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 __read_mostly spinlock_t nf_conntrack_locks_all_lock;
70 static __read_mostly bool nf_conntrack_locks_all;
72 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
75 while (unlikely(nf_conntrack_locks_all)) {
77 spin_unlock_wait(&nf_conntrack_locks_all_lock);
81 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
83 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
85 h1 %= CONNTRACK_LOCKS;
86 h2 %= CONNTRACK_LOCKS;
87 spin_unlock(&nf_conntrack_locks[h1]);
89 spin_unlock(&nf_conntrack_locks[h2]);
92 /* return true if we need to recompute hashes (in case hash table was resized) */
93 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
94 unsigned int h2, unsigned int sequence)
96 h1 %= CONNTRACK_LOCKS;
97 h2 %= CONNTRACK_LOCKS;
99 nf_conntrack_lock(&nf_conntrack_locks[h1]);
101 spin_lock_nested(&nf_conntrack_locks[h2],
102 SINGLE_DEPTH_NESTING);
104 nf_conntrack_lock(&nf_conntrack_locks[h2]);
105 spin_lock_nested(&nf_conntrack_locks[h1],
106 SINGLE_DEPTH_NESTING);
108 if (read_seqcount_retry(&net->ct.generation, sequence)) {
109 nf_conntrack_double_unlock(h1, h2);
115 static void nf_conntrack_all_lock(void)
119 spin_lock(&nf_conntrack_locks_all_lock);
120 nf_conntrack_locks_all = true;
122 for (i = 0; i < CONNTRACK_LOCKS; i++) {
123 spin_unlock_wait(&nf_conntrack_locks[i]);
127 static void nf_conntrack_all_unlock(void)
129 nf_conntrack_locks_all = false;
130 spin_unlock(&nf_conntrack_locks_all_lock);
133 unsigned int nf_conntrack_htable_size __read_mostly;
134 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
136 unsigned int nf_conntrack_max __read_mostly;
137 EXPORT_SYMBOL_GPL(nf_conntrack_max);
139 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
140 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
142 unsigned int nf_conntrack_hash_rnd __read_mostly;
143 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
145 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple)
149 /* The direction must be ignored, so we hash everything up to the
150 * destination ports (which is a multiple of 4) and treat the last
151 * three bytes manually.
153 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
154 return jhash2((u32 *)tuple, n, nf_conntrack_hash_rnd ^
155 (((__force __u16)tuple->dst.u.all << 16) |
156 tuple->dst.protonum));
159 static u32 __hash_bucket(u32 hash, unsigned int size)
161 return reciprocal_scale(hash, size);
164 static u32 hash_bucket(u32 hash, const struct net *net)
166 return __hash_bucket(hash, net->ct.htable_size);
169 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
172 return __hash_bucket(hash_conntrack_raw(tuple), size);
175 static inline u_int32_t hash_conntrack(const struct net *net,
176 const struct nf_conntrack_tuple *tuple)
178 return __hash_conntrack(tuple, net->ct.htable_size);
182 nf_ct_get_tuple(const struct sk_buff *skb,
184 unsigned int dataoff,
188 struct nf_conntrack_tuple *tuple,
189 const struct nf_conntrack_l3proto *l3proto,
190 const struct nf_conntrack_l4proto *l4proto)
192 memset(tuple, 0, sizeof(*tuple));
194 tuple->src.l3num = l3num;
195 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
198 tuple->dst.protonum = protonum;
199 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
201 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
203 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
205 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
207 struct net *net, struct nf_conntrack_tuple *tuple)
209 struct nf_conntrack_l3proto *l3proto;
210 struct nf_conntrack_l4proto *l4proto;
211 unsigned int protoff;
217 l3proto = __nf_ct_l3proto_find(l3num);
218 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
219 if (ret != NF_ACCEPT) {
224 l4proto = __nf_ct_l4proto_find(l3num, protonum);
226 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
232 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
235 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
236 const struct nf_conntrack_tuple *orig,
237 const struct nf_conntrack_l3proto *l3proto,
238 const struct nf_conntrack_l4proto *l4proto)
240 memset(inverse, 0, sizeof(*inverse));
242 inverse->src.l3num = orig->src.l3num;
243 if (l3proto->invert_tuple(inverse, orig) == 0)
246 inverse->dst.dir = !orig->dst.dir;
248 inverse->dst.protonum = orig->dst.protonum;
249 return l4proto->invert_tuple(inverse, orig);
251 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
254 clean_from_lists(struct nf_conn *ct)
256 pr_debug("clean_from_lists(%p)\n", ct);
257 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
258 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
260 /* Destroy all pending expectations */
261 nf_ct_remove_expectations(ct);
264 /* must be called with local_bh_disable */
265 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
267 struct ct_pcpu *pcpu;
269 /* add this conntrack to the (per cpu) dying list */
270 ct->cpu = smp_processor_id();
271 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
273 spin_lock(&pcpu->lock);
274 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
