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, u16 zone)
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, zone ^ 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,
154 u16 zone, unsigned int size)
156 return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
159 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
160 const struct nf_conntrack_tuple *tuple)
162 return __hash_conntrack(tuple, zone, 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, u16 zone, gfp_t flags)
293 struct nf_conn *tmpl;
295 tmpl = kzalloc(sizeof(struct nf_conn), GFP_KERNEL);
299 tmpl->status = IPS_TEMPLATE;
300 write_pnet(&tmpl->ct_net, net);
302 #ifdef CONFIG_NF_CONNTRACK_ZONES
304 struct nf_conntrack_zone *nf_ct_zone;
306 nf_ct_zone = nf_ct_ext_add(tmpl, NF_CT_EXT_ZONE, GFP_ATOMIC);
309 nf_ct_zone->id = zone;
312 atomic_set(&tmpl->ct_general.use, 0);
315 #ifdef CONFIG_NF_CONNTRACK_ZONES
321 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
323 static void nf_ct_tmpl_free(struct nf_conn *tmpl)
325 nf_ct_ext_destroy(tmpl);
326 nf_ct_ext_free(tmpl);
331 destroy_conntrack(struct nf_conntrack *nfct)
333 struct nf_conn *ct = (struct nf_conn *)nfct;
334 struct net *net = nf_ct_net(ct);
335 struct nf_conntrack_l4proto *l4proto;
337 pr_debug("destroy_conntrack(%p)\n", ct);
338 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
339 NF_CT_ASSERT(!timer_pending(&ct->timeout));
341 if (unlikely(nf_ct_is_template(ct))) {
346 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
347 if (l4proto && l4proto->destroy)
348 l4proto->destroy(ct);
353 /* Expectations will have been removed in clean_from_lists,
354 * except TFTP can create an expectation on the first packet,
355 * before connection is in the list, so we need to clean here,
358 nf_ct_remove_expectations(ct);
360 nf_ct_del_from_dying_or_unconfirmed_list(ct);
362 NF_CT_STAT_INC(net, delete);
366 nf_ct_put(ct->master);
368 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
369 nf_conntrack_free(ct);
372 static void nf_ct_delete_from_lists(struct nf_conn *ct)
374 struct net *net = nf_ct_net(ct);
375 unsigned int hash, reply_hash;
376 u16 zone = nf_ct_zone(ct);
377 unsigned int sequence;
379 nf_ct_helper_destroy(ct);
383 sequence = read_seqcount_begin(&net->ct.generation);
384 hash = hash_conntrack(net, zone,
385 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
386 reply_hash = hash_conntrack(net, zone,
387 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
388 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
390 clean_from_lists(ct);
391 nf_conntrack_double_unlock(hash, reply_hash);
393 nf_ct_add_to_dying_list(ct);
395 NF_CT_STAT_INC(net, delete_list);
399 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
401 struct nf_conn_tstamp *tstamp;
403 tstamp = nf_conn_tstamp_find(ct);
404 if (tstamp && tstamp->stop == 0)
405 tstamp->stop = ktime_get_real_ns();
407 if (nf_ct_is_dying(ct))
410 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
411 portid, report) < 0) {
412 /* destroy event was not delivered */
413 nf_ct_delete_from_lists(ct);
414 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
418 nf_conntrack_ecache_work(nf_ct_net(ct));
419 set_bit(IPS_DYING_BIT, &ct->status);
421 nf_ct_delete_from_lists(ct);
425 EXPORT_SYMBOL_GPL(nf_ct_delete);
427 static void death_by_timeout(unsigned long ul_conntrack)
429 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
433 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
434 const struct nf_conntrack_tuple *tuple,
437 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
439 /* A conntrack can be recreated with the equal tuple,
440 * so we need to check that the conntrack is confirmed
442 return nf_ct_tuple_equal(tuple, &h->tuple) &&
443 nf_ct_zone(ct) == zone &&
444 nf_ct_is_confirmed(ct);
449 * - Caller must take a reference on returned object
450 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
452 static struct nf_conntrack_tuple_hash *
453 ____nf_conntrack_find(struct net *net, u16 zone,
454 const struct nf_conntrack_tuple *tuple, u32 hash)
456 struct nf_conntrack_tuple_hash *h;
457 struct hlist_nulls_node *n;
458 unsigned int bucket = hash_bucket(hash, net);
460 /* Disable BHs the entire time since we normally need to disable them
461 * at least once for the stats anyway.
