1 /* Connection state tracking for netfilter. This is separated from,
2 but required by, the NAT layer; it can also be used by an iptables
5 /* (C) 1999-2001 Paul `Rusty' Russell
6 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
7 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
8 * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/types.h>
18 #include <linux/netfilter.h>
19 #include <linux/module.h>
20 #include <linux/sched.h>
21 #include <linux/skbuff.h>
22 #include <linux/proc_fs.h>
23 #include <linux/vmalloc.h>
24 #include <linux/stddef.h>
25 #include <linux/slab.h>
26 #include <linux/random.h>
27 #include <linux/jhash.h>
28 #include <linux/err.h>
29 #include <linux/percpu.h>
30 #include <linux/moduleparam.h>
31 #include <linux/notifier.h>
32 #include <linux/kernel.h>
33 #include <linux/netdevice.h>
34 #include <linux/socket.h>
36 #include <linux/nsproxy.h>
37 #include <linux/rculist_nulls.h>
39 #include <net/netfilter/nf_conntrack.h>
40 #include <net/netfilter/nf_conntrack_l3proto.h>
41 #include <net/netfilter/nf_conntrack_l4proto.h>
42 #include <net/netfilter/nf_conntrack_expect.h>
43 #include <net/netfilter/nf_conntrack_helper.h>
44 #include <net/netfilter/nf_conntrack_seqadj.h>
45 #include <net/netfilter/nf_conntrack_core.h>
46 #include <net/netfilter/nf_conntrack_extend.h>
47 #include <net/netfilter/nf_conntrack_acct.h>
48 #include <net/netfilter/nf_conntrack_ecache.h>
49 #include <net/netfilter/nf_conntrack_zones.h>
50 #include <net/netfilter/nf_conntrack_timestamp.h>
51 #include <net/netfilter/nf_conntrack_timeout.h>
52 #include <net/netfilter/nf_conntrack_labels.h>
53 #include <net/netfilter/nf_conntrack_synproxy.h>
54 #include <net/netfilter/nf_nat.h>
55 #include <net/netfilter/nf_nat_core.h>
56 #include <net/netfilter/nf_nat_helper.h>
57 #include <net/netns/hash.h>
59 #define NF_CONNTRACK_VERSION "0.5.0"
61 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
62 enum nf_nat_manip_type manip,
63 const struct nlattr *attr) __read_mostly;
64 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
66 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
67 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
69 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
70 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
72 struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
73 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
75 struct conntrack_gc_work {
76 struct delayed_work dwork;
82 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
83 static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
84 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
85 static __read_mostly bool nf_conntrack_locks_all;
87 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
88 #define GC_MAX_BUCKETS_DIV 128u
89 /* upper bound of full table scan */
90 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
91 /* desired ratio of entries found to be expired */
92 #define GC_EVICT_RATIO 50u
94 static struct conntrack_gc_work conntrack_gc_work;
96 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
99 while (unlikely(nf_conntrack_locks_all)) {
103 * Order the 'nf_conntrack_locks_all' load vs. the
104 * spin_unlock_wait() loads below, to ensure
105 * that 'nf_conntrack_locks_all_lock' is indeed held:
107 smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
108 spin_unlock_wait(&nf_conntrack_locks_all_lock);
112 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
114 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
116 h1 %= CONNTRACK_LOCKS;
117 h2 %= CONNTRACK_LOCKS;
118 spin_unlock(&nf_conntrack_locks[h1]);
120 spin_unlock(&nf_conntrack_locks[h2]);
123 /* return true if we need to recompute hashes (in case hash table was resized) */
124 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
125 unsigned int h2, unsigned int sequence)
127 h1 %= CONNTRACK_LOCKS;
128 h2 %= CONNTRACK_LOCKS;
130 nf_conntrack_lock(&nf_conntrack_locks[h1]);
132 spin_lock_nested(&nf_conntrack_locks[h2],
133 SINGLE_DEPTH_NESTING);
135 nf_conntrack_lock(&nf_conntrack_locks[h2]);
136 spin_lock_nested(&nf_conntrack_locks[h1],
137 SINGLE_DEPTH_NESTING);
139 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
140 nf_conntrack_double_unlock(h1, h2);
146 static void nf_conntrack_all_lock(void)
150 spin_lock(&nf_conntrack_locks_all_lock);
151 nf_conntrack_locks_all = true;
154 * Order the above store of 'nf_conntrack_locks_all' against
155 * the spin_unlock_wait() loads below, such that if
156 * nf_conntrack_lock() observes 'nf_conntrack_locks_all'
157 * we must observe nf_conntrack_locks[] held:
159 smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
161 for (i = 0; i < CONNTRACK_LOCKS; i++) {
162 spin_unlock_wait(&nf_conntrack_locks[i]);
166 static void nf_conntrack_all_unlock(void)
169 * All prior stores must be complete before we clear
170 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
171 * might observe the false value but not the entire
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 seqcount_t nf_conntrack_generation __read_mostly;
184 /* nf_conn must be 8 bytes aligned, as the 3 LSB bits are used
185 * for the nfctinfo. We cheat by (ab)using the PER CPU cache line
186 * alignment to enforce this.
188 DEFINE_PER_CPU_ALIGNED(struct nf_conn, nf_conntrack_untracked);
189 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
191 static unsigned int nf_conntrack_hash_rnd __read_mostly;
193 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
194 const struct net *net)
199 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
201 /* The direction must be ignored, so we hash everything up to the
202 * destination ports (which is a multiple of 4) and treat the last
203 * three bytes manually.
