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 #include <linux/types.h>
16 #include <linux/netfilter.h>
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/skbuff.h>
20 #include <linux/proc_fs.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stddef.h>
23 #include <linux/slab.h>
24 #include <linux/random.h>
25 #include <linux/jhash.h>
26 #include <linux/err.h>
27 #include <linux/percpu.h>
28 #include <linux/moduleparam.h>
29 #include <linux/notifier.h>
30 #include <linux/kernel.h>
31 #include <linux/netdevice.h>
32 #include <linux/socket.h>
34 #include <linux/nsproxy.h>
35 #include <linux/rculist_nulls.h>
37 #include <net/netfilter/nf_conntrack.h>
38 #include <net/netfilter/nf_conntrack_l3proto.h>
39 #include <net/netfilter/nf_conntrack_l4proto.h>
40 #include <net/netfilter/nf_conntrack_expect.h>
41 #include <net/netfilter/nf_conntrack_helper.h>
42 #include <net/netfilter/nf_conntrack_seqadj.h>
43 #include <net/netfilter/nf_conntrack_core.h>
44 #include <net/netfilter/nf_conntrack_extend.h>
45 #include <net/netfilter/nf_conntrack_acct.h>
46 #include <net/netfilter/nf_conntrack_ecache.h>
47 #include <net/netfilter/nf_conntrack_zones.h>
48 #include <net/netfilter/nf_conntrack_timestamp.h>
49 #include <net/netfilter/nf_conntrack_timeout.h>
50 #include <net/netfilter/nf_conntrack_labels.h>
51 #include <net/netfilter/nf_conntrack_synproxy.h>
52 #include <net/netfilter/nf_nat.h>
53 #include <net/netfilter/nf_nat_core.h>
54 #include <net/netfilter/nf_nat_helper.h>
56 #define NF_CONNTRACK_VERSION "0.5.0"
58 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
59 enum nf_nat_manip_type manip,
60 const struct nlattr *attr) __read_mostly;
61 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
63 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
64 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
66 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
67 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
69 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
71 h1 %= CONNTRACK_LOCKS;
72 h2 %= CONNTRACK_LOCKS;
73 spin_unlock(&nf_conntrack_locks[h1]);
75 spin_unlock(&nf_conntrack_locks[h2]);
78 /* return true if we need to recompute hashes (in case hash table was resized) */
79 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
80 unsigned int h2, unsigned int sequence)
82 h1 %= CONNTRACK_LOCKS;
83 h2 %= CONNTRACK_LOCKS;
85 spin_lock(&nf_conntrack_locks[h1]);
87 spin_lock_nested(&nf_conntrack_locks[h2],
88 SINGLE_DEPTH_NESTING);
90 spin_lock(&nf_conntrack_locks[h2]);
91 spin_lock_nested(&nf_conntrack_locks[h1],
92 SINGLE_DEPTH_NESTING);
94 if (read_seqcount_retry(&net->ct.generation, sequence)) {
95 nf_conntrack_double_unlock(h1, h2);
101 static void nf_conntrack_all_lock(void)
105 for (i = 0; i < CONNTRACK_LOCKS; i++)
106 spin_lock_nested(&nf_conntrack_locks[i], i);
109 static void nf_conntrack_all_unlock(void)
113 for (i = 0; i < CONNTRACK_LOCKS; i++)
114 spin_unlock(&nf_conntrack_locks[i]);
117 unsigned int nf_conntrack_htable_size __read_mostly;
118 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
120 unsigned int nf_conntrack_max __read_mostly;
121 EXPORT_SYMBOL_GPL(nf_conntrack_max);
123 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
124 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
126 unsigned int nf_conntrack_hash_rnd __read_mostly;
127 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
129 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple)
133 /* The direction must be ignored, so we hash everything up to the
134 * destination ports (which is a multiple of 4) and treat the last
135 * three bytes manually.