465 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
466 if (nf_ct_key_equal(h, tuple, zone)) {
467 NF_CT_STAT_INC(net, found);
471 NF_CT_STAT_INC(net, searched);
474 * if the nulls value we got at the end of this lookup is
475 * not the expected one, we must restart lookup.
476 * We probably met an item that was moved to another chain.
478 if (get_nulls_value(n) != bucket) {
479 NF_CT_STAT_INC(net, search_restart);
487 /* Find a connection corresponding to a tuple. */
488 static struct nf_conntrack_tuple_hash *
489 __nf_conntrack_find_get(struct net *net, u16 zone,
490 const struct nf_conntrack_tuple *tuple, u32 hash)
492 struct nf_conntrack_tuple_hash *h;
497 h = ____nf_conntrack_find(net, zone, tuple, hash);
499 ct = nf_ct_tuplehash_to_ctrack(h);
500 if (unlikely(nf_ct_is_dying(ct) ||
501 !atomic_inc_not_zero(&ct->ct_general.use)))
504 if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
515 struct nf_conntrack_tuple_hash *
516 nf_conntrack_find_get(struct net *net, u16 zone,
517 const struct nf_conntrack_tuple *tuple)
519 return __nf_conntrack_find_get(net, zone, tuple,
520 hash_conntrack_raw(tuple, zone));
522 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
524 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
526 unsigned int reply_hash)
528 struct net *net = nf_ct_net(ct);
530 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
531 &net->ct.hash[hash]);
532 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
533 &net->ct.hash[reply_hash]);
537 nf_conntrack_hash_check_insert(struct nf_conn *ct)
539 struct net *net = nf_ct_net(ct);
540 unsigned int hash, reply_hash;
541 struct nf_conntrack_tuple_hash *h;
542 struct hlist_nulls_node *n;
544 unsigned int sequence;
546 zone = nf_ct_zone(ct);
550 sequence = read_seqcount_begin(&net->ct.generation);
551 hash = hash_conntrack(net, zone,
552 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
553 reply_hash = hash_conntrack(net, zone,
554 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
555 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
557 /* See if there's one in the list already, including reverse */
558 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
559 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
561 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
563 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
564 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
566 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
569 add_timer(&ct->timeout);
571 /* The caller holds a reference to this object */
572 atomic_set(&ct->ct_general.use, 2);
573 __nf_conntrack_hash_insert(ct, hash, reply_hash);
574 nf_conntrack_double_unlock(hash, reply_hash);
575 NF_CT_STAT_INC(net, insert);
580 nf_conntrack_double_unlock(hash, reply_hash);
581 NF_CT_STAT_INC(net, insert_failed);
585 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
587 /* Confirm a connection given skb; places it in hash table */
589 __nf_conntrack_confirm(struct sk_buff *skb)
591 unsigned int hash, reply_hash;
592 struct nf_conntrack_tuple_hash *h;
594 struct nf_conn_help *help;
595 struct nf_conn_tstamp *tstamp;
596 struct hlist_nulls_node *n;
597 enum ip_conntrack_info ctinfo;
600 unsigned int sequence;
602 ct = nf_ct_get(skb, &ctinfo);
605 /* ipt_REJECT uses nf_conntrack_attach to attach related
606 ICMP/TCP RST packets in other direction. Actual packet
607 which created connection will be IP_CT_NEW or for an
608 expected connection, IP_CT_RELATED. */
609 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
612 zone = nf_ct_zone(ct);
616 sequence = read_seqcount_begin(&net->ct.generation);
617 /* reuse the hash saved before */
618 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
619 hash = hash_bucket(hash, net);
620 reply_hash = hash_conntrack(net, zone,
621 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
623 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
625 /* We're not in hash table, and we refuse to set up related
626 * connections for unconfirmed conns. But packet copies and
627 * REJECT will give spurious warnings here.
629 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
631 /* No external references means no one else could have
634 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
635 pr_debug("Confirming conntrack %p\n", ct);
636 /* We have to check the DYING flag after unlink to prevent
637 * a race against nf_ct_get_next_corpse() possibly called from
638 * user context, else we insert an already 'dead' hash, blocking
639 * further use of that particular connection -JM.
