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,
172 struct nf_conntrack_tuple *tuple,
173 const struct nf_conntrack_l3proto *l3proto,
174 const struct nf_conntrack_l4proto *l4proto)
176 memset(tuple, 0, sizeof(*tuple));
178 tuple->src.l3num = l3num;
179 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
182 tuple->dst.protonum = protonum;
183 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
185 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
187 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
189 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
191 struct net *net, struct nf_conntrack_tuple *tuple)
193 struct nf_conntrack_l3proto *l3proto;
194 struct nf_conntrack_l4proto *l4proto;
195 unsigned int protoff;
201 l3proto = __nf_ct_l3proto_find(l3num);
202 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
203 if (ret != NF_ACCEPT) {
208 l4proto = __nf_ct_l4proto_find(l3num, protonum);
210 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
216 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
219 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
220 const struct nf_conntrack_tuple *orig,
221 const struct nf_conntrack_l3proto *l3proto,
222 const struct nf_conntrack_l4proto *l4proto)
224 memset(inverse, 0, sizeof(*inverse));
226 inverse->src.l3num = orig->src.l3num;
227 if (l3proto->invert_tuple(inverse, orig) == 0)
230 inverse->dst.dir = !orig->dst.dir;
232 inverse->dst.protonum = orig->dst.protonum;
233 return l4proto->invert_tuple(inverse, orig);
235 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
238 clean_from_lists(struct nf_conn *ct)
240 pr_debug("clean_from_lists(%p)\n", ct);
241 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
242 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
244 /* Destroy all pending expectations */
245 nf_ct_remove_expectations(ct);
248 /* must be called with local_bh_disable */
249 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
251 struct ct_pcpu *pcpu;
253 /* add this conntrack to the (per cpu) dying list */
254 ct->cpu = smp_processor_id();
255 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
257 spin_lock(&pcpu->lock);
258 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
260 spin_unlock(&pcpu->lock);
263 /* must be called with local_bh_disable */
264 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
266 struct ct_pcpu *pcpu;
268 /* add this conntrack to the (per cpu) unconfirmed list */
269 ct->cpu = smp_processor_id();
270 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
272 spin_lock(&pcpu->lock);
273 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
275 spin_unlock(&pcpu->lock);
278 /* must be called with local_bh_disable */
279 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
281 struct ct_pcpu *pcpu;
283 /* We overload first tuple to link into unconfirmed or dying list.*/
284 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
286 spin_lock(&pcpu->lock);
287 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
288 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
289 spin_unlock(&pcpu->lock);
292 /* Released via destroy_conntrack() */
293 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
294 const struct nf_conntrack_zone *zone,
297 struct nf_conn *tmpl;
299 tmpl = kzalloc(sizeof(*tmpl), flags);
303 tmpl->status = IPS_TEMPLATE;
304 write_pnet(&tmpl->ct_net, net);
306 if (nf_ct_zone_add(tmpl, flags, zone) < 0)
309 atomic_set(&tmpl->ct_general.use, 0);
316 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
318 void nf_ct_tmpl_free(struct nf_conn *tmpl)
320 nf_ct_ext_destroy(tmpl);
321 nf_ct_ext_free(tmpl);
324 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
327 destroy_conntrack(struct nf_conntrack *nfct)
329 struct nf_conn *ct = (struct nf_conn *)nfct;
330 struct net *net = nf_ct_net(ct);
331 struct nf_conntrack_l4proto *l4proto;
333 pr_debug("destroy_conntrack(%p)\n", ct);
334 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
335 NF_CT_ASSERT(!timer_pending(&ct->timeout));
337 if (unlikely(nf_ct_is_template(ct))) {
342 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
343 if (l4proto && l4proto->destroy)
344 l4proto->destroy(ct);
349 /* Expectations will have been removed in clean_from_lists,
350 * except TFTP can create an expectation on the first packet,
351 * before connection is in the list, so we need to clean here,
354 nf_ct_remove_expectations(ct);
356 nf_ct_del_from_dying_or_unconfirmed_list(ct);
358 NF_CT_STAT_INC(net, delete);
362 nf_ct_put(ct->master);
364 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
365 nf_conntrack_free(ct);
368 static void nf_ct_delete_from_lists(struct nf_conn *ct)
370 struct net *net = nf_ct_net(ct);
371 unsigned int hash, reply_hash;
372 unsigned int sequence;
374 nf_ct_helper_destroy(ct);
378 sequence = read_seqcount_begin(&net->ct.generation);
379 hash = hash_conntrack(net,
380 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
381 reply_hash = hash_conntrack(net,
382 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
383 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
385 clean_from_lists(ct);
386 nf_conntrack_double_unlock(hash, reply_hash);
388 nf_ct_add_to_dying_list(ct);
390 NF_CT_STAT_INC(net, delete_list);
394 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
396 struct nf_conn_tstamp *tstamp;
398 tstamp = nf_conn_tstamp_find(ct);
399 if (tstamp && tstamp->stop == 0)
400 tstamp->stop = ktime_get_real_ns();
402 if (nf_ct_is_dying(ct))
405 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
406 portid, report) < 0) {