137 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
138 return jhash2((u32 *)tuple, n, nf_conntrack_hash_rnd ^
139 (((__force __u16)tuple->dst.u.all << 16) |
140 tuple->dst.protonum));
143 static u32 __hash_bucket(u32 hash, unsigned int size)
145 return reciprocal_scale(hash, size);
148 static u32 hash_bucket(u32 hash, const struct net *net)
150 return __hash_bucket(hash, net->ct.htable_size);
153 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
156 return __hash_bucket(hash_conntrack_raw(tuple), size);
159 static inline u_int32_t hash_conntrack(const struct net *net,
160 const struct nf_conntrack_tuple *tuple)
162 return __hash_conntrack(tuple, net->ct.htable_size);
166 nf_ct_get_tuple(const struct sk_buff *skb,
168 unsigned int dataoff,
171 struct nf_conntrack_tuple *tuple,
172 const struct nf_conntrack_l3proto *l3proto,
173 const struct nf_conntrack_l4proto *l4proto)
175 memset(tuple, 0, sizeof(*tuple));
177 tuple->src.l3num = l3num;
178 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
181 tuple->dst.protonum = protonum;
182 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
184 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
186 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
188 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
189 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
191 struct nf_conntrack_l3proto *l3proto;
192 struct nf_conntrack_l4proto *l4proto;
193 unsigned int protoff;
199 l3proto = __nf_ct_l3proto_find(l3num);
200 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
201 if (ret != NF_ACCEPT) {
206 l4proto = __nf_ct_l4proto_find(l3num, protonum);
208 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
214 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
217 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
218 const struct nf_conntrack_tuple *orig,
219 const struct nf_conntrack_l3proto *l3proto,
220 const struct nf_conntrack_l4proto *l4proto)
222 memset(inverse, 0, sizeof(*inverse));
224 inverse->src.l3num = orig->src.l3num;
225 if (l3proto->invert_tuple(inverse, orig) == 0)
228 inverse->dst.dir = !orig->dst.dir;
230 inverse->dst.protonum = orig->dst.protonum;
231 return l4proto->invert_tuple(inverse, orig);
233 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
236 clean_from_lists(struct nf_conn *ct)
238 pr_debug("clean_from_lists(%p)\n", ct);
239 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
240 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
242 /* Destroy all pending expectations */
243 nf_ct_remove_expectations(ct);
246 /* must be called with local_bh_disable */
247 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
249 struct ct_pcpu *pcpu;
251 /* add this conntrack to the (per cpu) dying list */
252 ct->cpu = smp_processor_id();
253 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
255 spin_lock(&pcpu->lock);
256 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
258 spin_unlock(&pcpu->lock);
261 /* must be called with local_bh_disable */
262 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
264 struct ct_pcpu *pcpu;
266 /* add this conntrack to the (per cpu) unconfirmed list */
267 ct->cpu = smp_processor_id();
268 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
270 spin_lock(&pcpu->lock);
271 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
273 spin_unlock(&pcpu->lock);
276 /* must be called with local_bh_disable */
277 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
279 struct ct_pcpu *pcpu;
281 /* We overload first tuple to link into unconfirmed or dying list.*/
282 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
284 spin_lock(&pcpu->lock);
285 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
286 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
287 spin_unlock(&pcpu->lock);
290 /* Released via destroy_conntrack() */
291 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
292 const struct nf_conntrack_zone *zone,
295 struct nf_conn *tmpl;
297 tmpl = kzalloc(sizeof(*tmpl), flags);
301 tmpl->status = IPS_TEMPLATE;
302 write_pnet(&tmpl->ct_net, net);
304 if (nf_ct_zone_add(tmpl, flags, zone) < 0)
307 atomic_set(&tmpl->ct_general.use, 0);
314 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
316 static void nf_ct_tmpl_free(struct nf_conn *tmpl)
318 nf_ct_ext_destroy(tmpl);
319 nf_ct_ext_free(tmpl);
324 destroy_conntrack(struct nf_conntrack *nfct)
326 struct nf_conn *ct = (struct nf_conn *)nfct;
327 struct net *net = nf_ct_net(ct);
328 struct nf_conntrack_l4proto *l4proto;
330 pr_debug("destroy_conntrack(%p)\n", ct);
331 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
332 NF_CT_ASSERT(!timer_pending(&ct->timeout));
334 if (unlikely(nf_ct_is_template(ct))) {
339 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
340 if (l4proto && l4proto->destroy)
341 l4proto->destroy(ct);
346 /* Expectations will have been removed in clean_from_lists,
347 * except TFTP can create an expectation on the first packet,
348 * before connection is in the list, so we need to clean here,
351 nf_ct_remove_expectations(ct);
353 nf_ct_del_from_dying_or_unconfirmed_list(ct);
355 NF_CT_STAT_INC(net, delete);
359 nf_ct_put(ct->master);
361 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
362 nf_conntrack_free(ct);
365 static void nf_ct_delete_from_lists(struct nf_conn *ct)
367 struct net *net = nf_ct_net(ct);
368 unsigned int hash, reply_hash;
369 unsigned int sequence;
371 nf_ct_helper_destroy(ct);
375 sequence = read_seqcount_begin(&net->ct.generation);
376 hash = hash_conntrack(net,
377 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
378 reply_hash = hash_conntrack(net,
379 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
380 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
382 clean_from_lists(ct);
383 nf_conntrack_double_unlock(hash, reply_hash);
385 nf_ct_add_to_dying_list(ct);
387 NF_CT_STAT_INC(net, delete_list);
391 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
393 struct nf_conn_tstamp *tstamp;
395 tstamp = nf_conn_tstamp_find(ct);
396 if (tstamp && tstamp->stop == 0)
397 tstamp->stop = ktime_get_real_ns();
399 if (nf_ct_is_dying(ct))
402 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
403 portid, report) < 0) {
404 /* destroy event was not delivered */
405 nf_ct_delete_from_lists(ct);
406 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
410 nf_conntrack_ecache_work(nf_ct_net(ct));
411 set_bit(IPS_DYING_BIT, &ct->status);
413 nf_ct_delete_from_lists(ct);
417 EXPORT_SYMBOL_GPL(nf_ct_delete);
419 static void death_by_timeout(unsigned long ul_conntrack)
421 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
425 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
426 const struct nf_conntrack_tuple *tuple,
427 const struct nf_conntrack_zone *zone)
429 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
431 /* A conntrack can be recreated with the equal tuple,
432 * so we need to check that the conntrack is confirmed
434 return nf_ct_tuple_equal(tuple, &h->tuple) &&
435 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
436 nf_ct_is_confirmed(ct);
441 * - Caller must take a reference on returned object
442 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
444 static struct nf_conntrack_tuple_hash *
445 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
446 const struct nf_conntrack_tuple *tuple, u32 hash)
448 struct nf_conntrack_tuple_hash *h;
449 struct hlist_nulls_node *n;
450 unsigned int bucket = hash_bucket(hash, net);
452 /* Disable BHs the entire time since we normally need to disable them
453 * at least once for the stats anyway.