641 nf_ct_del_from_dying_or_unconfirmed_list(ct);
643 if (unlikely(nf_ct_is_dying(ct)))
646 /* See if there's one in the list already, including reverse:
647 NAT could have grabbed it without realizing, since we're
648 not in the hash. If there is, we lost race. */
649 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
650 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
652 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
654 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
655 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
657 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
660 /* Timer relative to confirmation time, not original
661 setting time, otherwise we'd get timer wrap in
662 weird delay cases. */
663 ct->timeout.expires += jiffies;
664 add_timer(&ct->timeout);
665 atomic_inc(&ct->ct_general.use);
666 ct->status |= IPS_CONFIRMED;
668 /* set conntrack timestamp, if enabled. */
669 tstamp = nf_conn_tstamp_find(ct);
671 if (skb->tstamp.tv64 == 0)
672 __net_timestamp(skb);
674 tstamp->start = ktime_to_ns(skb->tstamp);
676 /* Since the lookup is lockless, hash insertion must be done after
677 * starting the timer and setting the CONFIRMED bit. The RCU barriers
678 * guarantee that no other CPU can find the conntrack before the above
679 * stores are visible.
681 __nf_conntrack_hash_insert(ct, hash, reply_hash);
682 nf_conntrack_double_unlock(hash, reply_hash);
683 NF_CT_STAT_INC(net, insert);
686 help = nfct_help(ct);
687 if (help && help->helper)
688 nf_conntrack_event_cache(IPCT_HELPER, ct);
690 nf_conntrack_event_cache(master_ct(ct) ?
691 IPCT_RELATED : IPCT_NEW, ct);
695 nf_ct_add_to_dying_list(ct);
696 nf_conntrack_double_unlock(hash, reply_hash);
697 NF_CT_STAT_INC(net, insert_failed);
701 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
703 /* Returns true if a connection correspondings to the tuple (required
706 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
707 const struct nf_conn *ignored_conntrack)
709 struct net *net = nf_ct_net(ignored_conntrack);
710 struct nf_conntrack_tuple_hash *h;
711 struct hlist_nulls_node *n;
713 u16 zone = nf_ct_zone(ignored_conntrack);
714 unsigned int hash = hash_conntrack(net, zone, tuple);
716 /* Disable BHs the entire time since we need to disable them at
717 * least once for the stats anyway.
720 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
721 ct = nf_ct_tuplehash_to_ctrack(h);
722 if (ct != ignored_conntrack &&
723 nf_ct_tuple_equal(tuple, &h->tuple) &&
724 nf_ct_zone(ct) == zone) {
725 NF_CT_STAT_INC(net, found);
726 rcu_read_unlock_bh();
729 NF_CT_STAT_INC(net, searched);
731 rcu_read_unlock_bh();
735 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
737 #define NF_CT_EVICTION_RANGE 8
739 /* There's a small race here where we may free a just-assured
740 connection. Too bad: we're in trouble anyway. */
741 static noinline int early_drop(struct net *net, unsigned int _hash)
743 /* Use oldest entry, which is roughly LRU */
744 struct nf_conntrack_tuple_hash *h;
745 struct nf_conn *ct = NULL, *tmp;
746 struct hlist_nulls_node *n;
747 unsigned int i = 0, cnt = 0;
749 unsigned int hash, sequence;
754 sequence = read_seqcount_begin(&net->ct.generation);
755 hash = hash_bucket(_hash, net);
756 for (; i < net->ct.htable_size; i++) {
757 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
759 if (read_seqcount_retry(&net->ct.generation, sequence)) {
763 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
765 tmp = nf_ct_tuplehash_to_ctrack(h);
766 if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
767 !nf_ct_is_dying(tmp) &&
768 atomic_inc_not_zero(&tmp->ct_general.use)) {
775 hash = (hash + 1) % net->ct.htable_size;
778 if (ct || cnt >= NF_CT_EVICTION_RANGE)
787 if (del_timer(&ct->timeout)) {
788 if (nf_ct_delete(ct, 0, 0)) {
790 NF_CT_STAT_INC_ATOMIC(net, early_drop);
797 void init_nf_conntrack_hash_rnd(void)
802 * Why not initialize nf_conntrack_rnd in a "init()" function ?
803 * Because there isn't enough entropy when system initializing,
804 * and we initialize it as late as possible.