276 spin_unlock(&pcpu->lock);
279 /* must be called with local_bh_disable */
280 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
282 struct ct_pcpu *pcpu;
284 /* add this conntrack to the (per cpu) unconfirmed list */
285 ct->cpu = smp_processor_id();
286 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
288 spin_lock(&pcpu->lock);
289 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
291 spin_unlock(&pcpu->lock);
294 /* must be called with local_bh_disable */
295 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
297 struct ct_pcpu *pcpu;
299 /* We overload first tuple to link into unconfirmed or dying list.*/
300 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
302 spin_lock(&pcpu->lock);
303 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
304 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
305 spin_unlock(&pcpu->lock);
308 /* Released via destroy_conntrack() */
309 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
310 const struct nf_conntrack_zone *zone,
313 struct nf_conn *tmpl;
315 tmpl = kzalloc(sizeof(*tmpl), flags);
319 tmpl->status = IPS_TEMPLATE;
320 write_pnet(&tmpl->ct_net, net);
322 if (nf_ct_zone_add(tmpl, flags, zone) < 0)
325 atomic_set(&tmpl->ct_general.use, 0);
332 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
334 void nf_ct_tmpl_free(struct nf_conn *tmpl)
336 nf_ct_ext_destroy(tmpl);
337 nf_ct_ext_free(tmpl);
340 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
343 destroy_conntrack(struct nf_conntrack *nfct)
345 struct nf_conn *ct = (struct nf_conn *)nfct;
346 struct net *net = nf_ct_net(ct);
347 struct nf_conntrack_l4proto *l4proto;
349 pr_debug("destroy_conntrack(%p)\n", ct);
350 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
351 NF_CT_ASSERT(!timer_pending(&ct->timeout));
353 if (unlikely(nf_ct_is_template(ct))) {
358 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
359 if (l4proto && l4proto->destroy)
360 l4proto->destroy(ct);
365 /* Expectations will have been removed in clean_from_lists,
366 * except TFTP can create an expectation on the first packet,
367 * before connection is in the list, so we need to clean here,
370 nf_ct_remove_expectations(ct);
372 nf_ct_del_from_dying_or_unconfirmed_list(ct);
374 NF_CT_STAT_INC(net, delete);
378 nf_ct_put(ct->master);
380 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
381 nf_conntrack_free(ct);
384 static void nf_ct_delete_from_lists(struct nf_conn *ct)
386 struct net *net = nf_ct_net(ct);
387 unsigned int hash, reply_hash;
388 unsigned int sequence;
390 nf_ct_helper_destroy(ct);
394 sequence = read_seqcount_begin(&net->ct.generation);
395 hash = hash_conntrack(net,
396 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
397 reply_hash = hash_conntrack(net,
398 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
399 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
401 clean_from_lists(ct);
402 nf_conntrack_double_unlock(hash, reply_hash);
404 nf_ct_add_to_dying_list(ct);
406 NF_CT_STAT_INC(net, delete_list);
410 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
412 struct nf_conn_tstamp *tstamp;
414 tstamp = nf_conn_tstamp_find(ct);
415 if (tstamp && tstamp->stop == 0)
416 tstamp->stop = ktime_get_real_ns();
418 if (nf_ct_is_dying(ct))
421 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
422 portid, report) < 0) {
423 /* destroy event was not delivered */
424 nf_ct_delete_from_lists(ct);
425 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
429 nf_conntrack_ecache_work(nf_ct_net(ct));
430 set_bit(IPS_DYING_BIT, &ct->status);
432 nf_ct_delete_from_lists(ct);
436 EXPORT_SYMBOL_GPL(nf_ct_delete);
438 static void death_by_timeout(unsigned long ul_conntrack)
440 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
444 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
445 const struct nf_conntrack_tuple *tuple,
446 const struct nf_conntrack_zone *zone)
448 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
450 /* A conntrack can be recreated with the equal tuple,
451 * so we need to check that the conntrack is confirmed
453 return nf_ct_tuple_equal(tuple, &h->tuple) &&
454 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
455 nf_ct_is_confirmed(ct);
460 * - Caller must take a reference on returned object
461 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
463 static struct nf_conntrack_tuple_hash *
464 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
465 const struct nf_conntrack_tuple *tuple, u32 hash)
467 struct nf_conntrack_tuple_hash *h;
468 struct hlist_nulls_node *n;
469 unsigned int bucket = hash_bucket(hash, net);
471 /* Disable BHs the entire time since we normally need to disable them
472 * at least once for the stats anyway.
476 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
477 if (nf_ct_key_equal(h, tuple, zone)) {
478 NF_CT_STAT_INC(net, found);
482 NF_CT_STAT_INC(net, searched);
485 * if the nulls value we got at the end of this lookup is
486 * not the expected one, we must restart lookup.
487 * We probably met an item that was moved to another chain.