457 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
458 if (nf_ct_key_equal(h, tuple, zone)) {
459 NF_CT_STAT_INC(net, found);
463 NF_CT_STAT_INC(net, searched);
466 * if the nulls value we got at the end of this lookup is
467 * not the expected one, we must restart lookup.
468 * We probably met an item that was moved to another chain.
470 if (get_nulls_value(n) != bucket) {
471 NF_CT_STAT_INC(net, search_restart);
479 /* Find a connection corresponding to a tuple. */
480 static struct nf_conntrack_tuple_hash *
481 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
482 const struct nf_conntrack_tuple *tuple, u32 hash)
484 struct nf_conntrack_tuple_hash *h;
489 h = ____nf_conntrack_find(net, zone, tuple, hash);
491 ct = nf_ct_tuplehash_to_ctrack(h);
492 if (unlikely(nf_ct_is_dying(ct) ||
493 !atomic_inc_not_zero(&ct->ct_general.use)))
496 if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
507 struct nf_conntrack_tuple_hash *
508 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
509 const struct nf_conntrack_tuple *tuple)
511 return __nf_conntrack_find_get(net, zone, tuple,
512 hash_conntrack_raw(tuple));
514 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
516 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
518 unsigned int reply_hash)
520 struct net *net = nf_ct_net(ct);
522 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
523 &net->ct.hash[hash]);
524 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
525 &net->ct.hash[reply_hash]);
529 nf_conntrack_hash_check_insert(struct nf_conn *ct)
531 const struct nf_conntrack_zone *zone;
532 struct net *net = nf_ct_net(ct);
533 unsigned int hash, reply_hash;
534 struct nf_conntrack_tuple_hash *h;
535 struct hlist_nulls_node *n;
536 unsigned int sequence;
538 zone = nf_ct_zone(ct);
542 sequence = read_seqcount_begin(&net->ct.generation);
543 hash = hash_conntrack(net,
544 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
545 reply_hash = hash_conntrack(net,
546 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
547 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
549 /* See if there's one in the list already, including reverse */
550 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
551 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
553 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
556 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
557 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
559 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
563 add_timer(&ct->timeout);
565 /* The caller holds a reference to this object */
566 atomic_set(&ct->ct_general.use, 2);
567 __nf_conntrack_hash_insert(ct, hash, reply_hash);
568 nf_conntrack_double_unlock(hash, reply_hash);
569 NF_CT_STAT_INC(net, insert);
574 nf_conntrack_double_unlock(hash, reply_hash);
575 NF_CT_STAT_INC(net, insert_failed);
579 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
581 /* Confirm a connection given skb; places it in hash table */
583 __nf_conntrack_confirm(struct sk_buff *skb)
585 const struct nf_conntrack_zone *zone;
586 unsigned int hash, reply_hash;
587 struct nf_conntrack_tuple_hash *h;
589 struct nf_conn_help *help;
590 struct nf_conn_tstamp *tstamp;
591 struct hlist_nulls_node *n;
592 enum ip_conntrack_info ctinfo;
594 unsigned int sequence;
596 ct = nf_ct_get(skb, &ctinfo);
599 /* ipt_REJECT uses nf_conntrack_attach to attach related
600 ICMP/TCP RST packets in other direction. Actual packet
601 which created connection will be IP_CT_NEW or for an
602 expected connection, IP_CT_RELATED. */
603 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
606 zone = nf_ct_zone(ct);
610 sequence = read_seqcount_begin(&net->ct.generation);
611 /* reuse the hash saved before */
612 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
613 hash = hash_bucket(hash, net);
614 reply_hash = hash_conntrack(net,
615 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
617 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
619 /* We're not in hash table, and we refuse to set up related
620 * connections for unconfirmed conns. But packet copies and
621 * REJECT will give spurious warnings here.
623 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
625 /* No external references means no one else could have
628 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
629 pr_debug("Confirming conntrack %p\n", ct);
630 /* We have to check the DYING flag after unlink to prevent
631 * a race against nf_ct_get_next_corpse() possibly called from
632 * user context, else we insert an already 'dead' hash, blocking
633 * further use of that particular connection -JM.
635 nf_ct_del_from_dying_or_unconfirmed_list(ct);
637 if (unlikely(nf_ct_is_dying(ct)))
640 /* See if there's one in the list already, including reverse:
641 NAT could have grabbed it without realizing, since we're
642 not in the hash. If there is, we lost race. */
643 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
644 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
646 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
649 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
650 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
652 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
656 /* Timer relative to confirmation time, not original
657 setting time, otherwise we'd get timer wrap in
658 weird delay cases. */
659 ct->timeout.expires += jiffies;
660 add_timer(&ct->timeout);
661 atomic_inc(&ct->ct_general.use);
662 ct->status |= IPS_CONFIRMED;
664 /* set conntrack timestamp, if enabled. */
665 tstamp = nf_conn_tstamp_find(ct);
667 if (skb->tstamp.tv64 == 0)
668 __net_timestamp(skb);
670 tstamp->start = ktime_to_ns(skb->tstamp);
672 /* Since the lookup is lockless, hash insertion must be done after
673 * starting the timer and setting the CONFIRMED bit. The RCU barriers
674 * guarantee that no other CPU can find the conntrack before the above
675 * stores are visible.