807 get_random_bytes(&rand, sizeof(rand));
809 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
812 static struct nf_conn *
813 __nf_conntrack_alloc(struct net *net, u16 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, zone);
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);
844 atomic_dec(&net->ct.count);
845 return ERR_PTR(-ENOMEM);
847 spin_lock_init(&ct->lock);
848 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
849 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
850 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
851 /* save hash for reusing when confirming */
852 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
854 /* Don't set timer yet: wait for confirmation */
855 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
856 write_pnet(&ct->ct_net, net);
857 memset(&ct->__nfct_init_offset[0], 0,
858 offsetof(struct nf_conn, proto) -
859 offsetof(struct nf_conn, __nfct_init_offset[0]));
860 #ifdef CONFIG_NF_CONNTRACK_ZONES
862 struct nf_conntrack_zone *nf_ct_zone;
864 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
867 nf_ct_zone->id = zone;
870 /* Because we use RCU lookups, we set ct_general.use to zero before
871 * this is inserted in any list.
873 atomic_set(&ct->ct_general.use, 0);
876 #ifdef CONFIG_NF_CONNTRACK_ZONES
878 atomic_dec(&net->ct.count);
879 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
880 return ERR_PTR(-ENOMEM);
884 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
885 const struct nf_conntrack_tuple *orig,
886 const struct nf_conntrack_tuple *repl,
889 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
891 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
893 void nf_conntrack_free(struct nf_conn *ct)
895 struct net *net = nf_ct_net(ct);
897 /* A freed object has refcnt == 0, that's
898 * the golden rule for SLAB_DESTROY_BY_RCU
900 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
902 nf_ct_ext_destroy(ct);
904 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
905 smp_mb__before_atomic();
906 atomic_dec(&net->ct.count);
908 EXPORT_SYMBOL_GPL(nf_conntrack_free);
911 /* Allocate a new conntrack: we return -ENOMEM if classification
912 failed due to stress. Otherwise it really is unclassifiable. */
913 static struct nf_conntrack_tuple_hash *
914 init_conntrack(struct net *net, struct nf_conn *tmpl,
915 const struct nf_conntrack_tuple *tuple,
916 struct nf_conntrack_l3proto *l3proto,
917 struct nf_conntrack_l4proto *l4proto,
919 unsigned int dataoff, u32 hash)
922 struct nf_conn_help *help;
923 struct nf_conntrack_tuple repl_tuple;
924 struct nf_conntrack_ecache *ecache;
925 struct nf_conntrack_expect *exp = NULL;
926 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
927 struct nf_conn_timeout *timeout_ext;
928 unsigned int *timeouts;
930 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
931 pr_debug("Can't invert tuple.\n");
935 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
938 return (struct nf_conntrack_tuple_hash *)ct;
940 if (tmpl && nfct_synproxy(tmpl)) {
941 nfct_seqadj_ext_add(ct);
942 nfct_synproxy_ext_add(ct);
945 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
947 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
949 timeouts = l4proto->get_timeouts(net);
951 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
952 nf_conntrack_free(ct);
953 pr_debug("init conntrack: can't track with proto module\n");
958 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
960 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
961 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
962 nf_ct_labels_ext_add(ct);
964 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
965 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
966 ecache ? ecache->expmask : 0,
970 if (net->ct.expect_count) {
971 spin_lock(&nf_conntrack_expect_lock);
972 exp = nf_ct_find_expectation(net, zone, tuple);
974 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
976 /* Welcome, Mr. Bond. We've been expecting you... */
977 __set_bit(IPS_EXPECTED_BIT, &ct->status);
978 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
979 ct->master = exp->master;
981 help = nf_ct_helper_ext_add(ct, exp->helper,
984 rcu_assign_pointer(help->helper, exp->helper);
987 #ifdef CONFIG_NF_CONNTRACK_MARK
988 ct->mark = exp->master->mark;
990 #ifdef CONFIG_NF_CONNTRACK_SECMARK
991 ct->secmark = exp->master->secmark;
993 NF_CT_STAT_INC(net, expect_new);
995 spin_unlock(&nf_conntrack_expect_lock);
998 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
999 NF_CT_STAT_INC(net, new);
1002 /* Now it is inserted into the unconfirmed list, bump refcount */
1003 nf_conntrack_get(&ct->ct_general);
1004 nf_ct_add_to_unconfirmed_list(ct);
1010 exp->expectfn(ct, exp);
1011 nf_ct_expect_put(exp);
1014 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1017 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1018 static inline struct nf_conn *
1019 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1020 struct sk_buff *skb,
