1 /* Connection state tracking for netfilter. This is separated from,
2 but required by, the NAT layer; it can also be used by an iptables
5 /* (C) 1999-2001 Paul `Rusty' Russell
6 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
7 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
8 * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/types.h>
18 #include <linux/netfilter.h>
19 #include <linux/module.h>
20 #include <linux/sched.h>
21 #include <linux/skbuff.h>
22 #include <linux/proc_fs.h>
23 #include <linux/vmalloc.h>
24 #include <linux/stddef.h>
25 #include <linux/slab.h>
26 #include <linux/random.h>
27 #include <linux/jhash.h>
28 #include <linux/err.h>
29 #include <linux/percpu.h>
30 #include <linux/moduleparam.h>
31 #include <linux/notifier.h>
32 #include <linux/kernel.h>
33 #include <linux/netdevice.h>
34 #include <linux/socket.h>
36 #include <linux/nsproxy.h>
37 #include <linux/rculist_nulls.h>
39 #include <net/netfilter/nf_conntrack.h>
40 #include <net/netfilter/nf_conntrack_l3proto.h>
41 #include <net/netfilter/nf_conntrack_l4proto.h>
42 #include <net/netfilter/nf_conntrack_expect.h>
43 #include <net/netfilter/nf_conntrack_helper.h>
44 #include <net/netfilter/nf_conntrack_seqadj.h>
45 #include <net/netfilter/nf_conntrack_core.h>
46 #include <net/netfilter/nf_conntrack_extend.h>
47 #include <net/netfilter/nf_conntrack_acct.h>
48 #include <net/netfilter/nf_conntrack_ecache.h>
49 #include <net/netfilter/nf_conntrack_zones.h>
50 #include <net/netfilter/nf_conntrack_timestamp.h>
51 #include <net/netfilter/nf_conntrack_timeout.h>
52 #include <net/netfilter/nf_conntrack_labels.h>
53 #include <net/netfilter/nf_conntrack_synproxy.h>
54 #include <net/netfilter/nf_nat.h>
55 #include <net/netfilter/nf_nat_core.h>
56 #include <net/netfilter/nf_nat_helper.h>
57 #include <net/netns/hash.h>
59 #define NF_CONNTRACK_VERSION "0.5.0"
61 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
62 enum nf_nat_manip_type manip,
63 const struct nlattr *attr) __read_mostly;
64 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
66 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
67 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
69 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
70 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
72 struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
73 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
75 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
76 static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
77 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
78 static __read_mostly bool nf_conntrack_locks_all;
80 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
83 while (unlikely(nf_conntrack_locks_all)) {
87 * Order the 'nf_conntrack_locks_all' load vs. the
88 * spin_unlock_wait() loads below, to ensure
89 * that 'nf_conntrack_locks_all_lock' is indeed held:
91 smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
92 spin_unlock_wait(&nf_conntrack_locks_all_lock);
96 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
98 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
100 h1 %= CONNTRACK_LOCKS;
101 h2 %= CONNTRACK_LOCKS;
102 spin_unlock(&nf_conntrack_locks[h1]);
104 spin_unlock(&nf_conntrack_locks[h2]);
107 /* return true if we need to recompute hashes (in case hash table was resized) */
108 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
109 unsigned int h2, unsigned int sequence)
111 h1 %= CONNTRACK_LOCKS;
112 h2 %= CONNTRACK_LOCKS;
114 nf_conntrack_lock(&nf_conntrack_locks[h1]);
116 spin_lock_nested(&nf_conntrack_locks[h2],
117 SINGLE_DEPTH_NESTING);
119 nf_conntrack_lock(&nf_conntrack_locks[h2]);
120 spin_lock_nested(&nf_conntrack_locks[h1],
121 SINGLE_DEPTH_NESTING);
123 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
124 nf_conntrack_double_unlock(h1, h2);
130 static void nf_conntrack_all_lock(void)
134 spin_lock(&nf_conntrack_locks_all_lock);
135 nf_conntrack_locks_all = true;
138 * Order the above store of 'nf_conntrack_locks_all' against
139 * the spin_unlock_wait() loads below, such that if
140 * nf_conntrack_lock() observes 'nf_conntrack_locks_all'
141 * we must observe nf_conntrack_locks[] held:
143 smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
145 for (i = 0; i < CONNTRACK_LOCKS; i++) {
146 spin_unlock_wait(&nf_conntrack_locks[i]);
150 static void nf_conntrack_all_unlock(void)
153 * All prior stores must be complete before we clear
154 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
155 * might observe the false value but not the entire
158 smp_store_release(&nf_conntrack_locks_all, false);
159 spin_unlock(&nf_conntrack_locks_all_lock);
162 unsigned int nf_conntrack_htable_size __read_mostly;
163 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
165 unsigned int nf_conntrack_max __read_mostly;
166 seqcount_t nf_conntrack_generation __read_mostly;
168 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
169 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
171 static unsigned int nf_conntrack_hash_rnd __read_mostly;
173 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
174 const struct net *net)
179 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
181 /* The direction must be ignored, so we hash everything up to the
182 * destination ports (which is a multiple of 4) and treat the last
183 * three bytes manually.