677 __nf_conntrack_hash_insert(ct, hash, reply_hash);
678 nf_conntrack_double_unlock(hash, reply_hash);
679 NF_CT_STAT_INC(net, insert);
682 help = nfct_help(ct);
683 if (help && help->helper)
684 nf_conntrack_event_cache(IPCT_HELPER, ct);
686 nf_conntrack_event_cache(master_ct(ct) ?
687 IPCT_RELATED : IPCT_NEW, ct);
691 nf_ct_add_to_dying_list(ct);
692 nf_conntrack_double_unlock(hash, reply_hash);
693 NF_CT_STAT_INC(net, insert_failed);
697 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
699 /* Returns true if a connection correspondings to the tuple (required
702 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
703 const struct nf_conn *ignored_conntrack)
705 struct net *net = nf_ct_net(ignored_conntrack);
706 const struct nf_conntrack_zone *zone;
707 struct nf_conntrack_tuple_hash *h;
708 struct hlist_nulls_node *n;
712 zone = nf_ct_zone(ignored_conntrack);
713 hash = hash_conntrack(net, tuple);
715 /* Disable BHs the entire time since we need to disable them at
716 * least once for the stats anyway.
719 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
720 ct = nf_ct_tuplehash_to_ctrack(h);
721 if (ct != ignored_conntrack &&
722 nf_ct_tuple_equal(tuple, &h->tuple) &&
723 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h))) {
724 NF_CT_STAT_INC(net, found);
725 rcu_read_unlock_bh();
728 NF_CT_STAT_INC(net, searched);
730 rcu_read_unlock_bh();
734 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
736 #define NF_CT_EVICTION_RANGE 8
738 /* There's a small race here where we may free a just-assured
739 connection. Too bad: we're in trouble anyway. */
740 static noinline int early_drop(struct net *net, unsigned int _hash)
742 /* Use oldest entry, which is roughly LRU */
743 struct nf_conntrack_tuple_hash *h;
744 struct nf_conn *ct = NULL, *tmp;
745 struct hlist_nulls_node *n;
746 unsigned int i = 0, cnt = 0;
748 unsigned int hash, sequence;
753 sequence = read_seqcount_begin(&net->ct.generation);
754 hash = hash_bucket(_hash, net);
755 for (; i < net->ct.htable_size; i++) {
756 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
758 if (read_seqcount_retry(&net->ct.generation, sequence)) {
762 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
764 tmp = nf_ct_tuplehash_to_ctrack(h);
765 if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
766 !nf_ct_is_dying(tmp) &&
767 atomic_inc_not_zero(&tmp->ct_general.use)) {
774 hash = (hash + 1) % net->ct.htable_size;
777 if (ct || cnt >= NF_CT_EVICTION_RANGE)
786 if (del_timer(&ct->timeout)) {
787 if (nf_ct_delete(ct, 0, 0)) {
789 NF_CT_STAT_INC_ATOMIC(net, early_drop);
796 void init_nf_conntrack_hash_rnd(void)
801 * Why not initialize nf_conntrack_rnd in a "init()" function ?
802 * Because there isn't enough entropy when system initializing,
803 * and we initialize it as late as possible.
806 get_random_bytes(&rand, sizeof(rand));
808 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
811 static struct nf_conn *
812 __nf_conntrack_alloc(struct net *net,
813 const struct nf_conntrack_zone *zone,
814 const struct nf_conntrack_tuple *orig,
815 const struct nf_conntrack_tuple *repl,
820 if (unlikely(!nf_conntrack_hash_rnd)) {
821 init_nf_conntrack_hash_rnd();
822 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
823 hash = hash_conntrack_raw(orig);
826 /* We don't want any race condition at early drop stage */
827 atomic_inc(&net->ct.count);
829 if (nf_conntrack_max &&
830 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
831 if (!early_drop(net, hash)) {
832 atomic_dec(&net->ct.count);
833 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
834 return ERR_PTR(-ENOMEM);
839 * Do not use kmem_cache_zalloc(), as this cache uses
840 * SLAB_DESTROY_BY_RCU.
842 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
846 spin_lock_init(&ct->lock);
847 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
848 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
849 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
850 /* save hash for reusing when confirming */
851 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
853 /* Don't set timer yet: wait for confirmation */
854 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
855 write_pnet(&ct->ct_net, net);
856 memset(&ct->__nfct_init_offset[0], 0,
857 offsetof(struct nf_conn, proto) -
858 offsetof(struct nf_conn, __nfct_init_offset[0]));
860 if (zone && nf_ct_zone_add(ct, GFP_ATOMIC, zone) < 0)
863 /* Because we use RCU lookups, we set ct_general.use to zero before
864 * this is inserted in any list.