1021 unsigned int dataoff,
1024 struct nf_conntrack_l3proto *l3proto,
1025 struct nf_conntrack_l4proto *l4proto,
1027 enum ip_conntrack_info *ctinfo)
1029 struct nf_conntrack_tuple tuple;
1030 struct nf_conntrack_tuple_hash *h;
1032 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
1035 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1036 dataoff, l3num, protonum, &tuple, l3proto,
1038 pr_debug("resolve_normal_ct: Can't get tuple\n");
1042 /* look for tuple match */
1043 hash = hash_conntrack_raw(&tuple, zone);
1044 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1046 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1047 skb, dataoff, hash);
1053 ct = nf_ct_tuplehash_to_ctrack(h);
1055 /* It exists; we have (non-exclusive) reference. */
1056 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1057 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1058 /* Please set reply bit if this packet OK */
1061 /* Once we've had two way comms, always ESTABLISHED. */
1062 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1063 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
1064 *ctinfo = IP_CT_ESTABLISHED;
1065 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1066 pr_debug("nf_conntrack_in: related packet for %p\n",
1068 *ctinfo = IP_CT_RELATED;
1070 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
1071 *ctinfo = IP_CT_NEW;
1075 skb->nfct = &ct->ct_general;
1076 skb->nfctinfo = *ctinfo;
1081 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1082 struct sk_buff *skb)
1084 struct nf_conn *ct, *tmpl = NULL;
1085 enum ip_conntrack_info ctinfo;
1086 struct nf_conntrack_l3proto *l3proto;
1087 struct nf_conntrack_l4proto *l4proto;
1088 unsigned int *timeouts;
1089 unsigned int dataoff;
1095 /* Previously seen (loopback or untracked)? Ignore. */
1096 tmpl = (struct nf_conn *)skb->nfct;
1097 if (!nf_ct_is_template(tmpl)) {
1098 NF_CT_STAT_INC_ATOMIC(net, ignore);
1104 /* rcu_read_lock()ed by nf_hook_slow */
1105 l3proto = __nf_ct_l3proto_find(pf);
1106 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1107 &dataoff, &protonum);
1109 pr_debug("not prepared to track yet or error occurred\n");
1110 NF_CT_STAT_INC_ATOMIC(net, error);
1111 NF_CT_STAT_INC_ATOMIC(net, invalid);
1116 l4proto = __nf_ct_l4proto_find(pf, protonum);
1118 /* It may be an special packet, error, unclean...
1119 * inverse of the return code tells to the netfilter
1120 * core what to do with the packet. */
1121 if (l4proto->error != NULL) {
1122 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1125 NF_CT_STAT_INC_ATOMIC(net, error);
1126 NF_CT_STAT_INC_ATOMIC(net, invalid);
1130 /* ICMP[v6] protocol trackers may assign one conntrack. */
1135 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1136 l3proto, l4proto, &set_reply, &ctinfo);
1138 /* Not valid part of a connection */
1139 NF_CT_STAT_INC_ATOMIC(net, invalid);
1145 /* Too stressed to deal. */
1146 NF_CT_STAT_INC_ATOMIC(net, drop);
1151 NF_CT_ASSERT(skb->nfct);
1153 /* Decide what timeout policy we want to apply to this flow. */
1154 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1156 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1158 /* Invalid: inverse of the return code tells
1159 * the netfilter core what to do */
1160 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1161 nf_conntrack_put(skb->nfct);
1163 NF_CT_STAT_INC_ATOMIC(net, invalid);
1164 if (ret == -NF_DROP)
1165 NF_CT_STAT_INC_ATOMIC(net, drop);
1170 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1171 nf_conntrack_event_cache(IPCT_REPLY, ct);
1174 /* Special case: we have to repeat this hook, assign the
1175 * template again to this packet. We assume that this packet
1176 * has no conntrack assigned. This is used by nf_ct_tcp. */
1177 if (ret == NF_REPEAT)
1178 skb->nfct = (struct nf_conntrack *)tmpl;
1185 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1187 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1188 const struct nf_conntrack_tuple *orig)
1193 ret = nf_ct_invert_tuple(inverse, orig,
1194 __nf_ct_l3proto_find(orig->src.l3num),
1195 __nf_ct_l4proto_find(orig->src.l3num,
1196 orig->dst.protonum));
1200 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1202 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1203 implicitly racy: see __nf_conntrack_confirm */
1204 void nf_conntrack_alter_reply(struct nf_conn *ct,
1205 const struct nf_conntrack_tuple *newreply)
1207 struct nf_conn_help *help = nfct_help(ct);
1209 /* Should be unconfirmed, so not in hash table yet */
1210 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1212 pr_debug("Altering reply tuple of %p to ", ct);
1213 nf_ct_dump_tuple(newreply);
1215 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1216 if (ct->master || (help && !hlist_empty(&help->expectations)))
1220 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1223 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1225 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1226 void __nf_ct_refresh_acct(struct nf_conn *ct,