407 /* destroy event was not delivered */
408 nf_ct_delete_from_lists(ct);
409 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
413 nf_conntrack_ecache_work(nf_ct_net(ct));
414 set_bit(IPS_DYING_BIT, &ct->status);
416 nf_ct_delete_from_lists(ct);
420 EXPORT_SYMBOL_GPL(nf_ct_delete);
422 static void death_by_timeout(unsigned long ul_conntrack)
424 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
428 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
429 const struct nf_conntrack_tuple *tuple,
430 const struct nf_conntrack_zone *zone)
432 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
434 /* A conntrack can be recreated with the equal tuple,
435 * so we need to check that the conntrack is confirmed
437 return nf_ct_tuple_equal(tuple, &h->tuple) &&
438 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
439 nf_ct_is_confirmed(ct);
444 * - Caller must take a reference on returned object
445 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
447 static struct nf_conntrack_tuple_hash *
448 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
449 const struct nf_conntrack_tuple *tuple, u32 hash)
451 struct nf_conntrack_tuple_hash *h;
452 struct hlist_nulls_node *n;
453 unsigned int bucket = hash_bucket(hash, net);
455 /* Disable BHs the entire time since we normally need to disable them
456 * at least once for the stats anyway.
460 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
461 if (nf_ct_key_equal(h, tuple, zone)) {
462 NF_CT_STAT_INC(net, found);
466 NF_CT_STAT_INC(net, searched);
469 * if the nulls value we got at the end of this lookup is
470 * not the expected one, we must restart lookup.
471 * We probably met an item that was moved to another chain.
473 if (get_nulls_value(n) != bucket) {
474 NF_CT_STAT_INC(net, search_restart);
482 /* Find a connection corresponding to a tuple. */
483 static struct nf_conntrack_tuple_hash *
484 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
485 const struct nf_conntrack_tuple *tuple, u32 hash)
487 struct nf_conntrack_tuple_hash *h;
492 h = ____nf_conntrack_find(net, zone, tuple, hash);
494 ct = nf_ct_tuplehash_to_ctrack(h);
495 if (unlikely(nf_ct_is_dying(ct) ||
496 !atomic_inc_not_zero(&ct->ct_general.use)))
499 if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
510 struct nf_conntrack_tuple_hash *
511 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
512 const struct nf_conntrack_tuple *tuple)
514 return __nf_conntrack_find_get(net, zone, tuple,
515 hash_conntrack_raw(tuple));
517 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
519 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
521 unsigned int reply_hash)
523 struct net *net = nf_ct_net(ct);
525 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
526 &net->ct.hash[hash]);
527 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
528 &net->ct.hash[reply_hash]);
532 nf_conntrack_hash_check_insert(struct nf_conn *ct)
534 const struct nf_conntrack_zone *zone;
535 struct net *net = nf_ct_net(ct);
536 unsigned int hash, reply_hash;
537 struct nf_conntrack_tuple_hash *h;
538 struct hlist_nulls_node *n;
539 unsigned int sequence;
541 zone = nf_ct_zone(ct);
545 sequence = read_seqcount_begin(&net->ct.generation);
546 hash = hash_conntrack(net,
547 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
548 reply_hash = hash_conntrack(net,
549 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
550 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
552 /* See if there's one in the list already, including reverse */
553 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
554 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
556 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
559 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
560 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
562 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
566 add_timer(&ct->timeout);
568 /* The caller holds a reference to this object */
569 atomic_set(&ct->ct_general.use, 2);
570 __nf_conntrack_hash_insert(ct, hash, reply_hash);
571 nf_conntrack_double_unlock(hash, reply_hash);
572 NF_CT_STAT_INC(net, insert);
577 nf_conntrack_double_unlock(hash, reply_hash);
578 NF_CT_STAT_INC(net, insert_failed);
582 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
584 /* Confirm a connection given skb; places it in hash table */
586 __nf_conntrack_confirm(struct sk_buff *skb)
588 const struct nf_conntrack_zone *zone;
589 unsigned int hash, reply_hash;
590 struct nf_conntrack_tuple_hash *h;
592 struct nf_conn_help *help;
593 struct nf_conn_tstamp *tstamp;
594 struct hlist_nulls_node *n;
595 enum ip_conntrack_info ctinfo;
597 unsigned int sequence;
599 ct = nf_ct_get(skb, &ctinfo);
602 /* ipt_REJECT uses nf_conntrack_attach to attach related
603 ICMP/TCP RST packets in other direction. Actual packet
604 which created connection will be IP_CT_NEW or for an
605 expected connection, IP_CT_RELATED. */
606 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
609 zone = nf_ct_zone(ct);
613 sequence = read_seqcount_begin(&net->ct.generation);
614 /* reuse the hash saved before */
615 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
616 hash = hash_bucket(hash, net);
617 reply_hash = hash_conntrack(net,
618 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
620 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
622 /* We're not in hash table, and we refuse to set up related
623 * connections for unconfirmed conns. But packet copies and
624 * REJECT will give spurious warnings here.