489 if (get_nulls_value(n) != bucket) {
490 NF_CT_STAT_INC(net, search_restart);
498 /* Find a connection corresponding to a tuple. */
499 static struct nf_conntrack_tuple_hash *
500 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
501 const struct nf_conntrack_tuple *tuple, u32 hash)
503 struct nf_conntrack_tuple_hash *h;
508 h = ____nf_conntrack_find(net, zone, tuple, hash);
510 ct = nf_ct_tuplehash_to_ctrack(h);
511 if (unlikely(nf_ct_is_dying(ct) ||
512 !atomic_inc_not_zero(&ct->ct_general.use)))
515 if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
526 struct nf_conntrack_tuple_hash *
527 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
528 const struct nf_conntrack_tuple *tuple)
530 return __nf_conntrack_find_get(net, zone, tuple,
531 hash_conntrack_raw(tuple));
533 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
535 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
537 unsigned int reply_hash)
539 struct net *net = nf_ct_net(ct);
541 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
542 &net->ct.hash[hash]);
543 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
544 &net->ct.hash[reply_hash]);
548 nf_conntrack_hash_check_insert(struct nf_conn *ct)
550 const struct nf_conntrack_zone *zone;
551 struct net *net = nf_ct_net(ct);
552 unsigned int hash, reply_hash;
553 struct nf_conntrack_tuple_hash *h;
554 struct hlist_nulls_node *n;
555 unsigned int sequence;
557 zone = nf_ct_zone(ct);
561 sequence = read_seqcount_begin(&net->ct.generation);
562 hash = hash_conntrack(net,
563 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
564 reply_hash = hash_conntrack(net,
565 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
566 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
568 /* See if there's one in the list already, including reverse */
569 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
570 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
572 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
575 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
576 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
578 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
582 add_timer(&ct->timeout);
584 /* The caller holds a reference to this object */
585 atomic_set(&ct->ct_general.use, 2);
586 __nf_conntrack_hash_insert(ct, hash, reply_hash);
587 nf_conntrack_double_unlock(hash, reply_hash);
588 NF_CT_STAT_INC(net, insert);
593 nf_conntrack_double_unlock(hash, reply_hash);
594 NF_CT_STAT_INC(net, insert_failed);
598 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
600 /* Confirm a connection given skb; places it in hash table */
602 __nf_conntrack_confirm(struct sk_buff *skb)
604 const struct nf_conntrack_zone *zone;
605 unsigned int hash, reply_hash;
606 struct nf_conntrack_tuple_hash *h;
608 struct nf_conn_help *help;
609 struct nf_conn_tstamp *tstamp;
610 struct hlist_nulls_node *n;
611 enum ip_conntrack_info ctinfo;
613 unsigned int sequence;
615 ct = nf_ct_get(skb, &ctinfo);
618 /* ipt_REJECT uses nf_conntrack_attach to attach related
619 ICMP/TCP RST packets in other direction. Actual packet
620 which created connection will be IP_CT_NEW or for an
621 expected connection, IP_CT_RELATED. */
622 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
625 zone = nf_ct_zone(ct);
629 sequence = read_seqcount_begin(&net->ct.generation);
630 /* reuse the hash saved before */
631 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
632 hash = hash_bucket(hash, net);
633 reply_hash = hash_conntrack(net,
634 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
636 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
638 /* We're not in hash table, and we refuse to set up related
639 * connections for unconfirmed conns. But packet copies and
640 * REJECT will give spurious warnings here.
642 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
644 /* No external references means no one else could have
647 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
648 pr_debug("Confirming conntrack %p\n", ct);
649 /* We have to check the DYING flag after unlink to prevent
650 * a race against nf_ct_get_next_corpse() possibly called from
651 * user context, else we insert an already 'dead' hash, blocking
652 * further use of that particular connection -JM.
654 nf_ct_del_from_dying_or_unconfirmed_list(ct);
656 if (unlikely(nf_ct_is_dying(ct)))
659 /* See if there's one in the list already, including reverse:
660 NAT could have grabbed it without realizing, since we're
661 not in the hash. If there is, we lost race. */
662 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
663 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
665 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
668 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
669 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
671 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
675 /* Timer relative to confirmation time, not original
676 setting time, otherwise we'd get timer wrap in
677 weird delay cases. */
678 ct->timeout.expires += jiffies;
679 add_timer(&ct->timeout);
680 atomic_inc(&ct->ct_general.use);
681 ct->status |= IPS_CONFIRMED;
683 /* set conntrack timestamp, if enabled. */
684 tstamp = nf_conn_tstamp_find(ct);
686 if (skb->tstamp.tv64 == 0)
687 __net_timestamp(skb);
689 tstamp->start = ktime_to_ns(skb->tstamp);
691 /* Since the lookup is lockless, hash insertion must be done after
692 * starting the timer and setting the CONFIRMED bit. The RCU barriers
693 * guarantee that no other CPU can find the conntrack before the above
694 * stores are visible.
696 __nf_conntrack_hash_insert(ct, hash, reply_hash);
697 nf_conntrack_double_unlock(hash, reply_hash);
698 NF_CT_STAT_INC(net, insert);
701 help = nfct_help(ct);
702 if (help && help->helper)
703 nf_conntrack_event_cache(IPCT_HELPER, ct);
705 nf_conntrack_event_cache(master_ct(ct) ?