866 atomic_set(&ct->ct_general.use, 0);
869 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
871 atomic_dec(&net->ct.count);
872 return ERR_PTR(-ENOMEM);
875 struct nf_conn *nf_conntrack_alloc(struct net *net,
876 const struct nf_conntrack_zone *zone,
877 const struct nf_conntrack_tuple *orig,
878 const struct nf_conntrack_tuple *repl,
881 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
883 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
885 void nf_conntrack_free(struct nf_conn *ct)
887 struct net *net = nf_ct_net(ct);
889 /* A freed object has refcnt == 0, that's
890 * the golden rule for SLAB_DESTROY_BY_RCU
892 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
894 nf_ct_ext_destroy(ct);
896 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
897 smp_mb__before_atomic();
898 atomic_dec(&net->ct.count);
900 EXPORT_SYMBOL_GPL(nf_conntrack_free);
903 /* Allocate a new conntrack: we return -ENOMEM if classification
904 failed due to stress. Otherwise it really is unclassifiable. */
905 static struct nf_conntrack_tuple_hash *
906 init_conntrack(struct net *net, struct nf_conn *tmpl,
907 const struct nf_conntrack_tuple *tuple,
908 struct nf_conntrack_l3proto *l3proto,
909 struct nf_conntrack_l4proto *l4proto,
911 unsigned int dataoff, u32 hash)
914 struct nf_conn_help *help;
915 struct nf_conntrack_tuple repl_tuple;
916 struct nf_conntrack_ecache *ecache;
917 struct nf_conntrack_expect *exp = NULL;
918 const struct nf_conntrack_zone *zone;
919 struct nf_conn_timeout *timeout_ext;
920 struct nf_conntrack_zone tmp;
921 unsigned int *timeouts;
923 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
924 pr_debug("Can't invert tuple.\n");
928 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
929 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
932 return (struct nf_conntrack_tuple_hash *)ct;
934 if (tmpl && nfct_synproxy(tmpl)) {
935 nfct_seqadj_ext_add(ct);
936 nfct_synproxy_ext_add(ct);
939 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
941 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
943 timeouts = l4proto->get_timeouts(net);
945 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
946 nf_conntrack_free(ct);
947 pr_debug("init conntrack: can't track with proto module\n");
952 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
954 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
955 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
956 nf_ct_labels_ext_add(ct);
958 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
959 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
960 ecache ? ecache->expmask : 0,
964 if (net->ct.expect_count) {
965 spin_lock(&nf_conntrack_expect_lock);
966 exp = nf_ct_find_expectation(net, zone, tuple);
968 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
970 /* Welcome, Mr. Bond. We've been expecting you... */
971 __set_bit(IPS_EXPECTED_BIT, &ct->status);
972 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
973 ct->master = exp->master;
975 help = nf_ct_helper_ext_add(ct, exp->helper,
978 rcu_assign_pointer(help->helper, exp->helper);
981 #ifdef CONFIG_NF_CONNTRACK_MARK
982 ct->mark = exp->master->mark;
984 #ifdef CONFIG_NF_CONNTRACK_SECMARK
985 ct->secmark = exp->master->secmark;
987 NF_CT_STAT_INC(net, expect_new);
989 spin_unlock(&nf_conntrack_expect_lock);
992 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
993 NF_CT_STAT_INC(net, new);
996 /* Now it is inserted into the unconfirmed list, bump refcount */
997 nf_conntrack_get(&ct->ct_general);
998 nf_ct_add_to_unconfirmed_list(ct);
1004 exp->expectfn(ct, exp);
1005 nf_ct_expect_put(exp);
1008 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1011 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1012 static inline struct nf_conn *
1013 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1014 struct sk_buff *skb,
1015 unsigned int dataoff,
1018 struct nf_conntrack_l3proto *l3proto,
1019 struct nf_conntrack_l4proto *l4proto,
1021 enum ip_conntrack_info *ctinfo)
1023 const struct nf_conntrack_zone *zone;
1024 struct nf_conntrack_tuple tuple;
1025 struct nf_conntrack_tuple_hash *h;
1026 struct nf_conntrack_zone tmp;
1030 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1031 dataoff, l3num, protonum, &tuple, l3proto,
1033 pr_debug("resolve_normal_ct: Can't get tuple\n");
1037 /* look for tuple match */
1038 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1039 hash = hash_conntrack_raw(&tuple);
1040 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1042 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1043 skb, dataoff, hash);
1049 ct = nf_ct_tuplehash_to_ctrack(h);
1051 /* It exists; we have (non-exclusive) reference. */
1052 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1053 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1054 /* Please set reply bit if this packet OK */
1057 /* Once we've had two way comms, always ESTABLISHED. */
1058 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1059 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
1060 *ctinfo = IP_CT_ESTABLISHED;
1061 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1062 pr_debug("nf_conntrack_in: related packet for %p\n",
1064 *ctinfo = IP_CT_RELATED;
1066 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
1067 *ctinfo = IP_CT_NEW;
1071 skb->nfct = &ct->ct_general;
1072 skb->nfctinfo = *ctinfo;
1077 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1078 struct sk_buff *skb)
1080 struct nf_conn *ct, *tmpl = NULL;
1081 enum ip_conntrack_info ctinfo;
1082 struct nf_conntrack_l3proto *l3proto;
1083 struct nf_conntrack_l4proto *l4proto;
1084 unsigned int *timeouts;
1085 unsigned int dataoff;
1091 /* Previously seen (loopback or untracked)? Ignore. */
1092 tmpl = (struct nf_conn *)skb->nfct;
1093 if (!nf_ct_is_template(tmpl)) {
1094 NF_CT_STAT_INC_ATOMIC(net, ignore);
1100 /* rcu_read_lock()ed by nf_hook_slow */
1101 l3proto = __nf_ct_l3proto_find(pf);
1102 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1103 &dataoff, &protonum);
1105 pr_debug("not prepared to track yet or error occurred\n");
1106 NF_CT_STAT_INC_ATOMIC(net, error);
1107 NF_CT_STAT_INC_ATOMIC(net, invalid);
1112 l4proto = __nf_ct_l4proto_find(pf, protonum);
1114 /* It may be an special packet, error, unclean...