626 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
628 /* No external references means no one else could have
631 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
632 pr_debug("Confirming conntrack %p\n", ct);
633 /* We have to check the DYING flag after unlink to prevent
634 * a race against nf_ct_get_next_corpse() possibly called from
635 * user context, else we insert an already 'dead' hash, blocking
636 * further use of that particular connection -JM.
638 nf_ct_del_from_dying_or_unconfirmed_list(ct);
640 if (unlikely(nf_ct_is_dying(ct)))
643 /* See if there's one in the list already, including reverse:
644 NAT could have grabbed it without realizing, since we're
645 not in the hash. If there is, we lost race. */
646 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
647 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
649 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
652 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
653 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
655 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
659 /* Timer relative to confirmation time, not original
660 setting time, otherwise we'd get timer wrap in
661 weird delay cases. */
662 ct->timeout.expires += jiffies;
663 add_timer(&ct->timeout);
664 atomic_inc(&ct->ct_general.use);
665 ct->status |= IPS_CONFIRMED;
667 /* set conntrack timestamp, if enabled. */
668 tstamp = nf_conn_tstamp_find(ct);
670 if (skb->tstamp.tv64 == 0)
671 __net_timestamp(skb);
673 tstamp->start = ktime_to_ns(skb->tstamp);
675 /* Since the lookup is lockless, hash insertion must be done after
676 * starting the timer and setting the CONFIRMED bit. The RCU barriers
677 * guarantee that no other CPU can find the conntrack before the above
678 * stores are visible.
680 __nf_conntrack_hash_insert(ct, hash, reply_hash);
681 nf_conntrack_double_unlock(hash, reply_hash);
682 NF_CT_STAT_INC(net, insert);
685 help = nfct_help(ct);
686 if (help && help->helper)
687 nf_conntrack_event_cache(IPCT_HELPER, ct);
689 nf_conntrack_event_cache(master_ct(ct) ?
690 IPCT_RELATED : IPCT_NEW, ct);
694 nf_ct_add_to_dying_list(ct);
695 nf_conntrack_double_unlock(hash, reply_hash);
696 NF_CT_STAT_INC(net, insert_failed);
700 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
702 /* Returns true if a connection correspondings to the tuple (required
705 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
706 const struct nf_conn *ignored_conntrack)
708 struct net *net = nf_ct_net(ignored_conntrack);
709 const struct nf_conntrack_zone *zone;
710 struct nf_conntrack_tuple_hash *h;
711 struct hlist_nulls_node *n;
715 zone = nf_ct_zone(ignored_conntrack);
716 hash = hash_conntrack(net, tuple);
718 /* Disable BHs the entire time since we need to disable them at
719 * least once for the stats anyway.
722 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
723 ct = nf_ct_tuplehash_to_ctrack(h);
724 if (ct != ignored_conntrack &&
725 nf_ct_tuple_equal(tuple, &h->tuple) &&
726 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h))) {
727 NF_CT_STAT_INC(net, found);
728 rcu_read_unlock_bh();
731 NF_CT_STAT_INC(net, searched);
733 rcu_read_unlock_bh();
737 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
739 #define NF_CT_EVICTION_RANGE 8
741 /* There's a small race here where we may free a just-assured
742 connection. Too bad: we're in trouble anyway. */
743 static noinline int early_drop(struct net *net, unsigned int _hash)
745 /* Use oldest entry, which is roughly LRU */
746 struct nf_conntrack_tuple_hash *h;
747 struct nf_conn *ct = NULL, *tmp;
748 struct hlist_nulls_node *n;
749 unsigned int i = 0, cnt = 0;
751 unsigned int hash, sequence;
756 sequence = read_seqcount_begin(&net->ct.generation);
757 hash = hash_bucket(_hash, net);
758 for (; i < net->ct.htable_size; i++) {
759 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
761 if (read_seqcount_retry(&net->ct.generation, sequence)) {
765 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
767 tmp = nf_ct_tuplehash_to_ctrack(h);
768 if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
769 !nf_ct_is_dying(tmp) &&
770 atomic_inc_not_zero(&tmp->ct_general.use)) {
777 hash = (hash + 1) % net->ct.htable_size;
780 if (ct || cnt >= NF_CT_EVICTION_RANGE)
789 if (del_timer(&ct->timeout)) {
790 if (nf_ct_delete(ct, 0, 0)) {
792 NF_CT_STAT_INC_ATOMIC(net, early_drop);
799 void init_nf_conntrack_hash_rnd(void)
804 * Why not initialize nf_conntrack_rnd in a "init()" function ?
