conntrack_alter_reply(struct nf_conn *ct,
1649 const struct nf_conntrack_tuple *newreply)
1651 struct nf_conn_help *help = nfct_help(ct);
1653 /* Should be unconfirmed, so not in hash table yet */
1654 WARN_ON(nf_ct_is_confirmed(ct));
1656 pr_debug("Altering reply tuple of %p to ", ct);
1657 nf_ct_dump_tuple(newreply);
1659 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1660 if (ct->master || (help && !hlist_empty(&help->expectations)))
1664 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1667 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1669 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1670 void __nf_ct_refresh_acct(struct nf_conn *ct,
1671 enum ip_conntrack_info ctinfo,
1672 const struct sk_buff *skb,
1673 unsigned long extra_jiffies,
1678 /* Only update if this is not a fixed timeout */
1679 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1682 /* If not in hash table, timer will not be active yet */
1683 if (nf_ct_is_confirmed(ct))
1684 extra_jiffies += nfct_time_stamp;
1686 ct->timeout = extra_jiffies;
1689 nf_ct_acct_update(ct, ctinfo, skb->len);
1691 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1693 bool nf_ct_kill_acct(struct nf_conn *ct,
1694 enum ip_conntrack_info ctinfo,
1695 const struct sk_buff *skb)
1697 nf_ct_acct_update(ct, ctinfo, skb->len);
1699 return nf_ct_delete(ct, 0, 0);
1701 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1703 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1705 #include <linux/netfilter/nfnetlink.h>
1706 #include <linux/netfilter/nfnetlink_conntrack.h>
1707 #include <linux/mutex.h>
1709 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1710 * in ip_conntrack_core, since we don't want the protocols to autoload
1711 * or depend on ctnetlink */
1712 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1713 const struct nf_conntrack_tuple *tuple)
1715 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1716 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1717 goto nla_put_failure;
1723 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1725 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1726 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1727 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1729 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1731 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1732 struct nf_conntrack_tuple *t)
1734 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1737 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1738 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1742 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1744 unsigned int nf_ct_port_nlattr_tuple_size(void)
1746 static unsigned int size __read_mostly;
1749 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1753 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1756 /* Used by ipt_REJECT and ip6t_REJECT. */
1757 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1760 enum ip_conntrack_info ctinfo;
1762 /* This ICMP is in reverse direction to the packet which caused it */
1763 ct = nf_ct_get(skb, &ctinfo);
1764 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1765 ctinfo = IP_CT_RELATED_REPLY;
1767 ctinfo = IP_CT_RELATED;
1769 /* Attach to new skbuff, and increment count */
1770 nf_ct_set(nskb, ct, ctinfo);
1771 nf_conntrack_get(skb_nfct(nskb));
1774 static int nf_conntrack_update(struct net *net, struct sk_buff *skb)
1776 const struct nf_conntrack_l4proto *l4proto;
1777 struct nf_conntrack_tuple_hash *h;
1778 struct nf_conntrack_tuple tuple;
1779 enum ip_conntrack_info ctinfo;
1780 struct nf_nat_hook *nat_hook;
1781 unsigned int status;
1787 ct = nf_ct_get(skb, &ctinfo);
1788 if (!ct || nf_ct_is_confirmed(ct))
1791 l3num = nf_ct_l3num(ct);
1793 dataoff = get_l4proto(skb, skb_network_offset(skb), l3num, &l4num);
1797 l4proto = nf_ct_l4proto_find_get(l4num);
1799 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
1800 l4num, net, &tuple, l4proto))
1803 if (ct->status & IPS_SRC_NAT) {
1804 memcpy(tuple.src.u3.all,
1805 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.all,
1806 sizeof(tuple.src.u3.all));
1808 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.all;
1811 if (ct->status & IPS_DST_NAT) {
1812 memcpy(tuple.dst.u3.all,
1813 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.all,
1814 sizeof(tuple.dst.u3.all));
1816 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u.all;
1819 h = nf_conntrack_find_get(net, nf_ct_zone(ct), &tuple);
1823 /* Store status bits of the conntrack that is clashing to re-do NAT
1824 * mangling according to what it has been done already to this packet.