1115 * inverse of the return code tells to the netfilter
1116 * core what to do with the packet. */
1117 if (l4proto->error != NULL) {
1118 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1121 NF_CT_STAT_INC_ATOMIC(net, error);
1122 NF_CT_STAT_INC_ATOMIC(net, invalid);
1126 /* ICMP[v6] protocol trackers may assign one conntrack. */
1131 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1132 l3proto, l4proto, &set_reply, &ctinfo);
1134 /* Not valid part of a connection */
1135 NF_CT_STAT_INC_ATOMIC(net, invalid);
1141 /* Too stressed to deal. */
1142 NF_CT_STAT_INC_ATOMIC(net, drop);
1147 NF_CT_ASSERT(skb->nfct);
1149 /* Decide what timeout policy we want to apply to this flow. */
1150 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1152 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1154 /* Invalid: inverse of the return code tells
1155 * the netfilter core what to do */
1156 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1157 nf_conntrack_put(skb->nfct);
1159 NF_CT_STAT_INC_ATOMIC(net, invalid);
1160 if (ret == -NF_DROP)
1161 NF_CT_STAT_INC_ATOMIC(net, drop);
1166 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1167 nf_conntrack_event_cache(IPCT_REPLY, ct);
1170 /* Special case: we have to repeat this hook, assign the
1171 * template again to this packet. We assume that this packet
1172 * has no conntrack assigned. This is used by nf_ct_tcp. */
1173 if (ret == NF_REPEAT)
1174 skb->nfct = (struct nf_conntrack *)tmpl;
1181 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1183 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1184 const struct nf_conntrack_tuple *orig)
1189 ret = nf_ct_invert_tuple(inverse, orig,
1190 __nf_ct_l3proto_find(orig->src.l3num),
1191 __nf_ct_l4proto_find(orig->src.l3num,
1192 orig->dst.protonum));
1196 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1198 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1199 implicitly racy: see __nf_conntrack_confirm */
1200 void nf_conntrack_alter_reply(struct nf_conn *ct,
1201 const struct nf_conntrack_tuple *newreply)
1203 struct nf_conn_help *help = nfct_help(ct);
1205 /* Should be unconfirmed, so not in hash table yet */
1206 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1208 pr_debug("Altering reply tuple of %p to ", ct);
1209 nf_ct_dump_tuple(newreply);
1211 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1212 if (ct->master || (help && !hlist_empty(&help->expectations)))
1216 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1219 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1221 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1222 void __nf_ct_refresh_acct(struct nf_conn *ct,
1223 enum ip_conntrack_info ctinfo,
1224 const struct sk_buff *skb,
1225 unsigned long extra_jiffies,
1228 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1231 /* Only update if this is not a fixed timeout */
1232 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1235 /* If not in hash table, timer will not be active yet */
1236 if (!nf_ct_is_confirmed(ct)) {
1237 ct->timeout.expires = extra_jiffies;
1239 unsigned long newtime = jiffies + extra_jiffies;
1241 /* Only update the timeout if the new timeout is at least
1242 HZ jiffies from the old timeout. Need del_timer for race
1243 avoidance (may already be dying). */
1244 if (newtime - ct->timeout.expires >= HZ)
1245 mod_timer_pending(&ct->timeout, newtime);
1250 struct nf_conn_acct *acct;
1252 acct = nf_conn_acct_find(ct);
1254 struct nf_conn_counter *counter = acct->counter;
1256 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1257 atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1261 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1263 bool __nf_ct_kill_acct(struct nf_conn *ct,
1264 enum ip_conntrack_info ctinfo,
1265 const struct sk_buff *skb,
1269 struct nf_conn_acct *acct;
1271 acct = nf_conn_acct_find(ct);
1273 struct nf_conn_counter *counter = acct->counter;
1275 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1276 atomic64_add(skb->len - skb_network_offset(skb),
1277 &counter[CTINFO2DIR(ctinfo)].bytes);
1281 if (del_timer(&ct->timeout)) {
1282 ct->timeout.function((unsigned long)ct);
1287 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1289 #ifdef CONFIG_NF_CONNTRACK_ZONES
1290 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1291 .len = sizeof(struct nf_conntrack_zone),
1292 .align = __alignof__(struct nf_conntrack_zone),
1293 .id = NF_CT_EXT_ZONE,
1297 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1299 #include <linux/netfilter/nfnetlink.h>
1300 #include <linux/netfilter/nfnetlink_conntrack.h>
1301 #include <linux/mutex.h>
1303 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1304 * in ip_conntrack_core, since we don't want the protocols to autoload
1305 * or depend on ctnetlink */
1306 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1307 const struct nf_conntrack_tuple *tuple)
1309 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1310 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1311 goto nla_put_failure;
1317 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1319 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1320 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1321 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1323 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1325 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1326 struct nf_conntrack_tuple *t)
1328 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1331 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1332 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1336 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1338 int nf_ct_port_nlattr_tuple_size(void)