805 * Because there isn't enough entropy when system initializing,
806 * and we initialize it as late as possible.
809 get_random_bytes(&rand, sizeof(rand));
811 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
814 static struct nf_conn *
815 __nf_conntrack_alloc(struct net *net,
816 const struct nf_conntrack_zone *zone,
817 const struct nf_conntrack_tuple *orig,
818 const struct nf_conntrack_tuple *repl,
823 if (unlikely(!nf_conntrack_hash_rnd)) {
824 init_nf_conntrack_hash_rnd();
825 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
826 hash = hash_conntrack_raw(orig);
829 /* We don't want any race condition at early drop stage */
830 atomic_inc(&net->ct.count);
832 if (nf_conntrack_max &&
833 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
834 if (!early_drop(net, hash)) {
835 atomic_dec(&net->ct.count);
836 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
837 return ERR_PTR(-ENOMEM);
842 * Do not use kmem_cache_zalloc(), as this cache uses
843 * SLAB_DESTROY_BY_RCU.
845 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
849 spin_lock_init(&ct->lock);
850 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
851 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
852 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
853 /* save hash for reusing when confirming */
854 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
856 /* Don't set timer yet: wait for confirmation */
857 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
858 write_pnet(&ct->ct_net, net);
859 memset(&ct->__nfct_init_offset[0], 0,
860 offsetof(struct nf_conn, proto) -
861 offsetof(struct nf_conn, __nfct_init_offset[0]));
863 if (zone && nf_ct_zone_add(ct, GFP_ATOMIC, zone) < 0)
866 /* Because we use RCU lookups, we set ct_general.use to zero before
867 * this is inserted in any list.
869 atomic_set(&ct->ct_general.use, 0);
872 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
874 atomic_dec(&net->ct.count);
875 return ERR_PTR(-ENOMEM);
878 struct nf_conn *nf_conntrack_alloc(struct net *net,
879 const struct nf_conntrack_zone *zone,
880 const struct nf_conntrack_tuple *orig,
881 const struct nf_conntrack_tuple *repl,
884 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
886 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
888 void nf_conntrack_free(struct nf_conn *ct)
890 struct net *net = nf_ct_net(ct);
892 /* A freed object has refcnt == 0, that's
893 * the golden rule for SLAB_DESTROY_BY_RCU
895 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
897 nf_ct_ext_destroy(ct);
899 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
900 smp_mb__before_atomic();
901 atomic_dec(&net->ct.count);
903 EXPORT_SYMBOL_GPL(nf_conntrack_free);
906 /* Allocate a new conntrack: we return -ENOMEM if classification
907 failed due to stress. Otherwise it really is unclassifiable. */
908 static struct nf_conntrack_tuple_hash *
909 init_conntrack(struct net *net, struct nf_conn *tmpl,
910 const struct nf_conntrack_tuple *tuple,
911 struct nf_conntrack_l3proto *l3proto,
912 struct nf_conntrack_l4proto *l4proto,
914 unsigned int dataoff, u32 hash)
917 struct nf_conn_help *help;
918 struct nf_conntrack_tuple repl_tuple;
919 struct nf_conntrack_ecache *ecache;
920 struct nf_conntrack_expect *exp = NULL;
921 const struct nf_conntrack_zone *zone;
922 struct nf_conn_timeout *timeout_ext;
923 struct nf_conntrack_zone tmp;
924 unsigned int *timeouts;
926 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
927 pr_debug("Can't invert tuple.\n");
931 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
932 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
935 return (struct nf_conntrack_tuple_hash *)ct;
937 if (tmpl && nfct_synproxy(tmpl)) {
938 nfct_seqadj_ext_add(ct);
939 nfct_synproxy_ext_add(ct);
942 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
944 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
946 timeouts = l4proto->get_timeouts(net);
948 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
949 nf_conntrack_free(ct);