205 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
206 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
207 return jhash2((u32 *)tuple, n, seed ^
208 (((__force __u16)tuple->dst.u.all << 16) |
209 tuple->dst.protonum));
212 static u32 scale_hash(u32 hash)
214 return reciprocal_scale(hash, nf_conntrack_htable_size);
217 static u32 __hash_conntrack(const struct net *net,
218 const struct nf_conntrack_tuple *tuple,
221 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
224 static u32 hash_conntrack(const struct net *net,
225 const struct nf_conntrack_tuple *tuple)
227 return scale_hash(hash_conntrack_raw(tuple, net));
231 nf_ct_get_tuple(const struct sk_buff *skb,
233 unsigned int dataoff,
237 struct nf_conntrack_tuple *tuple,
238 const struct nf_conntrack_l3proto *l3proto,
239 const struct nf_conntrack_l4proto *l4proto)
241 memset(tuple, 0, sizeof(*tuple));
243 tuple->src.l3num = l3num;
244 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
247 tuple->dst.protonum = protonum;
248 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
250 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
252 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
254 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
256 struct net *net, struct nf_conntrack_tuple *tuple)
258 struct nf_conntrack_l3proto *l3proto;
259 struct nf_conntrack_l4proto *l4proto;
260 unsigned int protoff;
266 l3proto = __nf_ct_l3proto_find(l3num);
267 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
268 if (ret != NF_ACCEPT) {
273 l4proto = __nf_ct_l4proto_find(l3num, protonum);
275 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
281 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
284 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
285 const struct nf_conntrack_tuple *orig,
286 const struct nf_conntrack_l3proto *l3proto,
287 const struct nf_conntrack_l4proto *l4proto)
289 memset(inverse, 0, sizeof(*inverse));
291 inverse->src.l3num = orig->src.l3num;
292 if (l3proto->invert_tuple(inverse, orig) == 0)
295 inverse->dst.dir = !orig->dst.dir;
297 inverse->dst.protonum = orig->dst.protonum;
298 return l4proto->invert_tuple(inverse, orig);
300 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
303 clean_from_lists(struct nf_conn *ct)
305 pr_debug("clean_from_lists(%p)\n", ct);
306 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
307 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
309 /* Destroy all pending expectations */
310 nf_ct_remove_expectations(ct);
313 /* must be called with local_bh_disable */
314 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
316 struct ct_pcpu *pcpu;
318 /* add this conntrack to the (per cpu) dying list */
319 ct->cpu = smp_processor_id();
320 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
322 spin_lock(&pcpu->lock);
323 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
325 spin_unlock(&pcpu->lock);
328 /* must be called with local_bh_disable */
329 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
331 struct ct_pcpu *pcpu;
333 /* add this conntrack to the (per cpu) unconfirmed list */
334 ct->cpu = smp_processor_id();
335 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
337 spin_lock(&pcpu->lock);
338 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
340 spin_unlock(&pcpu->lock);
343 /* must be called with local_bh_disable */
344 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
346 struct ct_pcpu *pcpu;
348 /* We overload first tuple to link into unconfirmed or dying list.*/
349 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
351 spin_lock(&pcpu->lock);
352 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
353 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
354 spin_unlock(&pcpu->lock);
357 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
359 /* Released via destroy_conntrack() */
360 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
361 const struct nf_conntrack_zone *zone,
364 struct nf_conn *tmpl, *p;
366 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
367 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
372 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
374 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
375 tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
378 tmpl = kzalloc(sizeof(*tmpl), flags);
383 tmpl->status = IPS_TEMPLATE;
384 write_pnet(&tmpl->ct_net, net);
385 nf_ct_zone_add(tmpl, zone);
386 atomic_set(&tmpl->ct_general.use, 0);
390 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
392 void nf_ct_tmpl_free(struct nf_conn *tmpl)
394 nf_ct_ext_destroy(tmpl);
395 nf_ct_ext_free(tmpl);
397 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
398 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
402 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
405 destroy_conntrack(struct nf_conntrack *nfct)
407 struct nf_conn *ct = (struct nf_conn *)nfct;
408 struct nf_conntrack_l4proto *l4proto;
410 pr_debug("destroy_conntrack(%p)\n", ct);
411 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
413 if (unlikely(nf_ct_is_template(ct))) {
418 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
419 if (l4proto->destroy)
420 l4proto->destroy(ct);
425 /* Expectations will have been removed in clean_from_lists,
426 * except TFTP can create an expectation on the first packet,
427 * before connection is in the list, so we need to clean here,
430 nf_ct_remove_expectations(ct);
432 nf_ct_del_from_dying_or_unconfirmed_list(ct);
437 nf_ct_put(ct->master);
439 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
440 nf_conntrack_free(ct);
443 static void nf_ct_delete_from_lists(struct nf_conn *ct)
445 struct net *net = nf_ct_net(ct);
446 unsigned int hash, reply_hash;
447 unsigned int sequence;
449 nf_ct_helper_destroy(ct);
453 sequence = read_seqcount_begin(&nf_conntrack_generation);
454 hash = hash_conntrack(net,
455 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
456 reply_hash = hash_conntrack(net,
457 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
458 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
460 clean_from_lists(ct);
461 nf_conntrack_double_unlock(hash, reply_hash);
463 nf_ct_add_to_dying_list(ct);
468 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
470 struct nf_conn_tstamp *tstamp;
472 if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
475 tstamp = nf_conn_tstamp_find(ct);
476 if (tstamp && tstamp->stop == 0)
477 tstamp->stop = ktime_get_real_ns();
479 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
480 portid, report) < 0) {
481 /* destroy event was not delivered. nf_ct_put will
482 * be done by event cache worker on redelivery.