1227 enum ip_conntrack_info ctinfo,
1228 const struct sk_buff *skb,
1229 unsigned long extra_jiffies,
1232 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1235 /* Only update if this is not a fixed timeout */
1236 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1239 /* If not in hash table, timer will not be active yet */
1240 if (!nf_ct_is_confirmed(ct)) {
1241 ct->timeout.expires = extra_jiffies;
1243 unsigned long newtime = jiffies + extra_jiffies;
1245 /* Only update the timeout if the new timeout is at least
1246 HZ jiffies from the old timeout. Need del_timer for race
1247 avoidance (may already be dying). */
1248 if (newtime - ct->timeout.expires >= HZ)
1249 mod_timer_pending(&ct->timeout, newtime);
1254 struct nf_conn_acct *acct;
1256 acct = nf_conn_acct_find(ct);
1258 struct nf_conn_counter *counter = acct->counter;
1260 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1261 atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1265 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1267 bool __nf_ct_kill_acct(struct nf_conn *ct,
1268 enum ip_conntrack_info ctinfo,
1269 const struct sk_buff *skb,
1273 struct nf_conn_acct *acct;
1275 acct = nf_conn_acct_find(ct);
1277 struct nf_conn_counter *counter = acct->counter;
1279 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1280 atomic64_add(skb->len - skb_network_offset(skb),
1281 &counter[CTINFO2DIR(ctinfo)].bytes);
1285 if (del_timer(&ct->timeout)) {
1286 ct->timeout.function((unsigned long)ct);
1291 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1293 #ifdef CONFIG_NF_CONNTRACK_ZONES
1294 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1295 .len = sizeof(struct nf_conntrack_zone),
1296 .align = __alignof__(struct nf_conntrack_zone),
1297 .id = NF_CT_EXT_ZONE,
1301 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1303 #include <linux/netfilter/nfnetlink.h>
1304 #include <linux/netfilter/nfnetlink_conntrack.h>
1305 #include <linux/mutex.h>
1307 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1308 * in ip_conntrack_core, since we don't want the protocols to autoload
1309 * or depend on ctnetlink */
1310 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1311 const struct nf_conntrack_tuple *tuple)
1313 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1314 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1315 goto nla_put_failure;
1321 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1323 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1324 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1325 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1327 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1329 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1330 struct nf_conntrack_tuple *t)
1332 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1335 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1336 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1340 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1342 int nf_ct_port_nlattr_tuple_size(void)
1344 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1346 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1349 /* Used by ipt_REJECT and ip6t_REJECT. */
1350 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1353 enum ip_conntrack_info ctinfo;
1355 /* This ICMP is in reverse direction to the packet which caused it */
1356 ct = nf_ct_get(skb, &ctinfo);
1357 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1358 ctinfo = IP_CT_RELATED_REPLY;
1360 ctinfo = IP_CT_RELATED;
1362 /* Attach to new skbuff, and increment count */
1363 nskb->nfct = &ct->ct_general;
1364 nskb->nfctinfo = ctinfo;
1365 nf_conntrack_get(nskb->nfct);
1368 /* Bring out ya dead! */
1369 static struct nf_conn *
1370 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1371 void *data, unsigned int *bucket)
1373 struct nf_conntrack_tuple_hash *h;
1375 struct hlist_nulls_node *n;
1379 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1380 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1383 if (*bucket < net->ct.htable_size) {
1384 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1385 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1387 ct = nf_ct_tuplehash_to_ctrack(h);
1396 for_each_possible_cpu(cpu) {
1397 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1399 spin_lock_bh(&pcpu->lock);
1400 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1401 ct = nf_ct_tuplehash_to_ctrack(h);
1403 set_bit(IPS_DYING_BIT, &ct->status);
1405 spin_unlock_bh(&pcpu->lock);
1409 atomic_inc(&ct->ct_general.use);
1415 void nf_ct_iterate_cleanup(struct net *net,
1416 int (*iter)(struct nf_conn *i, void *data),
1417 void *data, u32 portid, int report)
1420 unsigned int bucket = 0;
1422 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1423 /* Time to push up daises... */
1424 if (del_timer(&ct->timeout))