950 pr_debug("init conntrack: can't track with proto module\n");
955 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
957 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
958 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
959 nf_ct_labels_ext_add(ct);
961 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
962 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
963 ecache ? ecache->expmask : 0,
967 if (net->ct.expect_count) {
968 spin_lock(&nf_conntrack_expect_lock);
969 exp = nf_ct_find_expectation(net, zone, tuple);
971 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
973 /* Welcome, Mr. Bond. We've been expecting you... */
974 __set_bit(IPS_EXPECTED_BIT, &ct->status);
975 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
976 ct->master = exp->master;
978 help = nf_ct_helper_ext_add(ct, exp->helper,
981 rcu_assign_pointer(help->helper, exp->helper);
984 #ifdef CONFIG_NF_CONNTRACK_MARK
985 ct->mark = exp->master->mark;
987 #ifdef CONFIG_NF_CONNTRACK_SECMARK
988 ct->secmark = exp->master->secmark;
990 NF_CT_STAT_INC(net, expect_new);
992 spin_unlock(&nf_conntrack_expect_lock);
995 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
996 NF_CT_STAT_INC(net, new);
999 /* Now it is inserted into the unconfirmed list, bump refcount */
1000 nf_conntrack_get(&ct->ct_general);
1001 nf_ct_add_to_unconfirmed_list(ct);
1007 exp->expectfn(ct, exp);
1008 nf_ct_expect_put(exp);
1011 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1014 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1015 static inline struct nf_conn *
1016 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1017 struct sk_buff *skb,
1018 unsigned int dataoff,
1021 struct nf_conntrack_l3proto *l3proto,
1022 struct nf_conntrack_l4proto *l4proto,
1024 enum ip_conntrack_info *ctinfo)
1026 const struct nf_conntrack_zone *zone;
1027 struct nf_conntrack_tuple tuple;
1028 struct nf_conntrack_tuple_hash *h;
1029 struct nf_conntrack_zone tmp;
1033 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1034 dataoff, l3num, protonum, net, &tuple, l3proto,
1036 pr_debug("resolve_normal_ct: Can't get tuple\n");
1040 /* look for tuple match */
1041 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1042 hash = hash_conntrack_raw(&tuple);
1043 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1045 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1046 skb, dataoff, hash);
1052 ct = nf_ct_tuplehash_to_ctrack(h);
1054 /* It exists; we have (non-exclusive) reference. */
1055 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1056 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1057 /* Please set reply bit if this packet OK */
1060 /* Once we've had two way comms, always ESTABLISHED. */
1061 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1062 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
1063 *ctinfo = IP_CT_ESTABLISHED;
1064 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1065 pr_debug("nf_conntrack_in: related packet for %p\n",
1067 *ctinfo = IP_CT_RELATED;
1069 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
1070 *ctinfo = IP_CT_NEW;
1074 skb->nfct = &ct->ct_general;
1075 skb->nfctinfo = *ctinfo;
1080 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1081 struct sk_buff *skb)
1083 struct nf_conn *ct, *tmpl = NULL;
1084 enum ip_conntrack_info ctinfo;
1085 struct nf_conntrack_l3proto *l3proto;
1086 struct nf_conntrack_l4proto *l4proto;
1087 unsigned int *timeouts;
1088 unsigned int dataoff;
1094 /* Previously seen (loopback or untracked)? Ignore. */
1095 tmpl = (struct nf_conn *)skb->nfct;
1096 if (!nf_ct_is_template(tmpl)) {
1097 NF_CT_STAT_INC_ATOMIC(net, ignore);
1103 /* rcu_read_lock()ed by nf_hook_slow */
1104 l3proto = __nf_ct_l3proto_find(pf);
1105 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1106 &dataoff, &protonum);
1108 pr_debug("not prepared to track yet or error occurred\n");
1109 NF_CT_STAT_INC_ATOMIC(net, error);
1110 NF_CT_STAT_INC_ATOMIC(net, invalid);
1115 l4proto = __nf_ct_l4proto_find(pf, protonum);
1117 /* It may be an special packet, error, unclean...