706 IPCT_RELATED : IPCT_NEW, ct);
710 nf_ct_add_to_dying_list(ct);
711 nf_conntrack_double_unlock(hash, reply_hash);
712 NF_CT_STAT_INC(net, insert_failed);
716 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
718 /* Returns true if a connection correspondings to the tuple (required
721 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
722 const struct nf_conn *ignored_conntrack)
724 struct net *net = nf_ct_net(ignored_conntrack);
725 const struct nf_conntrack_zone *zone;
726 struct nf_conntrack_tuple_hash *h;
727 struct hlist_nulls_node *n;
731 zone = nf_ct_zone(ignored_conntrack);
732 hash = hash_conntrack(net, tuple);
734 /* Disable BHs the entire time since we need to disable them at
735 * least once for the stats anyway.
738 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
739 ct = nf_ct_tuplehash_to_ctrack(h);
740 if (ct != ignored_conntrack &&
741 nf_ct_tuple_equal(tuple, &h->tuple) &&
742 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h))) {
743 NF_CT_STAT_INC(net, found);
744 rcu_read_unlock_bh();
747 NF_CT_STAT_INC(net, searched);
749 rcu_read_unlock_bh();
753 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
755 #define NF_CT_EVICTION_RANGE 8
757 /* There's a small race here where we may free a just-assured
758 connection. Too bad: we're in trouble anyway. */
759 static noinline int early_drop(struct net *net, unsigned int _hash)
761 /* Use oldest entry, which is roughly LRU */
762 struct nf_conntrack_tuple_hash *h;
763 struct nf_conn *ct = NULL, *tmp;
764 struct hlist_nulls_node *n;
765 unsigned int i = 0, cnt = 0;
767 unsigned int hash, sequence;
772 sequence = read_seqcount_begin(&net->ct.generation);
773 hash = hash_bucket(_hash, net);
774 for (; i < net->ct.htable_size; i++) {
775 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
776 nf_conntrack_lock(lockp);
777 if (read_seqcount_retry(&net->ct.generation, sequence)) {
781 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
783 tmp = nf_ct_tuplehash_to_ctrack(h);
784 if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
785 !nf_ct_is_dying(tmp) &&
786 atomic_inc_not_zero(&tmp->ct_general.use)) {
793 hash = (hash + 1) % net->ct.htable_size;
796 if (ct || cnt >= NF_CT_EVICTION_RANGE)
805 if (del_timer(&ct->timeout)) {
806 if (nf_ct_delete(ct, 0, 0)) {
808 NF_CT_STAT_INC_ATOMIC(net, early_drop);
815 void init_nf_conntrack_hash_rnd(void)
820 * Why not initialize nf_conntrack_rnd in a "init()" function ?
821 * Because there isn't enough entropy when system initializing,
822 * and we initialize it as late as possible.
825 get_random_bytes(&rand, sizeof(rand));
827 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
830 static struct nf_conn *
831 __nf_conntrack_alloc(struct net *net,
832 const struct nf_conntrack_zone *zone,
833 const struct nf_conntrack_tuple *orig,
834 const struct nf_conntrack_tuple *repl,
839 if (unlikely(!nf_conntrack_hash_rnd)) {
840 init_nf_conntrack_hash_rnd();
841 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
842 hash = hash_conntrack_raw(orig);
845 /* We don't want any race condition at early drop stage */
846 atomic_inc(&net->ct.count);
848 if (nf_conntrack_max &&
849 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
850 if (!early_drop(net, hash)) {
851 atomic_dec(&net->ct.count);
852 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
853 return ERR_PTR(-ENOMEM);
858 * Do not use kmem_cache_zalloc(), as this cache uses
859 * SLAB_DESTROY_BY_RCU.
861 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
865 spin_lock_init(&ct->lock);
866 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
867 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
868 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
869 /* save hash for reusing when confirming */
870 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
872 /* Don't set timer yet: wait for confirmation */
873 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
874 write_pnet(&ct->ct_net, net);
875 memset(&ct->__nfct_init_offset[0], 0,
876 offsetof(struct nf_conn, proto) -
877 offsetof(struct nf_conn, __nfct_init_offset[0]));
879 if (zone && nf_ct_zone_add(ct, GFP_ATOMIC, zone) < 0)
882 /* Because we use RCU lookups, we set ct_general.use to zero before
883 * this is inserted in any list.