484 nf_ct_delete_from_lists(ct);
485 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
489 nf_conntrack_ecache_work(nf_ct_net(ct));
490 nf_ct_delete_from_lists(ct);
494 EXPORT_SYMBOL_GPL(nf_ct_delete);
497 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
498 const struct nf_conntrack_tuple *tuple,
499 const struct nf_conntrack_zone *zone,
500 const struct net *net)
502 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
504 /* A conntrack can be recreated with the equal tuple,
505 * so we need to check that the conntrack is confirmed
507 return nf_ct_tuple_equal(tuple, &h->tuple) &&
508 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
509 nf_ct_is_confirmed(ct) &&
510 net_eq(net, nf_ct_net(ct));
513 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
514 static void nf_ct_gc_expired(struct nf_conn *ct)
516 if (!atomic_inc_not_zero(&ct->ct_general.use))
519 if (nf_ct_should_gc(ct))
527 * - Caller must take a reference on returned object
528 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
530 static struct nf_conntrack_tuple_hash *
531 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
532 const struct nf_conntrack_tuple *tuple, u32 hash)
534 struct nf_conntrack_tuple_hash *h;
535 struct hlist_nulls_head *ct_hash;
536 struct hlist_nulls_node *n;
537 unsigned int bucket, hsize;
540 nf_conntrack_get_ht(&ct_hash, &hsize);
541 bucket = reciprocal_scale(hash, hsize);
543 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
546 ct = nf_ct_tuplehash_to_ctrack(h);
547 if (nf_ct_is_expired(ct)) {
548 nf_ct_gc_expired(ct);
552 if (nf_ct_is_dying(ct))
555 if (nf_ct_key_equal(h, tuple, zone, net))
559 * if the nulls value we got at the end of this lookup is
560 * not the expected one, we must restart lookup.
561 * We probably met an item that was moved to another chain.
563 if (get_nulls_value(n) != bucket) {
564 NF_CT_STAT_INC_ATOMIC(net, search_restart);
571 /* Find a connection corresponding to a tuple. */
572 static struct nf_conntrack_tuple_hash *
573 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
574 const struct nf_conntrack_tuple *tuple, u32 hash)
576 struct nf_conntrack_tuple_hash *h;
581 h = ____nf_conntrack_find(net, zone, tuple, hash);
583 ct = nf_ct_tuplehash_to_ctrack(h);
584 if (unlikely(nf_ct_is_dying(ct) ||
585 !atomic_inc_not_zero(&ct->ct_general.use)))
588 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
599 struct nf_conntrack_tuple_hash *
600 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
601 const struct nf_conntrack_tuple *tuple)
603 return __nf_conntrack_find_get(net, zone, tuple,
604 hash_conntrack_raw(tuple, net));
606 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
608 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
610 unsigned int reply_hash)
612 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
613 &nf_conntrack_hash[hash]);
614 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
615 &nf_conntrack_hash[reply_hash]);
619 nf_conntrack_hash_check_insert(struct nf_conn *ct)
621 const struct nf_conntrack_zone *zone;
622 struct net *net = nf_ct_net(ct);
623 unsigned int hash, reply_hash;
624 struct nf_conntrack_tuple_hash *h;
625 struct hlist_nulls_node *n;
626 unsigned int sequence;
628 zone = nf_ct_zone(ct);
632 sequence = read_seqcount_begin(&nf_conntrack_generation);
633 hash = hash_conntrack(net,
634 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
635 reply_hash = hash_conntrack(net,
636 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
637 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
639 /* See if there's one in the list already, including reverse */
640 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
641 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
645 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
646 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
651 /* The caller holds a reference to this object */
652 atomic_set(&ct->ct_general.use, 2);
653 __nf_conntrack_hash_insert(ct, hash, reply_hash);
654 nf_conntrack_double_unlock(hash, reply_hash);
655 NF_CT_STAT_INC(net, insert);
660 nf_conntrack_double_unlock(hash, reply_hash);
661 NF_CT_STAT_INC(net, insert_failed);
665 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
667 static inline void nf_ct_acct_update(struct nf_conn *ct,
668 enum ip_conntrack_info ctinfo,
671 struct nf_conn_acct *acct;
673 acct = nf_conn_acct_find(ct);
675 struct nf_conn_counter *counter = acct->counter;
677 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
678 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
682 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
683 const struct nf_conn *loser_ct)
685 struct nf_conn_acct *acct;
687 acct = nf_conn_acct_find(loser_ct);
689 struct nf_conn_counter *counter = acct->counter;
692 /* u32 should be fine since we must have seen one packet. */
693 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
694 nf_ct_acct_update(ct, ctinfo, bytes);
698 /* Resolve race on insertion if this protocol allows this. */
699 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
700 enum ip_conntrack_info ctinfo,
701 struct nf_conntrack_tuple_hash *h)
703 /* This is the conntrack entry already in hashes that won race. */
704 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
705 struct nf_conntrack_l4proto *l4proto;
707 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
708 if (l4proto->allow_clash &&
710 !nf_ct_is_dying(ct) &&
711 atomic_inc_not_zero(&ct->ct_general.use)) {
712 enum ip_conntrack_info oldinfo;
713 struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo);
715 nf_ct_acct_merge(ct, ctinfo, loser_ct);
716 nf_conntrack_put(&loser_ct->ct_general);
717 nf_ct_set(skb, ct, oldinfo);
720 NF_CT_STAT_INC(net, drop);
724 /* Confirm a connection given skb; places it in hash table */
726 __nf_conntrack_confirm(struct sk_buff *skb)
728 const struct nf_conntrack_zone *zone;
729 unsigned int hash, reply_hash;
730 struct nf_conntrack_tuple_hash *h;
732 struct nf_conn_help *help;
733 struct nf_conn_tstamp *tstamp;
734 struct hlist_nulls_node *n;
735 enum ip_conntrack_info ctinfo;
737 unsigned int sequence;
740 ct = nf_ct_get(skb, &ctinfo);
743 /* ipt_REJECT uses nf_conntrack_attach to attach related
744 ICMP/TCP RST packets in other direction. Actual packet
745 which created connection will be IP_CT_NEW or for an
746 expected connection, IP_CT_RELATED. */
747 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
750 zone = nf_ct_zone(ct);
754 sequence = read_seqcount_begin(&nf_conntrack_generation);
755 /* reuse the hash saved before */
756 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
757 hash = scale_hash(hash);
758 reply_hash = hash_conntrack(net,
759 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
761 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
763 /* We're not in hash table, and we refuse to set up related
764 * connections for unconfirmed conns. But packet copies and
765 * REJECT will give spurious warnings here.