1118 * inverse of the return code tells to the netfilter
1119 * core what to do with the packet. */
1120 if (l4proto->error != NULL) {
1121 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1124 NF_CT_STAT_INC_ATOMIC(net, error);
1125 NF_CT_STAT_INC_ATOMIC(net, invalid);
1129 /* ICMP[v6] protocol trackers may assign one conntrack. */
1134 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1135 l3proto, l4proto, &set_reply, &ctinfo);
1137 /* Not valid part of a connection */
1138 NF_CT_STAT_INC_ATOMIC(net, invalid);
1144 /* Too stressed to deal. */
1145 NF_CT_STAT_INC_ATOMIC(net, drop);
1150 NF_CT_ASSERT(skb->nfct);
1152 /* Decide what timeout policy we want to apply to this flow. */
1153 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1155 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1157 /* Invalid: inverse of the return code tells
1158 * the netfilter core what to do */
1159 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1160 nf_conntrack_put(skb->nfct);
1162 NF_CT_STAT_INC_ATOMIC(net, invalid);
1163 if (ret == -NF_DROP)
1164 NF_CT_STAT_INC_ATOMIC(net, drop);
1169 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1170 nf_conntrack_event_cache(IPCT_REPLY, ct);
1173 /* Special case: we have to repeat this hook, assign the
1174 * template again to this packet. We assume that this packet
1175 * has no conntrack assigned. This is used by nf_ct_tcp. */
1176 if (ret == NF_REPEAT)
1177 skb->nfct = (struct nf_conntrack *)tmpl;
1184 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1186 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1187 const struct nf_conntrack_tuple *orig)
1192 ret = nf_ct_invert_tuple(inverse, orig,
1193 __nf_ct_l3proto_find(orig->src.l3num),
1194 __nf_ct_l4proto_find(orig->src.l3num,
1195 orig->dst.protonum));
1199 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1201 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1202 implicitly racy: see __nf_conntrack_confirm */
1203 void nf_conntrack_alter_reply(struct nf_conn *ct,
1204 const struct nf_conntrack_tuple *newreply)
1206 struct nf_conn_help *help = nfct_help(ct);
1208 /* Should be unconfirmed, so not in hash table yet */
1209 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1211 pr_debug("Altering reply tuple of %p to ", ct);
1212 nf_ct_dump_tuple(newreply);
1214 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1215 if (ct->master || (help && !hlist_empty(&help->expectations)))
1219 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1222 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1224 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1225 void __nf_ct_refresh_acct(struct nf_conn *ct,
1226 enum ip_conntrack_info ctinfo,
1227 const struct sk_buff *skb,
1228 unsigned long extra_jiffies,
1231 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1234 /* Only update if this is not a fixed timeout */
1235 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1238 /* If not in hash table, timer will not be active yet */
1239 if (!nf_ct_is_confirmed(ct)) {
1240 ct->timeout.expires = extra_jiffies;
1242 unsigned long newtime = jiffies + extra_jiffies;
1244 /* Only update the timeout if the new timeout is at least
1245 HZ jiffies from the old timeout. Need del_timer for race
1246 avoidance (may already be dying). */
1247 if (newtime - ct->timeout.expires >= HZ)
1248 mod_timer_pending(&ct->timeout, newtime);
1253 struct nf_conn_acct *acct;
1255 acct = nf_conn_acct_find(ct);
1257 struct nf_conn_counter *counter = acct->counter;
1259 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1260 atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1264 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1266 bool __nf_ct_kill_acct(struct nf_conn *ct,
1267 enum ip_conntrack_info ctinfo,
1268 const struct sk_buff *skb,
1272 struct nf_conn_acct *acct;
1274 acct = nf_conn_acct_find(ct);
1276 struct nf_conn_counter *counter = acct->counter;
1278 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1279 atomic64_add(skb->len - skb_network_offset(skb),
1280 &counter[CTINFO2DIR(ctinfo)].bytes);
1284 if (del_timer(&ct->timeout)) {
1285 ct->timeout.function((unsigned long)ct);
1290 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1292 #ifdef CONFIG_NF_CONNTRACK_ZONES
1293 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1294 .len = sizeof(struct nf_conntrack_zone),
1295 .align = __alignof__(struct nf_conntrack_zone),
1296 .id = NF_CT_EXT_ZONE,
1300 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1302 #include <linux/netfilter/nfnetlink.h>
1303 #include <linux/netfilter/nfnetlink_conntrack.h>
1304 #include <linux/mutex.h>
1306 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1307 * in ip_conntrack_core, since we don't want the protocols to autoload
1308 * or depend on ctnetlink */
1309 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1310 const struct nf_conntrack_tuple *tuple)
1312 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1313 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1314 goto nla_put_failure;
1320 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1322 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1323 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1324 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1326 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1328 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1329 struct nf_conntrack_tuple *t)
1331 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1334 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1335 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1339 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1341 int nf_ct_port_nlattr_tuple_size(void)