885 atomic_set(&ct->ct_general.use, 0);
888 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
890 atomic_dec(&net->ct.count);
891 return ERR_PTR(-ENOMEM);
894 struct nf_conn *nf_conntrack_alloc(struct net *net,
895 const struct nf_conntrack_zone *zone,
896 const struct nf_conntrack_tuple *orig,
897 const struct nf_conntrack_tuple *repl,
900 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
902 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
904 void nf_conntrack_free(struct nf_conn *ct)
906 struct net *net = nf_ct_net(ct);
908 /* A freed object has refcnt == 0, that's
909 * the golden rule for SLAB_DESTROY_BY_RCU
911 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
913 nf_ct_ext_destroy(ct);
915 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
916 smp_mb__before_atomic();
917 atomic_dec(&net->ct.count);
919 EXPORT_SYMBOL_GPL(nf_conntrack_free);
922 /* Allocate a new conntrack: we return -ENOMEM if classification
923 failed due to stress. Otherwise it really is unclassifiable. */
924 static struct nf_conntrack_tuple_hash *
925 init_conntrack(struct net *net, struct nf_conn *tmpl,
926 const struct nf_conntrack_tuple *tuple,
927 struct nf_conntrack_l3proto *l3proto,
928 struct nf_conntrack_l4proto *l4proto,
930 unsigned int dataoff, u32 hash)
933 struct nf_conn_help *help;
934 struct nf_conntrack_tuple repl_tuple;
935 struct nf_conntrack_ecache *ecache;
936 struct nf_conntrack_expect *exp = NULL;
937 const struct nf_conntrack_zone *zone;
938 struct nf_conn_timeout *timeout_ext;
939 struct nf_conntrack_zone tmp;
940 unsigned int *timeouts;
942 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
943 pr_debug("Can't invert tuple.\n");
947 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
948 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
951 return (struct nf_conntrack_tuple_hash *)ct;
953 if (tmpl && nfct_synproxy(tmpl)) {
954 nfct_seqadj_ext_add(ct);
955 nfct_synproxy_ext_add(ct);
958 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
960 timeouts = nf_ct_timeout_data(timeout_ext);
961 if (unlikely(!timeouts))
962 timeouts = l4proto->get_timeouts(net);
964 timeouts = l4proto->get_timeouts(net);
967 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
968 nf_conntrack_free(ct);
969 pr_debug("init conntrack: can't track with proto module\n");
974 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
977 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
978 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
979 nf_ct_labels_ext_add(ct);
981 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
982 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
983 ecache ? ecache->expmask : 0,
987 if (net->ct.expect_count) {
988 spin_lock(&nf_conntrack_expect_lock);
989 exp = nf_ct_find_expectation(net, zone, tuple);
991 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
993 /* Welcome, Mr. Bond. We've been expecting you... */
994 __set_bit(IPS_EXPECTED_BIT, &ct->status);
995 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
996 ct->master = exp->master;
998 help = nf_ct_helper_ext_add(ct, exp->helper,
1001 rcu_assign_pointer(help->helper, exp->helper);
1004 #ifdef CONFIG_NF_CONNTRACK_MARK
1005 ct->mark = exp->master->mark;
1007 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1008 ct->secmark = exp->master->secmark;
1010 NF_CT_STAT_INC(net, expect_new);
1012 spin_unlock(&nf_conntrack_expect_lock);
1015 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1016 NF_CT_STAT_INC(net, new);
1019 /* Now it is inserted into the unconfirmed list, bump refcount */
1020 nf_conntrack_get(&ct->ct_general);
1021 nf_ct_add_to_unconfirmed_list(ct);
1027 exp->expectfn(ct, exp);
1028 nf_ct_expect_put(exp);
1031 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1034 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1035 static inline struct nf_conn *
1036 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1037 struct sk_buff *skb,
1038 unsigned int dataoff,
1041 struct nf_conntrack_l3proto *l3proto,
1042 struct nf_conntrack_l4proto *l4proto,
1044 enum ip_conntrack_info *ctinfo)
1046 const struct nf_conntrack_zone *zone;
1047 struct nf_conntrack_tuple tuple;
1048 struct nf_conntrack_tuple_hash *h;
1049 struct nf_conntrack_zone tmp;
1053 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1054 dataoff, l3num, protonum, net, &tuple, l3proto,
1056 pr_debug("resolve_normal_ct: Can't get tuple\n");
1060 /* look for tuple match */
1061 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1062 hash = hash_conntrack_raw(&tuple);
1063 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1065 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1066 skb, dataoff, hash);
1072 ct = nf_ct_tuplehash_to_ctrack(h);
1074 /* It exists; we have (non-exclusive) reference. */
1075 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1076 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1077 /* Please set reply bit if this packet OK */
1080 /* Once we've had two way comms, always ESTABLISHED. */
1081 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1082 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
1083 *ctinfo = IP_CT_ESTABLISHED;
1084 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1085 pr_debug("nf_conntrack_in: related packet for %p\n",
1087 *ctinfo = IP_CT_RELATED;
1089 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
1090 *ctinfo = IP_CT_NEW;
1094 skb->nfct = &ct->ct_general;
1095 skb->nfctinfo = *ctinfo;
1100 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1101 struct sk_buff *skb)
1103 struct nf_conn *ct, *tmpl = NULL;
1104 enum ip_conntrack_info ctinfo;
1105 struct nf_conntrack_l3proto *l3proto;
1106 struct nf_conntrack_l4proto *l4proto;
1107 unsigned int *timeouts;
1108 unsigned int dataoff;
1114 /* Previously seen (loopback or untracked)? Ignore. */
1115 tmpl = (struct nf_conn *)skb->nfct;
1116 if (!nf_ct_is_template(tmpl)) {
1117 NF_CT_STAT_INC_ATOMIC(net, ignore);
1123 /* rcu_read_lock()ed by nf_hook_slow */
1124 l3proto = __nf_ct_l3proto_find(pf);
1125 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1126 &dataoff, &protonum);
1128 pr_debug("not prepared to track yet or error occurred\n");
1129 NF_CT_STAT_INC_ATOMIC(net, error);
1130 NF_CT_STAT_INC_ATOMIC(net, invalid);
1135 l4proto = __nf_ct_l4proto_find(pf, protonum);
1137 /* It may be an special packet, error, unclean...