185 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
186 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
187 return jhash2((u32 *)tuple, n, seed ^
188 (((__force __u16)tuple->dst.u.all << 16) |
189 tuple->dst.protonum));
192 static u32 scale_hash(u32 hash)
194 return reciprocal_scale(hash, nf_conntrack_htable_size);
197 static u32 __hash_conntrack(const struct net *net,
198 const struct nf_conntrack_tuple *tuple,
201 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
204 static u32 hash_conntrack(const struct net *net,
205 const struct nf_conntrack_tuple *tuple)
207 return scale_hash(hash_conntrack_raw(tuple, net));
211 nf_ct_get_tuple(const struct sk_buff *skb,
213 unsigned int dataoff,
217 struct nf_conntrack_tuple *tuple,
218 const struct nf_conntrack_l3proto *l3proto,
219 const struct nf_conntrack_l4proto *l4proto)
221 memset(tuple, 0, sizeof(*tuple));
223 tuple->src.l3num = l3num;
224 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
227 tuple->dst.protonum = protonum;
228 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
230 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
232 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
234 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
236 struct net *net, struct nf_conntrack_tuple *tuple)
238 struct nf_conntrack_l3proto *l3proto;
239 struct nf_conntrack_l4proto *l4proto;
240 unsigned int protoff;
246 l3proto = __nf_ct_l3proto_find(l3num);
247 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
248 if (ret != NF_ACCEPT) {
253 l4proto = __nf_ct_l4proto_find(l3num, protonum);
255 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
261 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
264 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
265 const struct nf_conntrack_tuple *orig,
266 const struct nf_conntrack_l3proto *l3proto,
267 const struct nf_conntrack_l4proto *l4proto)
269 memset(inverse, 0, sizeof(*inverse));
271 inverse->src.l3num = orig->src.l3num;
272 if (l3proto->invert_tuple(inverse, orig) == 0)
275 inverse->dst.dir = !orig->dst.dir;
277 inverse->dst.protonum = orig->dst.protonum;
278 return l4proto->invert_tuple(inverse, orig);
280 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
283 clean_from_lists(struct nf_conn *ct)
285 pr_debug("clean_from_lists(%p)\n", ct);
286 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
287 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
289 /* Destroy all pending expectations */
290 nf_ct_remove_expectations(ct);
293 /* must be called with local_bh_disable */
294 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
296 struct ct_pcpu *pcpu;
298 /* add this conntrack to the (per cpu) dying list */
299 ct->cpu = smp_processor_id();
300 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
302 spin_lock(&pcpu->lock);
303 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
305 spin_unlock(&pcpu->lock);
308 /* must be called with local_bh_disable */
309 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
311 struct ct_pcpu *pcpu;
313 /* add this conntrack to the (per cpu) unconfirmed list */
314 ct->cpu = smp_processor_id();
315 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
317 spin_lock(&pcpu->lock);
318 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
320 spin_unlock(&pcpu->lock);
323 /* must be called with local_bh_disable */
324 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
326 struct ct_pcpu *pcpu;
328 /* We overload first tuple to link into unconfirmed or dying list.*/
329 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
331 spin_lock(&pcpu->lock);
332 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
333 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
334 spin_unlock(&pcpu->lock);
337 /* Released via destroy_conntrack() */
338 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
339 const struct nf_conntrack_zone *zone,
342 struct nf_conn *tmpl;
344 tmpl = kzalloc(sizeof(*tmpl), flags);
348 tmpl->status = IPS_TEMPLATE;
349 write_pnet(&tmpl->ct_net, net);
350 nf_ct_zone_add(tmpl, zone);