1343 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1345 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1348 /* Used by ipt_REJECT and ip6t_REJECT. */
1349 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1352 enum ip_conntrack_info ctinfo;
1354 /* This ICMP is in reverse direction to the packet which caused it */
1355 ct = nf_ct_get(skb, &ctinfo);
1356 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1357 ctinfo = IP_CT_RELATED_REPLY;
1359 ctinfo = IP_CT_RELATED;
1361 /* Attach to new skbuff, and increment count */
1362 nskb->nfct = &ct->ct_general;
1363 nskb->nfctinfo = ctinfo;
1364 nf_conntrack_get(nskb->nfct);
1367 /* Bring out ya dead! */
1368 static struct nf_conn *
1369 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1370 void *data, unsigned int *bucket)
1372 struct nf_conntrack_tuple_hash *h;
1374 struct hlist_nulls_node *n;
1378 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1379 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1382 if (*bucket < net->ct.htable_size) {
1383 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1384 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1386 ct = nf_ct_tuplehash_to_ctrack(h);
1395 for_each_possible_cpu(cpu) {
1396 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1398 spin_lock_bh(&pcpu->lock);
1399 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1400 ct = nf_ct_tuplehash_to_ctrack(h);
1402 set_bit(IPS_DYING_BIT, &ct->status);
1404 spin_unlock_bh(&pcpu->lock);
1408 atomic_inc(&ct->ct_general.use);
1414 void nf_ct_iterate_cleanup(struct net *net,
1415 int (*iter)(struct nf_conn *i, void *data),
1416 void *data, u32 portid, int report)
1419 unsigned int bucket = 0;
1421 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1422 /* Time to push up daises... */
1423 if (del_timer(&ct->timeout))
1424 nf_ct_delete(ct, portid, report);
1426 /* ... else the timer will get him soon. */
1431 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1433 static int kill_all(struct nf_conn *i, void *data)
1438 void nf_ct_free_hashtable(void *hash, unsigned int size)
1440 if (is_vmalloc_addr(hash))
1443 free_pages((unsigned long)hash,
1444 get_order(sizeof(struct hlist_head) * size));
1446 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1448 static int untrack_refs(void)
1452 for_each_possible_cpu(cpu) {
1453 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1455 cnt += atomic_read(&ct->ct_general.use) - 1;
1460 void nf_conntrack_cleanup_start(void)
1462 RCU_INIT_POINTER(ip_ct_attach, NULL);
1465 void nf_conntrack_cleanup_end(void)
1467 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1468 while (untrack_refs() > 0)
1471 #ifdef CONFIG_NF_CONNTRACK_ZONES
1472 nf_ct_extend_unregister(&nf_ct_zone_extend);
1474 nf_conntrack_proto_fini();
1475 nf_conntrack_seqadj_fini();
1476 nf_conntrack_labels_fini();
1477 nf_conntrack_helper_fini();
1478 nf_conntrack_timeout_fini();
1479 nf_conntrack_ecache_fini();
1480 nf_conntrack_tstamp_fini();
1481 nf_conntrack_acct_fini();
1482 nf_conntrack_expect_fini();
1486 * Mishearing the voices in his head, our hero wonders how he's
1487 * supposed to kill the mall.
1489 void nf_conntrack_cleanup_net(struct net *net)
1493 list_add(&net->exit_list, &single);
1494 nf_conntrack_cleanup_net_list(&single);
1497 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1503 * This makes sure all current packets have passed through
1504 * netfilter framework. Roll on, two-stage module
1510 list_for_each_entry(net, net_exit_list, exit_list) {
1511 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1512 if (atomic_read(&net->ct.count) != 0)
1517 goto i_see_dead_people;
1520 list_for_each_entry(net, net_exit_list, exit_list) {
1521 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1522 nf_conntrack_proto_pernet_fini(net);
1523 nf_conntrack_helper_pernet_fini(net);
1524 nf_conntrack_ecache_pernet_fini(net);
1525 nf_conntrack_tstamp_pernet_fini(net);
1526 nf_conntrack_acct_pernet_fini(net);
1527 nf_conntrack_expect_pernet_fini(net);
1528 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1529 kfree(net->ct.slabname);
1530 free_percpu(net->ct.stat);
1531 free_percpu(net->ct.pcpu_lists);
1535 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1537 struct hlist_nulls_head *hash;
1538 unsigned int nr_slots, i;
1541 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1542 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1543 sz = nr_slots * sizeof(struct hlist_nulls_head);
1544 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1550 for (i = 0; i < nr_slots; i++)
1551 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1555 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1557 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1560 unsigned int hashsize, old_size;
1561 struct hlist_nulls_head *hash, *old_hash;
1562 struct nf_conntrack_tuple_hash *h;
1565 if (current->nsproxy->net_ns != &init_net)
1568 /* On boot, we can set this without any fancy locking. */
1569 if (!nf_conntrack_htable_size)
1570 return param_set_uint(val, kp);
1572 rc = kstrtouint(val, 0, &hashsize);
1578 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1583 nf_conntrack_all_lock();
1584 write_seqcount_begin(&init_net.ct.generation);
1586 /* Lookups in the old hash might happen in parallel, which means we
1587 * might get false negatives during connection lookup. New connections
1588 * created because of a false negative won't make it into the hash
1589 * though since that required taking the locks.