1138 * inverse of the return code tells to the netfilter
1139 * core what to do with the packet. */
1140 if (l4proto->error != NULL) {
1141 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1144 NF_CT_STAT_INC_ATOMIC(net, error);
1145 NF_CT_STAT_INC_ATOMIC(net, invalid);
1149 /* ICMP[v6] protocol trackers may assign one conntrack. */
1154 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1155 l3proto, l4proto, &set_reply, &ctinfo);
1157 /* Not valid part of a connection */
1158 NF_CT_STAT_INC_ATOMIC(net, invalid);
1164 /* Too stressed to deal. */
1165 NF_CT_STAT_INC_ATOMIC(net, drop);
1170 NF_CT_ASSERT(skb->nfct);
1172 /* Decide what timeout policy we want to apply to this flow. */
1173 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1175 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1177 /* Invalid: inverse of the return code tells
1178 * the netfilter core what to do */
1179 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1180 nf_conntrack_put(skb->nfct);
1182 NF_CT_STAT_INC_ATOMIC(net, invalid);
1183 if (ret == -NF_DROP)
1184 NF_CT_STAT_INC_ATOMIC(net, drop);
1189 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1190 nf_conntrack_event_cache(IPCT_REPLY, ct);
1193 /* Special case: we have to repeat this hook, assign the
1194 * template again to this packet. We assume that this packet
1195 * has no conntrack assigned. This is used by nf_ct_tcp. */
1196 if (ret == NF_REPEAT)
1197 skb->nfct = (struct nf_conntrack *)tmpl;
1204 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1206 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1207 const struct nf_conntrack_tuple *orig)
1212 ret = nf_ct_invert_tuple(inverse, orig,
1213 __nf_ct_l3proto_find(orig->src.l3num),
1214 __nf_ct_l4proto_find(orig->src.l3num,
1215 orig->dst.protonum));
1219 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1221 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1222 implicitly racy: see __nf_conntrack_confirm */
1223 void nf_conntrack_alter_reply(struct nf_conn *ct,
1224 const struct nf_conntrack_tuple *newreply)
1226 struct nf_conn_help *help = nfct_help(ct);
1228 /* Should be unconfirmed, so not in hash table yet */
1229 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1231 pr_debug("Altering reply tuple of %p to ", ct);
1232 nf_ct_dump_tuple(newreply);
1234 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1235 if (ct->master || (help && !hlist_empty(&help->expectations)))
1239 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1242 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1244 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1245 void __nf_ct_refresh_acct(struct nf_conn *ct,
1246 enum ip_conntrack_info ctinfo,
1247 const struct sk_buff *skb,
1248 unsigned long extra_jiffies,
1251 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1254 /* Only update if this is not a fixed timeout */
1255 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1258 /* If not in hash table, timer will not be active yet */
1259 if (!nf_ct_is_confirmed(ct)) {
1260 ct->timeout.expires = extra_jiffies;
1262 unsigned long newtime = jiffies + extra_jiffies;
1264 /* Only update the timeout if the new timeout is at least
1265 HZ jiffies from the old timeout. Need del_timer for race
1266 avoidance (may already be dying). */
1267 if (newtime - ct->timeout.expires >= HZ)
1268 mod_timer_pending(&ct->timeout, newtime);
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, &counter[CTINFO2DIR(ctinfo)].bytes);
1284 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1286 bool __nf_ct_kill_acct(struct nf_conn *ct,
1287 enum ip_conntrack_info ctinfo,
1288 const struct sk_buff *skb,
1292 struct nf_conn_acct *acct;
1294 acct = nf_conn_acct_find(ct);
1296 struct nf_conn_counter *counter = acct->counter;
1298 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1299 atomic64_add(skb->len - skb_network_offset(skb),
1300 &counter[CTINFO2DIR(ctinfo)].bytes);
1304 if (del_timer(&ct->timeout)) {
1305 ct->timeout.function((unsigned long)ct);
1310 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1312 #ifdef CONFIG_NF_CONNTRACK_ZONES
1313 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1314 .len = sizeof(struct nf_conntrack_zone),
1315 .align = __alignof__(struct nf_conntrack_zone),
1316 .id = NF_CT_EXT_ZONE,
1320 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1322 #include <linux/netfilter/nfnetlink.h>
1323 #include <linux/netfilter/nfnetlink_conntrack.h>
1324 #include <linux/mutex.h>
1326 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1327 * in ip_conntrack_core, since we don't want the protocols to autoload
1328 * or depend on ctnetlink */
1329 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1330 const struct nf_conntrack_tuple *tuple)
1332 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1333 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1334 goto nla_put_failure;
1340 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1342 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1343 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1344 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1346 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1348 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1349 struct nf_conntrack_tuple *t)
1351 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1354 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1355 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1359 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1361 int nf_ct_port_nlattr_tuple_size(void)