351 atomic_set(&tmpl->ct_general.use, 0);
355 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
357 void nf_ct_tmpl_free(struct nf_conn *tmpl)
359 nf_ct_ext_destroy(tmpl);
360 nf_ct_ext_free(tmpl);
363 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
366 destroy_conntrack(struct nf_conntrack *nfct)
368 struct nf_conn *ct = (struct nf_conn *)nfct;
369 struct net *net = nf_ct_net(ct);
370 struct nf_conntrack_l4proto *l4proto;
372 pr_debug("destroy_conntrack(%p)\n", ct);
373 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
375 if (unlikely(nf_ct_is_template(ct))) {
380 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
381 if (l4proto->destroy)
382 l4proto->destroy(ct);
387 /* Expectations will have been removed in clean_from_lists,
388 * except TFTP can create an expectation on the first packet,
389 * before connection is in the list, so we need to clean here,
392 nf_ct_remove_expectations(ct);
394 nf_ct_del_from_dying_or_unconfirmed_list(ct);
396 NF_CT_STAT_INC(net, delete);
400 nf_ct_put(ct->master);
402 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
403 nf_conntrack_free(ct);
406 static void nf_ct_delete_from_lists(struct nf_conn *ct)
408 struct net *net = nf_ct_net(ct);
409 unsigned int hash, reply_hash;
410 unsigned int sequence;
412 nf_ct_helper_destroy(ct);
416 sequence = read_seqcount_begin(&nf_conntrack_generation);
417 hash = hash_conntrack(net,
418 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
419 reply_hash = hash_conntrack(net,
420 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
421 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
423 clean_from_lists(ct);
424 nf_conntrack_double_unlock(hash, reply_hash);
426 nf_ct_add_to_dying_list(ct);
428 NF_CT_STAT_INC(net, delete_list);
432 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
434 struct nf_conn_tstamp *tstamp;
436 if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
439 tstamp = nf_conn_tstamp_find(ct);
440 if (tstamp && tstamp->stop == 0)
441 tstamp->stop = ktime_get_real_ns();
443 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
444 portid, report) < 0) {
445 /* destroy event was not delivered. nf_ct_put will
446 * be done by event cache worker on redelivery.
448 nf_ct_delete_from_lists(ct);
449 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
453 nf_conntrack_ecache_work(nf_ct_net(ct));
454 nf_ct_delete_from_lists(ct);
458 EXPORT_SYMBOL_GPL(nf_ct_delete);
461 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
462 const struct nf_conntrack_tuple *tuple,
463 const struct nf_conntrack_zone *zone,
464 const struct net *net)
466 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
468 /* A conntrack can be recreated with the equal tuple,
469 * so we need to check that the conntrack is confirmed
471 return nf_ct_tuple_equal(tuple, &h->tuple) &&
472 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
473 nf_ct_is_confirmed(ct) &&
474 net_eq(net, nf_ct_net(ct));
477 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
478 static void nf_ct_gc_expired(struct nf_conn *ct)
480 if (!atomic_inc_not_zero(&ct->ct_general.use))
483 if (nf_ct_should_gc(ct))
491 * - Caller must take a reference on returned object
492 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
494 static struct nf_conntrack_tuple_hash *
495 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
496 const struct nf_conntrack_tuple *tuple, u32 hash)
498 struct nf_conntrack_tuple_hash *h;
499 struct hlist_nulls_head *ct_hash;
500 struct hlist_nulls_node *n;
501 unsigned int bucket, hsize;
504 nf_conntrack_get_ht(&ct_hash, &hsize);
505 bucket = reciprocal_scale(hash, hsize);
507 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
510 ct = nf_ct_tuplehash_to_ctrack(h);
511 if (nf_ct_is_expired(ct)) {
512 nf_ct_gc_expired(ct);
516 if (nf_ct_is_dying(ct))
519 if (nf_ct_key_equal(h, tuple, zone, net)) {
520 NF_CT_STAT_INC_ATOMIC(net, found);
523 NF_CT_STAT_INC_ATOMIC(net, searched);
526 * if the nulls value we got at the end of this lookup is
527 * not the expected one, we must restart lookup.