1340 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1342 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1345 /* Used by ipt_REJECT and ip6t_REJECT. */
1346 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1349 enum ip_conntrack_info ctinfo;
1351 /* This ICMP is in reverse direction to the packet which caused it */
1352 ct = nf_ct_get(skb, &ctinfo);
1353 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1354 ctinfo = IP_CT_RELATED_REPLY;
1356 ctinfo = IP_CT_RELATED;
1358 /* Attach to new skbuff, and increment count */
1359 nskb->nfct = &ct->ct_general;
1360 nskb->nfctinfo = ctinfo;
1361 nf_conntrack_get(nskb->nfct);
1364 /* Bring out ya dead! */
1365 static struct nf_conn *
1366 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1367 void *data, unsigned int *bucket)
1369 struct nf_conntrack_tuple_hash *h;
1371 struct hlist_nulls_node *n;
1375 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1376 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1379 if (*bucket < net->ct.htable_size) {
1380 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1381 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1383 ct = nf_ct_tuplehash_to_ctrack(h);
1392 for_each_possible_cpu(cpu) {
1393 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1395 spin_lock_bh(&pcpu->lock);
1396 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1397 ct = nf_ct_tuplehash_to_ctrack(h);
1399 set_bit(IPS_DYING_BIT, &ct->status);
1401 spin_unlock_bh(&pcpu->lock);
1405 atomic_inc(&ct->ct_general.use);
1411 void nf_ct_iterate_cleanup(struct net *net,
1412 int (*iter)(struct nf_conn *i, void *data),
1413 void *data, u32 portid, int report)
1416 unsigned int bucket = 0;
1418 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1419 /* Time to push up daises... */
1420 if (del_timer(&ct->timeout))
1421 nf_ct_delete(ct, portid, report);
1423 /* ... else the timer will get him soon. */
1428 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1430 static int kill_all(struct nf_conn *i, void *data)
1435 void nf_ct_free_hashtable(void *hash, unsigned int size)
1437 if (is_vmalloc_addr(hash))
1440 free_pages((unsigned long)hash,
1441 get_order(sizeof(struct hlist_head) * size));
1443 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1445 static int untrack_refs(void)
1449 for_each_possible_cpu(cpu) {
1450 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1452 cnt += atomic_read(&ct->ct_general.use) - 1;
1457 void nf_conntrack_cleanup_start(void)
1459 RCU_INIT_POINTER(ip_ct_attach, NULL);
1462 void nf_conntrack_cleanup_end(void)
1464 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1465 while (untrack_refs() > 0)
1468 #ifdef CONFIG_NF_CONNTRACK_ZONES
1469 nf_ct_extend_unregister(&nf_ct_zone_extend);
1471 nf_conntrack_proto_fini();
1472 nf_conntrack_seqadj_fini();
1473 nf_conntrack_labels_fini();
1474 nf_conntrack_helper_fini();
1475 nf_conntrack_timeout_fini();
1476 nf_conntrack_ecache_fini();
1477 nf_conntrack_tstamp_fini();
1478 nf_conntrack_acct_fini();
1479 nf_conntrack_expect_fini();
1483 * Mishearing the voices in his head, our hero wonders how he's
1484 * supposed to kill the mall.
1486 void nf_conntrack_cleanup_net(struct net *net)
1490 list_add(&net->exit_list, &single);
1491 nf_conntrack_cleanup_net_list(&single);
1494 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1500 * This makes sure all current packets have passed through
1501 * netfilter framework. Roll on, two-stage module
1507 list_for_each_entry(net, net_exit_list, exit_list) {
1508 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1509 if (atomic_read(&net->ct.count) != 0)
1514 goto i_see_dead_people;
1517 list_for_each_entry(net, net_exit_list, exit_list) {
1518 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1519 nf_conntrack_proto_pernet_fini(net);
1520 nf_conntrack_helper_pernet_fini(net);
1521 nf_conntrack_ecache_pernet_fini(net);
1522 nf_conntrack_tstamp_pernet_fini(net);
1523 nf_conntrack_acct_pernet_fini(net);
1524 nf_conntrack_expect_pernet_fini(net);
1525 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1526 kfree(net->ct.slabname);
1527 free_percpu(net->ct.stat);
1528 free_percpu(net->ct.pcpu_lists);
1532 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1534 struct hlist_nulls_head *hash;
1535 unsigned int nr_slots, i;
1538 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1539 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1540 sz = nr_slots * sizeof(struct hlist_nulls_head);
1541 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1547 for (i = 0; i < nr_slots; i++)
1548 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1552 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1554 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1557 unsigned int hashsize, old_size;
1558 struct hlist_nulls_head *hash, *old_hash;
1559 struct nf_conntrack_tuple_hash *h;
1562 if (current->nsproxy->net_ns != &init_net)
1565 /* On boot, we can set this without any fancy locking. */
1566 if (!nf_conntrack_htable_size)
1567 return param_set_uint(val, kp);
1569 rc = kstrtouint(val, 0, &hashsize);
1575 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1580 nf_conntrack_all_lock();
1581 write_seqcount_begin(&init_net.ct.generation);
1583 /* Lookups in the old hash might happen in parallel, which means we
1584 * might get false negatives during connection lookup. New connections
1585 * created because of a false negative won't make it into the hash
1586 * though since that required taking the locks.