767 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
769 /* No external references means no one else could have
772 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
773 pr_debug("Confirming conntrack %p\n", ct);
774 /* We have to check the DYING flag after unlink to prevent
775 * a race against nf_ct_get_next_corpse() possibly called from
776 * user context, else we insert an already 'dead' hash, blocking
777 * further use of that particular connection -JM.
779 nf_ct_del_from_dying_or_unconfirmed_list(ct);
781 if (unlikely(nf_ct_is_dying(ct))) {
782 nf_ct_add_to_dying_list(ct);
786 /* See if there's one in the list already, including reverse:
787 NAT could have grabbed it without realizing, since we're
788 not in the hash. If there is, we lost race. */
789 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
790 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
794 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
795 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
799 /* Timer relative to confirmation time, not original
800 setting time, otherwise we'd get timer wrap in
801 weird delay cases. */
802 ct->timeout += nfct_time_stamp;
803 atomic_inc(&ct->ct_general.use);
804 ct->status |= IPS_CONFIRMED;
806 /* set conntrack timestamp, if enabled. */
807 tstamp = nf_conn_tstamp_find(ct);
809 if (skb->tstamp == 0)
810 __net_timestamp(skb);
812 tstamp->start = ktime_to_ns(skb->tstamp);
814 /* Since the lookup is lockless, hash insertion must be done after
815 * starting the timer and setting the CONFIRMED bit. The RCU barriers
816 * guarantee that no other CPU can find the conntrack before the above
817 * stores are visible.
819 __nf_conntrack_hash_insert(ct, hash, reply_hash);
820 nf_conntrack_double_unlock(hash, reply_hash);
823 help = nfct_help(ct);
824 if (help && help->helper)
825 nf_conntrack_event_cache(IPCT_HELPER, ct);
827 nf_conntrack_event_cache(master_ct(ct) ?
828 IPCT_RELATED : IPCT_NEW, ct);
832 nf_ct_add_to_dying_list(ct);
833 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
835 nf_conntrack_double_unlock(hash, reply_hash);
836 NF_CT_STAT_INC(net, insert_failed);
840 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
842 /* Returns true if a connection correspondings to the tuple (required
845 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
846 const struct nf_conn *ignored_conntrack)
848 struct net *net = nf_ct_net(ignored_conntrack);
849 const struct nf_conntrack_zone *zone;
850 struct nf_conntrack_tuple_hash *h;
851 struct hlist_nulls_head *ct_hash;
852 unsigned int hash, hsize;
853 struct hlist_nulls_node *n;
856 zone = nf_ct_zone(ignored_conntrack);
860 nf_conntrack_get_ht(&ct_hash, &hsize);
861 hash = __hash_conntrack(net, tuple, hsize);
863 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
864 ct = nf_ct_tuplehash_to_ctrack(h);
866 if (ct == ignored_conntrack)
869 if (nf_ct_is_expired(ct)) {
870 nf_ct_gc_expired(ct);
874 if (nf_ct_key_equal(h, tuple, zone, net)) {
875 NF_CT_STAT_INC_ATOMIC(net, found);
881 if (get_nulls_value(n) != hash) {
882 NF_CT_STAT_INC_ATOMIC(net, search_restart);
890 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
892 #define NF_CT_EVICTION_RANGE 8
894 /* There's a small race here where we may free a just-assured
895 connection. Too bad: we're in trouble anyway. */
896 static unsigned int early_drop_list(struct net *net,
897 struct hlist_nulls_head *head)
899 struct nf_conntrack_tuple_hash *h;
900 struct hlist_nulls_node *n;
901 unsigned int drops = 0;
904 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
905 tmp = nf_ct_tuplehash_to_ctrack(h);
907 if (nf_ct_is_expired(tmp)) {
908 nf_ct_gc_expired(tmp);
912 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
913 !net_eq(nf_ct_net(tmp), net) ||
917 if (!atomic_inc_not_zero(&tmp->ct_general.use))
920 /* kill only if still in same netns -- might have moved due to
921 * SLAB_DESTROY_BY_RCU rules.
923 * We steal the timer reference. If that fails timer has
924 * already fired or someone else deleted it. Just drop ref
925 * and move to next entry.
927 if (net_eq(nf_ct_net(tmp), net) &&
928 nf_ct_is_confirmed(tmp) &&
929 nf_ct_delete(tmp, 0, 0))
938 static noinline int early_drop(struct net *net, unsigned int _hash)
942 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
943 struct hlist_nulls_head *ct_hash;
944 unsigned int hash, hsize, drops;
947 nf_conntrack_get_ht(&ct_hash, &hsize);
948 hash = reciprocal_scale(_hash++, hsize);
950 drops = early_drop_list(net, &ct_hash[hash]);
954 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
962 static void gc_worker(struct work_struct *work)
964 unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
965 unsigned int i, goal, buckets = 0, expired_count = 0;
966 struct conntrack_gc_work *gc_work;
967 unsigned int ratio, scanned = 0;
968 unsigned long next_run;
970 gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
972 goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
973 i = gc_work->last_bucket;
976 struct nf_conntrack_tuple_hash *h;
977 struct hlist_nulls_head *ct_hash;
978 struct hlist_nulls_node *n;
985 nf_conntrack_get_ht(&ct_hash, &hashsz);
989 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
990 tmp = nf_ct_tuplehash_to_ctrack(h);
993 if (nf_ct_is_expired(tmp)) {
994 nf_ct_gc_expired(tmp);
1000 /* could check get_nulls_value() here and restart if ct
1001 * was moved to another chain. But given gc is best-effort
1002 * we will just continue with next hash slot.