1425 nf_ct_delete(ct, portid, report);
1427 /* ... else the timer will get him soon. */
1432 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1434 static int kill_all(struct nf_conn *i, void *data)
1439 void nf_ct_free_hashtable(void *hash, unsigned int size)
1441 if (is_vmalloc_addr(hash))
1444 free_pages((unsigned long)hash,
1445 get_order(sizeof(struct hlist_head) * size));
1447 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1449 static int untrack_refs(void)
1453 for_each_possible_cpu(cpu) {
1454 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1456 cnt += atomic_read(&ct->ct_general.use) - 1;
1461 void nf_conntrack_cleanup_start(void)
1463 RCU_INIT_POINTER(ip_ct_attach, NULL);
1466 void nf_conntrack_cleanup_end(void)
1468 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1469 while (untrack_refs() > 0)
1472 #ifdef CONFIG_NF_CONNTRACK_ZONES
1473 nf_ct_extend_unregister(&nf_ct_zone_extend);
1475 nf_conntrack_proto_fini();
1476 nf_conntrack_seqadj_fini();
1477 nf_conntrack_labels_fini();
1478 nf_conntrack_helper_fini();
1479 nf_conntrack_timeout_fini();
1480 nf_conntrack_ecache_fini();
1481 nf_conntrack_tstamp_fini();
1482 nf_conntrack_acct_fini();
1483 nf_conntrack_expect_fini();
1487 * Mishearing the voices in his head, our hero wonders how he's
1488 * supposed to kill the mall.
1490 void nf_conntrack_cleanup_net(struct net *net)
1494 list_add(&net->exit_list, &single);
1495 nf_conntrack_cleanup_net_list(&single);
1498 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1504 * This makes sure all current packets have passed through
1505 * netfilter framework. Roll on, two-stage module
1511 list_for_each_entry(net, net_exit_list, exit_list) {
1512 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1513 if (atomic_read(&net->ct.count) != 0)
1518 goto i_see_dead_people;
1521 list_for_each_entry(net, net_exit_list, exit_list) {
1522 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1523 nf_conntrack_proto_pernet_fini(net);
1524 nf_conntrack_helper_pernet_fini(net);
1525 nf_conntrack_ecache_pernet_fini(net);
1526 nf_conntrack_tstamp_pernet_fini(net);
1527 nf_conntrack_acct_pernet_fini(net);
1528 nf_conntrack_expect_pernet_fini(net);
1529 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1530 kfree(net->ct.slabname);
1531 free_percpu(net->ct.stat);
1532 free_percpu(net->ct.pcpu_lists);
1536 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1538 struct hlist_nulls_head *hash;
1539 unsigned int nr_slots, i;
1542 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1543 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1544 sz = nr_slots * sizeof(struct hlist_nulls_head);
1545 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1551 for (i = 0; i < nr_slots; i++)
1552 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1556 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1558 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1561 unsigned int hashsize, old_size;
1562 struct hlist_nulls_head *hash, *old_hash;
1563 struct nf_conntrack_tuple_hash *h;
1566 if (current->nsproxy->net_ns != &init_net)
1569 /* On boot, we can set this without any fancy locking. */
1570 if (!nf_conntrack_htable_size)
1571 return param_set_uint(val, kp);
1573 rc = kstrtouint(val, 0, &hashsize);
1579 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1584 nf_conntrack_all_lock();
1585 write_seqcount_begin(&init_net.ct.generation);
1587 /* Lookups in the old hash might happen in parallel, which means we
1588 * might get false negatives during connection lookup. New connections
1589 * created because of a false negative won't make it into the hash
1590 * though since that required taking the locks.
1593 for (i = 0; i < init_net.ct.htable_size; i++) {
1594 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1595 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1596 struct nf_conntrack_tuple_hash, hnnode);
1597 ct = nf_ct_tuplehash_to_ctrack(h);
1598 hlist_nulls_del_rcu(&h->hnnode);
1599 bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1601 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1604 old_size = init_net.ct.htable_size;
1605 old_hash = init_net.ct.hash;
1607 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1608 init_net.ct.hash = hash;
1610 write_seqcount_end(&init_net.ct.generation);
1611 nf_conntrack_all_unlock();
1614 nf_ct_free_hashtable(old_hash, old_size);
1617 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1619 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1620 &nf_conntrack_htable_size, 0600);
1622 void nf_ct_untracked_status_or(unsigned long bits)
1626 for_each_possible_cpu(cpu)
1627 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1629 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1631 int nf_conntrack_init_start(void)
1636 for (i = 0; i < CONNTRACK_LOCKS; i++)
1637 spin_lock_init(&nf_conntrack_locks[i]);
1639 if (!nf_conntrack_htable_size) {
1640 /* Idea from tcp.c: use 1/16384 of memory.