1589 for (i = 0; i < init_net.ct.htable_size; i++) {
1590 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1591 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1592 struct nf_conntrack_tuple_hash, hnnode);
1593 ct = nf_ct_tuplehash_to_ctrack(h);
1594 hlist_nulls_del_rcu(&h->hnnode);
1595 bucket = __hash_conntrack(&h->tuple, hashsize);
1596 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1599 old_size = init_net.ct.htable_size;
1600 old_hash = init_net.ct.hash;
1602 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1603 init_net.ct.hash = hash;
1605 write_seqcount_end(&init_net.ct.generation);
1606 nf_conntrack_all_unlock();
1609 nf_ct_free_hashtable(old_hash, old_size);
1612 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1614 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1615 &nf_conntrack_htable_size, 0600);
1617 void nf_ct_untracked_status_or(unsigned long bits)
1621 for_each_possible_cpu(cpu)
1622 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1624 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1626 int nf_conntrack_init_start(void)
1631 for (i = 0; i < CONNTRACK_LOCKS; i++)
1632 spin_lock_init(&nf_conntrack_locks[i]);
1634 if (!nf_conntrack_htable_size) {
1635 /* Idea from tcp.c: use 1/16384 of memory.
1636 * On i386: 32MB machine has 512 buckets.
1637 * >= 1GB machines have 16384 buckets.
1638 * >= 4GB machines have 65536 buckets.
1640 nf_conntrack_htable_size
1641 = (((totalram_pages << PAGE_SHIFT) / 16384)
1642 / sizeof(struct hlist_head));
1643 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1644 nf_conntrack_htable_size = 65536;
1645 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1646 nf_conntrack_htable_size = 16384;
1647 if (nf_conntrack_htable_size < 32)
1648 nf_conntrack_htable_size = 32;
1650 /* Use a max. factor of four by default to get the same max as
1651 * with the old struct list_heads. When a table size is given
1652 * we use the old value of 8 to avoid reducing the max.
1656 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1658 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1659 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1662 ret = nf_conntrack_expect_init();
1666 ret = nf_conntrack_acct_init();
1670 ret = nf_conntrack_tstamp_init();
1674 ret = nf_conntrack_ecache_init();
1678 ret = nf_conntrack_timeout_init();
1682 ret = nf_conntrack_helper_init();
1686 ret = nf_conntrack_labels_init();
1690 ret = nf_conntrack_seqadj_init();
1694 #ifdef CONFIG_NF_CONNTRACK_ZONES
1695 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1699 ret = nf_conntrack_proto_init();
1703 /* Set up fake conntrack: to never be deleted, not in any hashes */
1704 for_each_possible_cpu(cpu) {
1705 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1706 write_pnet(&ct->ct_net, &init_net);
1707 atomic_set(&ct->ct_general.use, 1);
1709 /* - and look it like as a confirmed connection */
1710 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1714 #ifdef CONFIG_NF_CONNTRACK_ZONES
1715 nf_ct_extend_unregister(&nf_ct_zone_extend);
1718 nf_conntrack_seqadj_fini();
1720 nf_conntrack_labels_fini();
1722 nf_conntrack_helper_fini();
1724 nf_conntrack_timeout_fini();
1726 nf_conntrack_ecache_fini();
1728 nf_conntrack_tstamp_fini();
1730 nf_conntrack_acct_fini();
1732 nf_conntrack_expect_fini();
1737 void nf_conntrack_init_end(void)
1739 /* For use by REJECT target */
1740 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1741 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1745 * We need to use special "null" values, not used in hash table
1747 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1748 #define DYING_NULLS_VAL ((1<<30)+1)
1749 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1751 int nf_conntrack_init_net(struct net *net)
1756 atomic_set(&net->ct.count, 0);
1757 seqcount_init(&net->ct.generation);
1759 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1760 if (!net->ct.pcpu_lists)
1763 for_each_possible_cpu(cpu) {
1764 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1766 spin_lock_init(&pcpu->lock);
1767 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1768 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1771 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1773 goto err_pcpu_lists;
1775 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1776 if (!net->ct.slabname)
1779 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1780 sizeof(struct nf_conn), 0,
1781 SLAB_DESTROY_BY_RCU, NULL);
1782 if (!net->ct.nf_conntrack_cachep) {
1783 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1787 net->ct.htable_size = nf_conntrack_htable_size;
1788 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1789 if (!net->ct.hash) {
1790 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1793 ret = nf_conntrack_expect_pernet_init(net);
1796 ret = nf_conntrack_acct_pernet_init(net);
1799 ret = nf_conntrack_tstamp_pernet_init(net);
1802 ret = nf_conntrack_ecache_pernet_init(net);
1805 ret = nf_conntrack_helper_pernet_init(net);
1808 ret = nf_conntrack_proto_pernet_init(net);
1814 nf_conntrack_helper_pernet_fini(net);
1816 nf_conntrack_ecache_pernet_fini(net);
1818 nf_conntrack_tstamp_pernet_fini(net);
1820 nf_conntrack_acct_pernet_fini(net);
1822 nf_conntrack_expect_pernet_fini(net);
1824 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1826 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1828 kfree(net->ct.slabname);
1830 free_percpu(net->ct.stat);
1832 free_percpu(net->ct.pcpu_lists);