1826 status = ct->status;
1829 ct = nf_ct_tuplehash_to_ctrack(h);
1830 nf_ct_set(skb, ct, ctinfo);
1832 nat_hook = rcu_dereference(nf_nat_hook);
1836 if (status & IPS_SRC_NAT &&
1837 nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_SRC,
1838 IP_CT_DIR_ORIGINAL) == NF_DROP)
1841 if (status & IPS_DST_NAT &&
1842 nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_DST,
1843 IP_CT_DIR_ORIGINAL) == NF_DROP)
1849 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
1850 const struct sk_buff *skb)
1852 const struct nf_conntrack_tuple *src_tuple;
1853 const struct nf_conntrack_tuple_hash *hash;
1854 struct nf_conntrack_tuple srctuple;
1855 enum ip_conntrack_info ctinfo;
1858 ct = nf_ct_get(skb, &ctinfo);
1860 src_tuple = nf_ct_tuple(ct, CTINFO2DIR(ctinfo));
1861 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
1865 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
1866 NFPROTO_IPV4, dev_net(skb->dev),
1870 hash = nf_conntrack_find_get(dev_net(skb->dev),
1876 ct = nf_ct_tuplehash_to_ctrack(hash);
1877 src_tuple = nf_ct_tuple(ct, !hash->tuple.dst.dir);
1878 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
1884 /* Bring out ya dead! */
1885 static struct nf_conn *
1886 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
1887 void *data, unsigned int *bucket)
1889 struct nf_conntrack_tuple_hash *h;
1891 struct hlist_nulls_node *n;
1894 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1895 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1897 nf_conntrack_lock(lockp);
1898 if (*bucket < nf_conntrack_htable_size) {
1899 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1900 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1902 ct = nf_ct_tuplehash_to_ctrack(h);
1914 atomic_inc(&ct->ct_general.use);
1920 static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data),
1921 void *data, u32 portid, int report)
1923 unsigned int bucket = 0, sequence;
1929 sequence = read_seqcount_begin(&nf_conntrack_generation);
1931 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
1932 /* Time to push up daises... */
1934 nf_ct_delete(ct, portid, report);
1939 if (!read_seqcount_retry(&nf_conntrack_generation, sequence))
1946 int (*iter)(struct nf_conn *i, void *data);
1951 static int iter_net_only(struct nf_conn *i, void *data)
1953 struct iter_data *d = data;
1955 if (!net_eq(d->net, nf_ct_net(i)))
1958 return d->iter(i, d->data);
1962 __nf_ct_unconfirmed_destroy(struct net *net)
1966 for_each_possible_cpu(cpu) {
1967 struct nf_conntrack_tuple_hash *h;
1968 struct hlist_nulls_node *n;
1969 struct ct_pcpu *pcpu;
1971 pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1973 spin_lock_bh(&pcpu->lock);
1974 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1977 ct = nf_ct_tuplehash_to_ctrack(h);
1979 /* we cannot call iter() on unconfirmed list, the
1980 * owning cpu can reallocate ct->ext at any time.
1982 set_bit(IPS_DYING_BIT, &ct->status);
1984 spin_unlock_bh(&pcpu->lock);
1989 void nf_ct_unconfirmed_destroy(struct net *net)
1993 if (atomic_read(&net->ct.count) > 0) {
1994 __nf_ct_unconfirmed_destroy(net);
1995 nf_queue_nf_hook_drop(net);
1999 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy);
2001 void nf_ct_iterate_cleanup_net(struct net *net,
2002 int (*iter)(struct nf_conn *i, void *data),
2003 void *data, u32 portid, int report)
2009 if (atomic_read(&net->ct.count) == 0)
2016 nf_ct_iterate_cleanup(iter_net_only, &d, portid, report);
2018 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net);
2021 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2022 * @iter: callback to invoke for each conntrack
2023 * @data: data to pass to @iter
2025 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2026 * unconfirmed list as dying (so they will not be inserted into
2029 * Can only be called in module exit path.
2032 nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
2036 down_read(&net_rwsem);
2038 if (atomic_read(&net->ct.count) == 0)
2040 __nf_ct_unconfirmed_destroy(net);
2041 nf_queue_nf_hook_drop(net);
2043 up_read(&net_rwsem);
2045 /* Need to wait for netns cleanup worker to finish, if its
2046 * running -- it might have deleted a net namespace from
2047 * the global list, so our __nf_ct_unconfirmed_destroy() might
2048 * not have affected all namespaces.
2052 /* a conntrack could have been unlinked from unconfirmed list
2053 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2054 * This makes sure its inserted into conntrack table.