1005 cond_resched_rcu_qs();
1006 } while (++buckets < goal);
1008 if (gc_work->exiting)
1012 * Eviction will normally happen from the packet path, and not
1013 * from this gc worker.
1015 * This worker is only here to reap expired entries when system went
1016 * idle after a busy period.
1018 * The heuristics below are supposed to balance conflicting goals:
1020 * 1. Minimize time until we notice a stale entry
1021 * 2. Maximize scan intervals to not waste cycles
1023 * Normally, expire ratio will be close to 0.
1025 * As soon as a sizeable fraction of the entries have expired
1026 * increase scan frequency.
1028 ratio = scanned ? expired_count * 100 / scanned : 0;
1029 if (ratio > GC_EVICT_RATIO) {
1030 gc_work->next_gc_run = min_interval;
1032 unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV;
1034 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0);
1036 gc_work->next_gc_run += min_interval;
1037 if (gc_work->next_gc_run > max)
1038 gc_work->next_gc_run = max;
1041 next_run = gc_work->next_gc_run;
1042 gc_work->last_bucket = i;
1043 queue_delayed_work(system_long_wq, &gc_work->dwork, next_run);
1046 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1048 INIT_DELAYED_WORK(&gc_work->dwork, gc_worker);
1049 gc_work->next_gc_run = HZ;
1050 gc_work->exiting = false;
1053 static struct nf_conn *
1054 __nf_conntrack_alloc(struct net *net,
1055 const struct nf_conntrack_zone *zone,
1056 const struct nf_conntrack_tuple *orig,
1057 const struct nf_conntrack_tuple *repl,
1058 gfp_t gfp, u32 hash)
1062 /* We don't want any race condition at early drop stage */
1063 atomic_inc(&net->ct.count);
1065 if (nf_conntrack_max &&
1066 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1067 if (!early_drop(net, hash)) {
1068 atomic_dec(&net->ct.count);
1069 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1070 return ERR_PTR(-ENOMEM);
1075 * Do not use kmem_cache_zalloc(), as this cache uses
1076 * SLAB_DESTROY_BY_RCU.
1078 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1082 spin_lock_init(&ct->lock);
1083 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1084 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1085 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1086 /* save hash for reusing when confirming */
1087 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1089 write_pnet(&ct->ct_net, net);
1090 memset(&ct->__nfct_init_offset[0], 0,
1091 offsetof(struct nf_conn, proto) -
1092 offsetof(struct nf_conn, __nfct_init_offset[0]));
1094 nf_ct_zone_add(ct, zone);
1096 /* Because we use RCU lookups, we set ct_general.use to zero before
1097 * this is inserted in any list.
1099 atomic_set(&ct->ct_general.use, 0);
1102 atomic_dec(&net->ct.count);
1103 return ERR_PTR(-ENOMEM);
1106 struct nf_conn *nf_conntrack_alloc(struct net *net,
1107 const struct nf_conntrack_zone *zone,
1108 const struct nf_conntrack_tuple *orig,
1109 const struct nf_conntrack_tuple *repl,
1112 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1114 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1116 void nf_conntrack_free(struct nf_conn *ct)
1118 struct net *net = nf_ct_net(ct);
1120 /* A freed object has refcnt == 0, that's
1121 * the golden rule for SLAB_DESTROY_BY_RCU
1123 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
1125 nf_ct_ext_destroy(ct);
1127 kmem_cache_free(nf_conntrack_cachep, ct);
1128 smp_mb__before_atomic();
1129 atomic_dec(&net->ct.count);
1131 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1134 /* Allocate a new conntrack: we return -ENOMEM if classification
1135 failed due to stress. Otherwise it really is unclassifiable. */
1136 static noinline struct nf_conntrack_tuple_hash *
1137 init_conntrack(struct net *net, struct nf_conn *tmpl,
1138 const struct nf_conntrack_tuple *tuple,
1139 struct nf_conntrack_l3proto *l3proto,
1140 struct nf_conntrack_l4proto *l4proto,
1141 struct sk_buff *skb,
1142 unsigned int dataoff, u32 hash)
1145 struct nf_conn_help *help;
1146 struct nf_conntrack_tuple repl_tuple;
1147 struct nf_conntrack_ecache *ecache;
1148 struct nf_conntrack_expect *exp = NULL;
1149 const struct nf_conntrack_zone *zone;
1150 struct nf_conn_timeout *timeout_ext;
1151 struct nf_conntrack_zone tmp;
1152 unsigned int *timeouts;
1154 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1155 pr_debug("Can't invert tuple.\n");
1159 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1160 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1163 return (struct nf_conntrack_tuple_hash *)ct;
1165 if (!nf_ct_add_synproxy(ct, tmpl)) {
1166 nf_conntrack_free(ct);
1167 return ERR_PTR(-ENOMEM);
1170 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1172 timeouts = nf_ct_timeout_data(timeout_ext);
1173 if (unlikely(!timeouts))
1174 timeouts = l4proto->get_timeouts(net);
1176 timeouts = l4proto->get_timeouts(net);
1179 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1180 nf_conntrack_free(ct);
1181 pr_debug("can't track with proto module\n");
1186 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1189 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1190 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1191 nf_ct_labels_ext_add(ct);
1193 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1194 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1195 ecache ? ecache->expmask : 0,
1199 if (net->ct.expect_count) {
1200 spin_lock(&nf_conntrack_expect_lock);
1201 exp = nf_ct_find_expectation(net, zone, tuple);
1203 pr_debug("expectation arrives ct=%p exp=%p\n",
1205 /* Welcome, Mr. Bond. We've been expecting you... */
1206 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1207 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1208 ct->master = exp->master;
1210 help = nf_ct_helper_ext_add(ct, exp->helper,
1213 rcu_assign_pointer(help->helper, exp->helper);
1216 #ifdef CONFIG_NF_CONNTRACK_MARK
1217 ct->mark = exp->master->mark;
1219 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1220 ct->secmark = exp->master->secmark;
1222 NF_CT_STAT_INC(net, expect_new);
1224 spin_unlock(&nf_conntrack_expect_lock);
1227 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1229 /* Now it is inserted into the unconfirmed list, bump refcount */
1230 nf_conntrack_get(&ct->ct_general);
1231 nf_ct_add_to_unconfirmed_list(ct);