1641 * On i386: 32MB machine has 512 buckets.
1642 * >= 1GB machines have 16384 buckets.
1643 * >= 4GB machines have 65536 buckets.
1645 nf_conntrack_htable_size
1646 = (((totalram_pages << PAGE_SHIFT) / 16384)
1647 / sizeof(struct hlist_head));
1648 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1649 nf_conntrack_htable_size = 65536;
1650 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1651 nf_conntrack_htable_size = 16384;
1652 if (nf_conntrack_htable_size < 32)
1653 nf_conntrack_htable_size = 32;
1655 /* Use a max. factor of four by default to get the same max as
1656 * with the old struct list_heads. When a table size is given
1657 * we use the old value of 8 to avoid reducing the max.
1661 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1663 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1664 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1667 ret = nf_conntrack_expect_init();
1671 ret = nf_conntrack_acct_init();
1675 ret = nf_conntrack_tstamp_init();
1679 ret = nf_conntrack_ecache_init();
1683 ret = nf_conntrack_timeout_init();
1687 ret = nf_conntrack_helper_init();
1691 ret = nf_conntrack_labels_init();
1695 ret = nf_conntrack_seqadj_init();
1699 #ifdef CONFIG_NF_CONNTRACK_ZONES
1700 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1704 ret = nf_conntrack_proto_init();
1708 /* Set up fake conntrack: to never be deleted, not in any hashes */
1709 for_each_possible_cpu(cpu) {
1710 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1711 write_pnet(&ct->ct_net, &init_net);
1712 atomic_set(&ct->ct_general.use, 1);
1714 /* - and look it like as a confirmed connection */
1715 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1719 #ifdef CONFIG_NF_CONNTRACK_ZONES
1720 nf_ct_extend_unregister(&nf_ct_zone_extend);
1723 nf_conntrack_seqadj_fini();
1725 nf_conntrack_labels_fini();
1727 nf_conntrack_helper_fini();
1729 nf_conntrack_timeout_fini();
1731 nf_conntrack_ecache_fini();
1733 nf_conntrack_tstamp_fini();
1735 nf_conntrack_acct_fini();
1737 nf_conntrack_expect_fini();
1742 void nf_conntrack_init_end(void)
1744 /* For use by REJECT target */
1745 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1746 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1750 * We need to use special "null" values, not used in hash table
1752 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1753 #define DYING_NULLS_VAL ((1<<30)+1)
1754 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1756 int nf_conntrack_init_net(struct net *net)
1761 atomic_set(&net->ct.count, 0);
1762 seqcount_init(&net->ct.generation);
1764 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1765 if (!net->ct.pcpu_lists)
1768 for_each_possible_cpu(cpu) {
1769 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1771 spin_lock_init(&pcpu->lock);
1772 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1773 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1776 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1778 goto err_pcpu_lists;
1780 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1781 if (!net->ct.slabname)
1784 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1785 sizeof(struct nf_conn), 0,
1786 SLAB_DESTROY_BY_RCU, NULL);
1787 if (!net->ct.nf_conntrack_cachep) {
1788 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1792 net->ct.htable_size = nf_conntrack_htable_size;
1793 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1794 if (!net->ct.hash) {
1795 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1798 ret = nf_conntrack_expect_pernet_init(net);
1801 ret = nf_conntrack_acct_pernet_init(net);
1804 ret = nf_conntrack_tstamp_pernet_init(net);
1807 ret = nf_conntrack_ecache_pernet_init(net);
1810 ret = nf_conntrack_helper_pernet_init(net);
1813 ret = nf_conntrack_proto_pernet_init(net);
1819 nf_conntrack_helper_pernet_fini(net);
1821 nf_conntrack_ecache_pernet_fini(net);
1823 nf_conntrack_tstamp_pernet_fini(net);
1825 nf_conntrack_acct_pernet_fini(net);
1827 nf_conntrack_expect_pernet_fini(net);
1829 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1831 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1833 kfree(net->ct.slabname);
1835 free_percpu(net->ct.stat);
1837 free_percpu(net->ct.pcpu_lists);