2058 nf_ct_iterate_cleanup(iter, data, 0, 0);
2060 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
2062 static int kill_all(struct nf_conn *i, void *data)
2064 return net_eq(nf_ct_net(i), data);
2067 void nf_conntrack_cleanup_start(void)
2069 conntrack_gc_work.exiting = true;
2070 RCU_INIT_POINTER(ip_ct_attach, NULL);
2073 void nf_conntrack_cleanup_end(void)
2075 RCU_INIT_POINTER(nf_ct_hook, NULL);
2076 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
2077 kvfree(nf_conntrack_hash);
2079 nf_conntrack_proto_fini();
2080 nf_conntrack_seqadj_fini();
2081 nf_conntrack_labels_fini();
2082 nf_conntrack_helper_fini();
2083 nf_conntrack_timeout_fini();
2084 nf_conntrack_ecache_fini();
2085 nf_conntrack_tstamp_fini();
2086 nf_conntrack_acct_fini();
2087 nf_conntrack_expect_fini();
2089 kmem_cache_destroy(nf_conntrack_cachep);
2093 * Mishearing the voices in his head, our hero wonders how he's
2094 * supposed to kill the mall.
2096 void nf_conntrack_cleanup_net(struct net *net)
2100 list_add(&net->exit_list, &single);
2101 nf_conntrack_cleanup_net_list(&single);
2104 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
2110 * This makes sure all current packets have passed through
2111 * netfilter framework. Roll on, two-stage module
2117 list_for_each_entry(net, net_exit_list, exit_list) {
2118 nf_ct_iterate_cleanup(kill_all, net, 0, 0);
2119 if (atomic_read(&net->ct.count) != 0)
2124 goto i_see_dead_people;
2127 list_for_each_entry(net, net_exit_list, exit_list) {
2128 nf_conntrack_proto_pernet_fini(net);
2129 nf_conntrack_ecache_pernet_fini(net);
2130 nf_conntrack_expect_pernet_fini(net);
2131 free_percpu(net->ct.stat);
2132 free_percpu(net->ct.pcpu_lists);
2136 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
2138 struct hlist_nulls_head *hash;
2139 unsigned int nr_slots, i;
2141 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
2144 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
2145 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
2147 hash = kvmalloc_array(nr_slots, sizeof(struct hlist_nulls_head),
2148 GFP_KERNEL | __GFP_ZERO);
2151 for (i = 0; i < nr_slots; i++)
2152 INIT_HLIST_NULLS_HEAD(&hash[i], i);
2156 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
2158 int nf_conntrack_hash_resize(unsigned int hashsize)
2161 unsigned int old_size;
2162 struct hlist_nulls_head *hash, *old_hash;
2163 struct nf_conntrack_tuple_hash *h;
2169 hash = nf_ct_alloc_hashtable(&hashsize, 1);
2173 old_size = nf_conntrack_htable_size;
2174 if (old_size == hashsize) {
2180 nf_conntrack_all_lock();
2181 write_seqcount_begin(&nf_conntrack_generation);
2183 /* Lookups in the old hash might happen in parallel, which means we
2184 * might get false negatives during connection lookup. New connections
2185 * created because of a false negative won't make it into the hash
2186 * though since that required taking the locks.
2189 for (i = 0; i < nf_conntrack_htable_size; i++) {
2190 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
2191 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
2192 struct nf_conntrack_tuple_hash, hnnode);
2193 ct = nf_ct_tuplehash_to_ctrack(h);
2194 hlist_nulls_del_rcu(&h->hnnode);
2195 bucket = __hash_conntrack(nf_ct_net(ct),
2196 &h->tuple, hashsize);
2197 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
2200 old_size = nf_conntrack_htable_size;
2201 old_hash = nf_conntrack_hash;
2203 nf_conntrack_hash = hash;
2204 nf_conntrack_htable_size = hashsize;
2206 write_seqcount_end(&nf_conntrack_generation);
2207 nf_conntrack_all_unlock();
2215 int nf_conntrack_set_hashsize(const char *val, const struct kernel_param *kp)
2217 unsigned int hashsize;
2220 if (current->nsproxy->net_ns != &init_net)
2223 /* On boot, we can set this without any fancy locking. */
2224 if (!nf_conntrack_hash)
2225 return param_set_uint(val, kp);
2227 rc = kstrtouint(val, 0, &hashsize);
2231 return nf_conntrack_hash_resize(hashsize);
2233 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
2235 static __always_inline unsigned int total_extension_size(void)
2237 /* remember to add new extensions below */
2238 BUILD_BUG_ON(NF_CT_EXT_NUM > 9);
2240 return sizeof(struct nf_ct_ext) +
2241 sizeof(struct nf_conn_help)
2242 #if IS_ENABLED(CONFIG_NF_NAT)
2243 + sizeof(struct nf_conn_nat)
2245 + sizeof(struct nf_conn_seqadj)
2246 + sizeof(struct nf_conn_acct)
2247 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2248 + sizeof(struct nf_conntrack_ecache)
2250 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2251 + sizeof(struct nf_conn_tstamp)
2253 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2254 + sizeof(struct nf_conn_timeout)
2256 #ifdef CONFIG_NF_CONNTRACK_LABELS
2257 + sizeof(struct nf_conn_labels)
2259 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2260 + sizeof(struct nf_conn_synproxy)
2265 int nf_conntrack_init_start(void)
2267 unsigned long nr_pages = totalram_pages();
2272 /* struct nf_ct_ext uses u8 to store offsets/size */
2273 BUILD_BUG_ON(total_extension_size() > 255u);
2275 seqcount_init(&nf_conntrack_generation);
2277 for (i = 0; i < CONNTRACK_LOCKS; i++)
2278 spin_lock_init(&nf_conntrack_locks[i]);
2280 if (!nf_conntrack_htable_size) {
2281 /* Idea from tcp.c: use 1/16384 of memory.