1237 exp->expectfn(ct, exp);
1238 nf_ct_expect_put(exp);
1241 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1244 /* On success, returns 0, sets skb->_nfct | ctinfo */
1246 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1247 struct sk_buff *skb,
1248 unsigned int dataoff,
1251 struct nf_conntrack_l3proto *l3proto,
1252 struct nf_conntrack_l4proto *l4proto)
1254 const struct nf_conntrack_zone *zone;
1255 struct nf_conntrack_tuple tuple;
1256 struct nf_conntrack_tuple_hash *h;
1257 enum ip_conntrack_info ctinfo;
1258 struct nf_conntrack_zone tmp;
1262 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1263 dataoff, l3num, protonum, net, &tuple, l3proto,
1265 pr_debug("Can't get tuple\n");
1269 /* look for tuple match */
1270 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1271 hash = hash_conntrack_raw(&tuple, net);
1272 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1274 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1275 skb, dataoff, hash);
1281 ct = nf_ct_tuplehash_to_ctrack(h);
1283 /* It exists; we have (non-exclusive) reference. */
1284 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1285 ctinfo = IP_CT_ESTABLISHED_REPLY;
1287 /* Once we've had two way comms, always ESTABLISHED. */
1288 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1289 pr_debug("normal packet for %p\n", ct);
1290 ctinfo = IP_CT_ESTABLISHED;
1291 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1292 pr_debug("related packet for %p\n", ct);
1293 ctinfo = IP_CT_RELATED;
1295 pr_debug("new packet for %p\n", ct);
1299 nf_ct_set(skb, ct, ctinfo);
1304 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1305 struct sk_buff *skb)
1307 struct nf_conn *ct, *tmpl;
1308 enum ip_conntrack_info ctinfo;
1309 struct nf_conntrack_l3proto *l3proto;
1310 struct nf_conntrack_l4proto *l4proto;
1311 unsigned int *timeouts;
1312 unsigned int dataoff;
1316 tmpl = nf_ct_get(skb, &ctinfo);
1318 /* Previously seen (loopback or untracked)? Ignore. */
1319 if (!nf_ct_is_template(tmpl)) {
1320 NF_CT_STAT_INC_ATOMIC(net, ignore);
1326 /* rcu_read_lock()ed by nf_hook_thresh */
1327 l3proto = __nf_ct_l3proto_find(pf);
1328 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1329 &dataoff, &protonum);
1331 pr_debug("not prepared to track yet or error occurred\n");
1332 NF_CT_STAT_INC_ATOMIC(net, error);
1333 NF_CT_STAT_INC_ATOMIC(net, invalid);
1338 l4proto = __nf_ct_l4proto_find(pf, protonum);
1340 /* It may be an special packet, error, unclean...
1341 * inverse of the return code tells to the netfilter
1342 * core what to do with the packet. */
1343 if (l4proto->error != NULL) {
1344 ret = l4proto->error(net, tmpl, skb, dataoff, pf, hooknum);
1346 NF_CT_STAT_INC_ATOMIC(net, error);
1347 NF_CT_STAT_INC_ATOMIC(net, invalid);
1351 /* ICMP[v6] protocol trackers may assign one conntrack. */
1356 ret = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1359 /* Too stressed to deal. */
1360 NF_CT_STAT_INC_ATOMIC(net, drop);
1365 ct = nf_ct_get(skb, &ctinfo);
1367 /* Not valid part of a connection */
1368 NF_CT_STAT_INC_ATOMIC(net, invalid);
1373 /* Decide what timeout policy we want to apply to this flow. */
1374 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1376 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1378 /* Invalid: inverse of the return code tells
1379 * the netfilter core what to do */
1380 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1381 nf_conntrack_put(&ct->ct_general);
1383 NF_CT_STAT_INC_ATOMIC(net, invalid);
1384 if (ret == -NF_DROP)
1385 NF_CT_STAT_INC_ATOMIC(net, drop);
1386 /* Special case: TCP tracker reports an attempt to reopen a
1387 * closed/aborted connection. We have to go back and create a
1390 if (ret == -NF_REPEAT)
1396 if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
1397 !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1398 nf_conntrack_event_cache(IPCT_REPLY, ct);
1405 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1407 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1408 const struct nf_conntrack_tuple *orig)
1413 ret = nf_ct_invert_tuple(inverse, orig,
1414 __nf_ct_l3proto_find(orig->src.l3num),
1415 __nf_ct_l4proto_find(orig->src.l3num,
1416 orig->dst.protonum));
1420 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1422 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1423 implicitly racy: see __nf_conntrack_confirm */
1424 void nf_conntrack_alter_reply(struct nf_conn *ct,
1425 const struct nf_conntrack_tuple *newreply)
1427 struct nf_conn_help *help = nfct_help(ct);
1429 /* Should be unconfirmed, so not in hash table yet */
1430 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1432 pr_debug("Altering reply tuple of %p to ", ct);
1433 nf_ct_dump_tuple(newreply);
1435 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1436 if (ct->master || (help && !hlist_empty(&help->expectations)))
1440 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1443 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1445 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1446 void __nf_ct_refresh_acct(struct nf_conn *ct,
1447 enum ip_conntrack_info ctinfo,
1448 const struct sk_buff *skb,
1449 unsigned long extra_jiffies,
1454 /* Only update if this is not a fixed timeout */
1455 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1458 /* If not in hash table, timer will not be active yet */
1459 if (nf_ct_is_confirmed(ct))
1460 extra_jiffies += nfct_time_stamp;
1462 ct->timeout = extra_jiffies;
1465 nf_ct_acct_update(ct, ctinfo, skb->len);
1467 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1469 bool nf_ct_kill_acct(struct nf_conn *ct,
1470 enum ip_conntrack_info ctinfo,
1471 const struct sk_buff *skb)
1473 nf_ct_acct_update(ct, ctinfo, skb->len);
1475 return nf_ct_delete(ct, 0, 0);
1477 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1479 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1481 #include <linux/netfilter/nfnetlink.h>
1482 #include <linux/netfilter/nfnetlink_conntrack.h>
1483 #include <linux/mutex.h>
1485 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1486 * in ip_conntrack_core, since we don't want the protocols to autoload
1487 * or depend on ctnetlink */
1488 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1489 const struct nf_conntrack_tuple *tuple)