1363 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1365 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1368 /* Used by ipt_REJECT and ip6t_REJECT. */
1369 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1372 enum ip_conntrack_info ctinfo;
1374 /* This ICMP is in reverse direction to the packet which caused it */
1375 ct = nf_ct_get(skb, &ctinfo);
1376 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1377 ctinfo = IP_CT_RELATED_REPLY;
1379 ctinfo = IP_CT_RELATED;
1381 /* Attach to new skbuff, and increment count */
1382 nskb->nfct = &ct->ct_general;
1383 nskb->nfctinfo = ctinfo;
1384 nf_conntrack_get(nskb->nfct);
1387 /* Bring out ya dead! */
1388 static struct nf_conn *
1389 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1390 void *data, unsigned int *bucket)
1392 struct nf_conntrack_tuple_hash *h;
1394 struct hlist_nulls_node *n;
1398 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1399 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1401 nf_conntrack_lock(lockp);
1402 if (*bucket < net->ct.htable_size) {
1403 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1404 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1406 ct = nf_ct_tuplehash_to_ctrack(h);
1416 for_each_possible_cpu(cpu) {
1417 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1419 spin_lock_bh(&pcpu->lock);
1420 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1421 ct = nf_ct_tuplehash_to_ctrack(h);
1423 set_bit(IPS_DYING_BIT, &ct->status);
1425 spin_unlock_bh(&pcpu->lock);
1430 atomic_inc(&ct->ct_general.use);
1436 void nf_ct_iterate_cleanup(struct net *net,
1437 int (*iter)(struct nf_conn *i, void *data),
1438 void *data, u32 portid, int report)
1441 unsigned int bucket = 0;
1445 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1446 /* Time to push up daises... */
1447 if (del_timer(&ct->timeout))
1448 nf_ct_delete(ct, portid, report);
1450 /* ... else the timer will get him soon. */
1456 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1458 static int kill_all(struct nf_conn *i, void *data)
1463 void nf_ct_free_hashtable(void *hash, unsigned int size)
1465 if (is_vmalloc_addr(hash))
1468 free_pages((unsigned long)hash,
1469 get_order(sizeof(struct hlist_head) * size));
1471 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1473 static int untrack_refs(void)
1477 for_each_possible_cpu(cpu) {
1478 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1480 cnt += atomic_read(&ct->ct_general.use) - 1;
1485 void nf_conntrack_cleanup_start(void)
1487 RCU_INIT_POINTER(ip_ct_attach, NULL);
1490 void nf_conntrack_cleanup_end(void)
1492 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1493 while (untrack_refs() > 0)
1496 #ifdef CONFIG_NF_CONNTRACK_ZONES
1497 nf_ct_extend_unregister(&nf_ct_zone_extend);
1499 nf_conntrack_proto_fini();
1500 nf_conntrack_seqadj_fini();
1501 nf_conntrack_labels_fini();
1502 nf_conntrack_helper_fini();
1503 nf_conntrack_timeout_fini();
1504 nf_conntrack_ecache_fini();
1505 nf_conntrack_tstamp_fini();
1506 nf_conntrack_acct_fini();
1507 nf_conntrack_expect_fini();
1511 * Mishearing the voices in his head, our hero wonders how he's
1512 * supposed to kill the mall.
1514 void nf_conntrack_cleanup_net(struct net *net)
1518 list_add(&net->exit_list, &single);
1519 nf_conntrack_cleanup_net_list(&single);
1522 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1528 * This makes sure all current packets have passed through
1529 * netfilter framework. Roll on, two-stage module
1535 list_for_each_entry(net, net_exit_list, exit_list) {
1536 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1537 if (atomic_read(&net->ct.count) != 0)
1542 goto i_see_dead_people;
1545 list_for_each_entry(net, net_exit_list, exit_list) {
1546 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1547 nf_conntrack_proto_pernet_fini(net);
1548 nf_conntrack_helper_pernet_fini(net);
1549 nf_conntrack_ecache_pernet_fini(net);
1550 nf_conntrack_tstamp_pernet_fini(net);
1551 nf_conntrack_acct_pernet_fini(net);
1552 nf_conntrack_expect_pernet_fini(net);
1553 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1554 kfree(net->ct.slabname);
1555 free_percpu(net->ct.stat);
1556 free_percpu(net->ct.pcpu_lists);
1560 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1562 struct hlist_nulls_head *hash;
1563 unsigned int nr_slots, i;
1566 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1567 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1568 sz = nr_slots * sizeof(struct hlist_nulls_head);
1569 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1575 for (i = 0; i < nr_slots; i++)
1576 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1580 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1582 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1585 unsigned int hashsize, old_size;
1586 struct hlist_nulls_head *hash, *old_hash;
1587 struct nf_conntrack_tuple_hash *h;
1590 if (current->nsproxy->net_ns != &init_net)
1593 /* On boot, we can set this without any fancy locking. */
1594 if (!nf_conntrack_htable_size)
1595 return param_set_uint(val, kp);
1597 rc = kstrtouint(val, 0, &hashsize);
1603 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1608 nf_conntrack_all_lock();
1609 write_seqcount_begin(&init_net.ct.generation);
1611 /* Lookups in the old hash might happen in parallel, which means we
1612 * might get false negatives during connection lookup. New connections
1613 * created because of a false negative won't make it into the hash
1614 * though since that required taking the locks.