2282 * On i386: 32MB machine has 512 buckets.
2283 * >= 1GB machines have 16384 buckets.
2284 * >= 4GB machines have 65536 buckets.
2286 nf_conntrack_htable_size
2287 = (((nr_pages << PAGE_SHIFT) / 16384)
2288 / sizeof(struct hlist_head));
2289 if (nr_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
2290 nf_conntrack_htable_size = 65536;
2291 else if (nr_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
2292 nf_conntrack_htable_size = 16384;
2293 if (nf_conntrack_htable_size < 32)
2294 nf_conntrack_htable_size = 32;
2296 /* Use a max. factor of four by default to get the same max as
2297 * with the old struct list_heads. When a table size is given
2298 * we use the old value of 8 to avoid reducing the max.
2303 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
2304 if (!nf_conntrack_hash)
2307 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
2309 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
2310 sizeof(struct nf_conn),
2312 SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
2313 if (!nf_conntrack_cachep)
2316 ret = nf_conntrack_expect_init();
2320 ret = nf_conntrack_acct_init();
2324 ret = nf_conntrack_tstamp_init();
2328 ret = nf_conntrack_ecache_init();
2332 ret = nf_conntrack_timeout_init();
2336 ret = nf_conntrack_helper_init();
2340 ret = nf_conntrack_labels_init();
2344 ret = nf_conntrack_seqadj_init();
2348 ret = nf_conntrack_proto_init();
2352 conntrack_gc_work_init(&conntrack_gc_work);
2353 queue_delayed_work(system_power_efficient_wq, &conntrack_gc_work.dwork, HZ);
2358 nf_conntrack_seqadj_fini();
2360 nf_conntrack_labels_fini();
2362 nf_conntrack_helper_fini();
2364 nf_conntrack_timeout_fini();
2366 nf_conntrack_ecache_fini();
2368 nf_conntrack_tstamp_fini();
2370 nf_conntrack_acct_fini();
2372 nf_conntrack_expect_fini();
2374 kmem_cache_destroy(nf_conntrack_cachep);
2376 kvfree(nf_conntrack_hash);
2380 static struct nf_ct_hook nf_conntrack_hook = {
2381 .update = nf_conntrack_update,
2382 .destroy = destroy_conntrack,
2383 .get_tuple_skb = nf_conntrack_get_tuple_skb,
2386 void nf_conntrack_init_end(void)
2388 /* For use by REJECT target */
2389 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
2390 RCU_INIT_POINTER(nf_ct_hook, &nf_conntrack_hook);
2394 * We need to use special "null" values, not used in hash table
2396 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2397 #define DYING_NULLS_VAL ((1<<30)+1)
2398 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
2400 int nf_conntrack_init_net(struct net *net)
2405 BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
2406 atomic_set(&net->ct.count, 0);
2408 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
2409 if (!net->ct.pcpu_lists)
2412 for_each_possible_cpu(cpu) {
2413 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2415 spin_lock_init(&pcpu->lock);
2416 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
2417 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
2420 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
2422 goto err_pcpu_lists;
2424 ret = nf_conntrack_expect_pernet_init(net);
2428 nf_conntrack_acct_pernet_init(net);
2429 nf_conntrack_tstamp_pernet_init(net);
2430 nf_conntrack_ecache_pernet_init(net);
2431 nf_conntrack_helper_pernet_init(net);
2433 ret = nf_conntrack_proto_pernet_init(net);
2439 nf_conntrack_ecache_pernet_fini(net);
2440 nf_conntrack_expect_pernet_fini(net);
2442 free_percpu(net->ct.stat);
2444 free_percpu(net->ct.pcpu_lists);