1491 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1492 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1493 goto nla_put_failure;
1499 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1501 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1502 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1503 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1505 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1507 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1508 struct nf_conntrack_tuple *t)
1510 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1513 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1514 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1518 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1520 int nf_ct_port_nlattr_tuple_size(void)
1522 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1524 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1527 /* Used by ipt_REJECT and ip6t_REJECT. */
1528 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1531 enum ip_conntrack_info ctinfo;
1533 /* This ICMP is in reverse direction to the packet which caused it */
1534 ct = nf_ct_get(skb, &ctinfo);
1535 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1536 ctinfo = IP_CT_RELATED_REPLY;
1538 ctinfo = IP_CT_RELATED;
1540 /* Attach to new skbuff, and increment count */
1541 nf_ct_set(nskb, ct, ctinfo);
1542 nf_conntrack_get(skb_nfct(nskb));
1545 /* Bring out ya dead! */
1546 static struct nf_conn *
1547 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1548 void *data, unsigned int *bucket)
1550 struct nf_conntrack_tuple_hash *h;
1552 struct hlist_nulls_node *n;
1556 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1557 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1559 nf_conntrack_lock(lockp);
1560 if (*bucket < nf_conntrack_htable_size) {
1561 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1562 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1564 ct = nf_ct_tuplehash_to_ctrack(h);
1565 if (net_eq(nf_ct_net(ct), net) &&
1575 for_each_possible_cpu(cpu) {
1576 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1578 spin_lock_bh(&pcpu->lock);
1579 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1580 ct = nf_ct_tuplehash_to_ctrack(h);
1582 set_bit(IPS_DYING_BIT, &ct->status);
1584 spin_unlock_bh(&pcpu->lock);
1589 atomic_inc(&ct->ct_general.use);
1595 void nf_ct_iterate_cleanup(struct net *net,
1596 int (*iter)(struct nf_conn *i, void *data),
1597 void *data, u32 portid, int report)
1600 unsigned int bucket = 0;
1604 if (atomic_read(&net->ct.count) == 0)
1607 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1608 /* Time to push up daises... */
1610 nf_ct_delete(ct, portid, report);
1615 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1617 static int kill_all(struct nf_conn *i, void *data)
1622 void nf_ct_free_hashtable(void *hash, unsigned int size)
1624 if (is_vmalloc_addr(hash))
1627 free_pages((unsigned long)hash,
1628 get_order(sizeof(struct hlist_head) * size));
1630 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1632 static int untrack_refs(void)
1636 for_each_possible_cpu(cpu) {
1637 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1639 cnt += atomic_read(&ct->ct_general.use) - 1;
1644 void nf_conntrack_cleanup_start(void)
1646 conntrack_gc_work.exiting = true;
1647 RCU_INIT_POINTER(ip_ct_attach, NULL);
1650 void nf_conntrack_cleanup_end(void)
1652 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1653 while (untrack_refs() > 0)
1656 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
1657 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1659 nf_conntrack_proto_fini();
1660 nf_conntrack_seqadj_fini();
1661 nf_conntrack_labels_fini();
1662 nf_conntrack_helper_fini();
1663 nf_conntrack_timeout_fini();
1664 nf_conntrack_ecache_fini();
1665 nf_conntrack_tstamp_fini();
1666 nf_conntrack_acct_fini();
1667 nf_conntrack_expect_fini();
1669 kmem_cache_destroy(nf_conntrack_cachep);
1673 * Mishearing the voices in his head, our hero wonders how he's
1674 * supposed to kill the mall.
1676 void nf_conntrack_cleanup_net(struct net *net)
1680 list_add(&net->exit_list, &single);
1681 nf_conntrack_cleanup_net_list(&single);
1684 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1690 * This makes sure all current packets have passed through
1691 * netfilter framework. Roll on, two-stage module
1697 list_for_each_entry(net, net_exit_list, exit_list) {
1698 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1699 if (atomic_read(&net->ct.count) != 0)
1704 goto i_see_dead_people;
1707 list_for_each_entry(net, net_exit_list, exit_list) {
1708 nf_conntrack_proto_pernet_fini(net);
1709 nf_conntrack_helper_pernet_fini(net);
1710 nf_conntrack_ecache_pernet_fini(net);
1711 nf_conntrack_tstamp_pernet_fini(net);
1712 nf_conntrack_acct_pernet_fini(net);
1713 nf_conntrack_expect_pernet_fini(net);
1714 free_percpu(net->ct.stat);
1715 free_percpu(net->ct.pcpu_lists);
1719 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1721 struct hlist_nulls_head *hash;
1722 unsigned int nr_slots, i;
1725 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1728 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1729 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1731 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1734 sz = nr_slots * sizeof(struct hlist_nulls_head);
1735 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1741 for (i = 0; i < nr_slots; i++)
1742 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1746 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1748 int nf_conntrack_hash_resize(unsigned int hashsize)
1751 unsigned int old_size;
1752 struct hlist_nulls_head *hash, *old_hash;
1753 struct nf_conntrack_tuple_hash *h;
1759 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1763 old_size = nf_conntrack_htable_size;
1764 if (old_size == hashsize) {
1765 nf_ct_free_hashtable(hash, hashsize);
1770 nf_conntrack_all_lock();
1771 write_seqcount_begin(&nf_conntrack_generation);
1773 /* Lookups in the old hash might happen in parallel, which means we
1774 * might get false negatives during connection lookup. New connections
1775 * created because of a false negative won't make it into the hash
1776 * though since that required taking the locks.