1592 for (i = 0; i < init_net.ct.htable_size; i++) {
1593 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1594 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1595 struct nf_conntrack_tuple_hash, hnnode);
1596 ct = nf_ct_tuplehash_to_ctrack(h);
1597 hlist_nulls_del_rcu(&h->hnnode);
1598 bucket = __hash_conntrack(&h->tuple, hashsize);
1599 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1602 old_size = init_net.ct.htable_size;
1603 old_hash = init_net.ct.hash;
1605 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1606 init_net.ct.hash = hash;
1608 write_seqcount_end(&init_net.ct.generation);
1609 nf_conntrack_all_unlock();
1612 nf_ct_free_hashtable(old_hash, old_size);
1615 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1617 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1618 &nf_conntrack_htable_size, 0600);
1620 void nf_ct_untracked_status_or(unsigned long bits)
1624 for_each_possible_cpu(cpu)
1625 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1627 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1629 int nf_conntrack_init_start(void)
1634 for (i = 0; i < CONNTRACK_LOCKS; i++)
1635 spin_lock_init(&nf_conntrack_locks[i]);
1637 if (!nf_conntrack_htable_size) {
1638 /* Idea from tcp.c: use 1/16384 of memory.
1639 * On i386: 32MB machine has 512 buckets.
1640 * >= 1GB machines have 16384 buckets.
1641 * >= 4GB machines have 65536 buckets.
1643 nf_conntrack_htable_size
1644 = (((totalram_pages << PAGE_SHIFT) / 16384)
1645 / sizeof(struct hlist_head));
1646 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1647 nf_conntrack_htable_size = 65536;
1648 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1649 nf_conntrack_htable_size = 16384;
1650 if (nf_conntrack_htable_size < 32)
1651 nf_conntrack_htable_size = 32;
1653 /* Use a max. factor of four by default to get the same max as
1654 * with the old struct list_heads. When a table size is given
1655 * we use the old value of 8 to avoid reducing the max.
1659 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1661 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1662 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1665 ret = nf_conntrack_expect_init();
1669 ret = nf_conntrack_acct_init();
1673 ret = nf_conntrack_tstamp_init();
1677 ret = nf_conntrack_ecache_init();
1681 ret = nf_conntrack_timeout_init();
1685 ret = nf_conntrack_helper_init();
1689 ret = nf_conntrack_labels_init();
1693 ret = nf_conntrack_seqadj_init();
1697 #ifdef CONFIG_NF_CONNTRACK_ZONES
1698 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1702 ret = nf_conntrack_proto_init();
1706 /* Set up fake conntrack: to never be deleted, not in any hashes */
1707 for_each_possible_cpu(cpu) {
1708 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1709 write_pnet(&ct->ct_net, &init_net);
1710 atomic_set(&ct->ct_general.use, 1);
1712 /* - and look it like as a confirmed connection */
1713 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1717 #ifdef CONFIG_NF_CONNTRACK_ZONES
1718 nf_ct_extend_unregister(&nf_ct_zone_extend);
1721 nf_conntrack_seqadj_fini();
1723 nf_conntrack_labels_fini();
1725 nf_conntrack_helper_fini();
1727 nf_conntrack_timeout_fini();
1729 nf_conntrack_ecache_fini();
1731 nf_conntrack_tstamp_fini();
1733 nf_conntrack_acct_fini();
1735 nf_conntrack_expect_fini();
1740 void nf_conntrack_init_end(void)
1742 /* For use by REJECT target */
1743 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1744 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1748 * We need to use special "null" values, not used in hash table
1750 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1751 #define DYING_NULLS_VAL ((1<<30)+1)
1752 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1754 int nf_conntrack_init_net(struct net *net)
1759 atomic_set(&net->ct.count, 0);
1760 seqcount_init(&net->ct.generation);
1762 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1763 if (!net->ct.pcpu_lists)
1766 for_each_possible_cpu(cpu) {
1767 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1769 spin_lock_init(&pcpu->lock);
1770 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1771 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1774 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1776 goto err_pcpu_lists;
1778 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1779 if (!net->ct.slabname)
1782 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1783 sizeof(struct nf_conn), 0,
1784 SLAB_DESTROY_BY_RCU, NULL);
1785 if (!net->ct.nf_conntrack_cachep) {
1786 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1790 net->ct.htable_size = nf_conntrack_htable_size;
1791 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1792 if (!net->ct.hash) {
1793 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1796 ret = nf_conntrack_expect_pernet_init(net);
1799 ret = nf_conntrack_acct_pernet_init(net);
1802 ret = nf_conntrack_tstamp_pernet_init(net);
1805 ret = nf_conntrack_ecache_pernet_init(net);
1808 ret = nf_conntrack_helper_pernet_init(net);
1811 ret = nf_conntrack_proto_pernet_init(net);
1817 nf_conntrack_helper_pernet_fini(net);
1819 nf_conntrack_ecache_pernet_fini(net);
1821 nf_conntrack_tstamp_pernet_fini(net);
1823 nf_conntrack_acct_pernet_fini(net);
1825 nf_conntrack_expect_pernet_fini(net);
1827 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1829 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1831 kfree(net->ct.slabname);
1833 free_percpu(net->ct.stat);
1835 free_percpu(net->ct.pcpu_lists);