1617 for (i = 0; i < init_net.ct.htable_size; i++) {
1618 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1619 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1620 struct nf_conntrack_tuple_hash, hnnode);
1621 ct = nf_ct_tuplehash_to_ctrack(h);
1622 hlist_nulls_del_rcu(&h->hnnode);
1623 bucket = __hash_conntrack(&h->tuple, hashsize);
1624 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1627 old_size = init_net.ct.htable_size;
1628 old_hash = init_net.ct.hash;
1630 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1631 init_net.ct.hash = hash;
1633 write_seqcount_end(&init_net.ct.generation);
1634 nf_conntrack_all_unlock();
1637 nf_ct_free_hashtable(old_hash, old_size);
1640 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1642 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1643 &nf_conntrack_htable_size, 0600);
1645 void nf_ct_untracked_status_or(unsigned long bits)
1649 for_each_possible_cpu(cpu)
1650 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1652 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1654 int nf_conntrack_init_start(void)
1659 for (i = 0; i < CONNTRACK_LOCKS; i++)
1660 spin_lock_init(&nf_conntrack_locks[i]);
1662 if (!nf_conntrack_htable_size) {
1663 /* Idea from tcp.c: use 1/16384 of memory.
1664 * On i386: 32MB machine has 512 buckets.
1665 * >= 1GB machines have 16384 buckets.
1666 * >= 4GB machines have 65536 buckets.
1668 nf_conntrack_htable_size
1669 = (((totalram_pages << PAGE_SHIFT) / 16384)
1670 / sizeof(struct hlist_head));
1671 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1672 nf_conntrack_htable_size = 65536;
1673 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1674 nf_conntrack_htable_size = 16384;
1675 if (nf_conntrack_htable_size < 32)
1676 nf_conntrack_htable_size = 32;
1678 /* Use a max. factor of four by default to get the same max as
1679 * with the old struct list_heads. When a table size is given
1680 * we use the old value of 8 to avoid reducing the max.
1684 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1686 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1687 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1690 ret = nf_conntrack_expect_init();
1694 ret = nf_conntrack_acct_init();
1698 ret = nf_conntrack_tstamp_init();
1702 ret = nf_conntrack_ecache_init();
1706 ret = nf_conntrack_timeout_init();
1710 ret = nf_conntrack_helper_init();
1714 ret = nf_conntrack_labels_init();
1718 ret = nf_conntrack_seqadj_init();
1722 #ifdef CONFIG_NF_CONNTRACK_ZONES
1723 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1727 ret = nf_conntrack_proto_init();
1731 /* Set up fake conntrack: to never be deleted, not in any hashes */
1732 for_each_possible_cpu(cpu) {
1733 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1734 write_pnet(&ct->ct_net, &init_net);
1735 atomic_set(&ct->ct_general.use, 1);
1737 /* - and look it like as a confirmed connection */
1738 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1742 #ifdef CONFIG_NF_CONNTRACK_ZONES
1743 nf_ct_extend_unregister(&nf_ct_zone_extend);
1746 nf_conntrack_seqadj_fini();
1748 nf_conntrack_labels_fini();
1750 nf_conntrack_helper_fini();
1752 nf_conntrack_timeout_fini();
1754 nf_conntrack_ecache_fini();
1756 nf_conntrack_tstamp_fini();
1758 nf_conntrack_acct_fini();
1760 nf_conntrack_expect_fini();
1765 void nf_conntrack_init_end(void)
1767 /* For use by REJECT target */
1768 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1769 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1773 * We need to use special "null" values, not used in hash table
1775 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1776 #define DYING_NULLS_VAL ((1<<30)+1)
1777 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1779 int nf_conntrack_init_net(struct net *net)
1784 atomic_set(&net->ct.count, 0);
1785 seqcount_init(&net->ct.generation);
1787 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1788 if (!net->ct.pcpu_lists)
1791 for_each_possible_cpu(cpu) {
1792 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1794 spin_lock_init(&pcpu->lock);
1795 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1796 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1799 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1801 goto err_pcpu_lists;
1803 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1804 if (!net->ct.slabname)
1807 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1808 sizeof(struct nf_conn), 0,
1809 SLAB_DESTROY_BY_RCU, NULL);
1810 if (!net->ct.nf_conntrack_cachep) {
1811 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1815 net->ct.htable_size = nf_conntrack_htable_size;
1816 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1817 if (!net->ct.hash) {
1818 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1821 ret = nf_conntrack_expect_pernet_init(net);
1824 ret = nf_conntrack_acct_pernet_init(net);
1827 ret = nf_conntrack_tstamp_pernet_init(net);
1830 ret = nf_conntrack_ecache_pernet_init(net);
1833 ret = nf_conntrack_helper_pernet_init(net);
1836 ret = nf_conntrack_proto_pernet_init(net);
1842 nf_conntrack_helper_pernet_fini(net);
1844 nf_conntrack_ecache_pernet_fini(net);
1846 nf_conntrack_tstamp_pernet_fini(net);
1848 nf_conntrack_acct_pernet_fini(net);
1850 nf_conntrack_expect_pernet_fini(net);
1852 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1854 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1856 kfree(net->ct.slabname);
1858 free_percpu(net->ct.stat);
1860 free_percpu(net->ct.pcpu_lists);