1779 for (i = 0; i < nf_conntrack_htable_size; i++) {
1780 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1781 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1782 struct nf_conntrack_tuple_hash, hnnode);
1783 ct = nf_ct_tuplehash_to_ctrack(h);
1784 hlist_nulls_del_rcu(&h->hnnode);
1785 bucket = __hash_conntrack(nf_ct_net(ct),
1786 &h->tuple, hashsize);
1787 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1790 old_size = nf_conntrack_htable_size;
1791 old_hash = nf_conntrack_hash;
1793 nf_conntrack_hash = hash;
1794 nf_conntrack_htable_size = hashsize;
1796 write_seqcount_end(&nf_conntrack_generation);
1797 nf_conntrack_all_unlock();
1801 nf_ct_free_hashtable(old_hash, old_size);
1805 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1807 unsigned int hashsize;
1810 if (current->nsproxy->net_ns != &init_net)
1813 /* On boot, we can set this without any fancy locking. */
1814 if (!nf_conntrack_htable_size)
1815 return param_set_uint(val, kp);
1817 rc = kstrtouint(val, 0, &hashsize);
1821 return nf_conntrack_hash_resize(hashsize);
1823 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1825 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1826 &nf_conntrack_htable_size, 0600);
1828 void nf_ct_untracked_status_or(unsigned long bits)
1832 for_each_possible_cpu(cpu)
1833 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1835 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1837 int nf_conntrack_init_start(void)
1843 seqcount_init(&nf_conntrack_generation);
1845 for (i = 0; i < CONNTRACK_LOCKS; i++)
1846 spin_lock_init(&nf_conntrack_locks[i]);
1848 if (!nf_conntrack_htable_size) {
1849 /* Idea from tcp.c: use 1/16384 of memory.
1850 * On i386: 32MB machine has 512 buckets.
1851 * >= 1GB machines have 16384 buckets.
1852 * >= 4GB machines have 65536 buckets.
1854 nf_conntrack_htable_size
1855 = (((totalram_pages << PAGE_SHIFT) / 16384)
1856 / sizeof(struct hlist_head));
1857 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1858 nf_conntrack_htable_size = 65536;
1859 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1860 nf_conntrack_htable_size = 16384;
1861 if (nf_conntrack_htable_size < 32)
1862 nf_conntrack_htable_size = 32;
1864 /* Use a max. factor of four by default to get the same max as
1865 * with the old struct list_heads. When a table size is given
1866 * we use the old value of 8 to avoid reducing the max.
1871 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1872 if (!nf_conntrack_hash)
1875 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1877 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1878 sizeof(struct nf_conn),
1880 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1881 if (!nf_conntrack_cachep)
1884 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1885 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1888 ret = nf_conntrack_expect_init();
1892 ret = nf_conntrack_acct_init();
1896 ret = nf_conntrack_tstamp_init();
1900 ret = nf_conntrack_ecache_init();
1904 ret = nf_conntrack_timeout_init();
1908 ret = nf_conntrack_helper_init();
1912 ret = nf_conntrack_labels_init();
1916 ret = nf_conntrack_seqadj_init();
1920 ret = nf_conntrack_proto_init();
1924 /* Set up fake conntrack: to never be deleted, not in any hashes */
1925 for_each_possible_cpu(cpu) {
1926 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1927 write_pnet(&ct->ct_net, &init_net);
1928 atomic_set(&ct->ct_general.use, 1);
1930 /* - and look it like as a confirmed connection */
1931 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1933 conntrack_gc_work_init(&conntrack_gc_work);
1934 queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, HZ);
1939 nf_conntrack_seqadj_fini();
1941 nf_conntrack_labels_fini();
1943 nf_conntrack_helper_fini();
1945 nf_conntrack_timeout_fini();
1947 nf_conntrack_ecache_fini();
1949 nf_conntrack_tstamp_fini();
1951 nf_conntrack_acct_fini();
1953 nf_conntrack_expect_fini();
1955 kmem_cache_destroy(nf_conntrack_cachep);
1957 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1961 void nf_conntrack_init_end(void)
1963 /* For use by REJECT target */
1964 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1965 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1969 * We need to use special "null" values, not used in hash table
1971 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1972 #define DYING_NULLS_VAL ((1<<30)+1)
1973 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1975 int nf_conntrack_init_net(struct net *net)
1980 atomic_set(&net->ct.count, 0);
1982 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1983 if (!net->ct.pcpu_lists)
1986 for_each_possible_cpu(cpu) {
1987 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1989 spin_lock_init(&pcpu->lock);
1990 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1991 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1994 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1996 goto err_pcpu_lists;
1998 ret = nf_conntrack_expect_pernet_init(net);
2001 ret = nf_conntrack_acct_pernet_init(net);
2004 ret = nf_conntrack_tstamp_pernet_init(net);
2007 ret = nf_conntrack_ecache_pernet_init(net);
2010 ret = nf_conntrack_helper_pernet_init(net);
2013 ret = nf_conntrack_proto_pernet_init(net);
2019 nf_conntrack_helper_pernet_fini(net);
2021 nf_conntrack_ecache_pernet_fini(net);
2023 nf_conntrack_tstamp_pernet_fini(net);
2025 nf_conntrack_acct_pernet_fini(net);
2027 nf_conntrack_expect_pernet_fini(net);
2029 free_percpu(net->ct.stat);
2031 free_percpu(net->ct.pcpu_lists);