528 * We probably met an item that was moved to another chain.
530 if (get_nulls_value(n) != bucket) {
531 NF_CT_STAT_INC_ATOMIC(net, search_restart);
538 /* Find a connection corresponding to a tuple. */
539 static struct nf_conntrack_tuple_hash *
540 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
541 const struct nf_conntrack_tuple *tuple, u32 hash)
543 struct nf_conntrack_tuple_hash *h;
548 h = ____nf_conntrack_find(net, zone, tuple, hash);
550 ct = nf_ct_tuplehash_to_ctrack(h);
551 if (unlikely(nf_ct_is_dying(ct) ||
552 !atomic_inc_not_zero(&ct->ct_general.use)))
555 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
566 struct nf_conntrack_tuple_hash *
567 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
568 const struct nf_conntrack_tuple *tuple)
570 return __nf_conntrack_find_get(net, zone, tuple,
571 hash_conntrack_raw(tuple, net));
573 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
575 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
577 unsigned int reply_hash)
579 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
580 &nf_conntrack_hash[hash]);
581 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
582 &nf_conntrack_hash[reply_hash]);
586 nf_conntrack_hash_check_insert(struct nf_conn *ct)
588 const struct nf_conntrack_zone *zone;
589 struct net *net = nf_ct_net(ct);
590 unsigned int hash, reply_hash;
591 struct nf_conntrack_tuple_hash *h;
592 struct hlist_nulls_node *n;
593 unsigned int sequence;
595 zone = nf_ct_zone(ct);
599 sequence = read_seqcount_begin(&nf_conntrack_generation);
600 hash = hash_conntrack(net,
601 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
602 reply_hash = hash_conntrack(net,
603 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
604 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
606 /* See if there's one in the list already, including reverse */
607 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
608 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
612 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
613 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
618 /* The caller holds a reference to this object */
619 atomic_set(&ct->ct_general.use, 2);
620 __nf_conntrack_hash_insert(ct, hash, reply_hash);
621 nf_conntrack_double_unlock(hash, reply_hash);
622 NF_CT_STAT_INC(net, insert);
627 nf_conntrack_double_unlock(hash, reply_hash);
628 NF_CT_STAT_INC(net, insert_failed);
632 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
634 static inline void nf_ct_acct_update(struct nf_conn *ct,
635 enum ip_conntrack_info ctinfo,
638 struct nf_conn_acct *acct;
640 acct = nf_conn_acct_find(ct);
642 struct nf_conn_counter *counter = acct->counter;
644 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
645 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
649 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
650 const struct nf_conn *loser_ct)
652 struct nf_conn_acct *acct;
654 acct = nf_conn_acct_find(loser_ct);
656 struct nf_conn_counter *counter = acct->counter;
659 /* u32 should be fine since we must have seen one packet. */
660 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
661 nf_ct_acct_update(ct, ctinfo, bytes);
665 /* Resolve race on insertion if this protocol allows this. */
666 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
667 enum ip_conntrack_info ctinfo,
668 struct nf_conntrack_tuple_hash *h)
670 /* This is the conntrack entry already in hashes that won race. */
671 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
672 struct nf_conntrack_l4proto *l4proto;
674 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
675 if (l4proto->allow_clash &&
677 !nf_ct_is_dying(ct) &&
678 atomic_inc_not_zero(&ct->ct_general.use)) {
679 nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct);
680 nf_conntrack_put(skb->nfct);
681 /* Assign conntrack already in hashes to this skbuff. Don't
682 * modify skb->nfctinfo to ensure consistent stateful filtering.
684 skb->nfct = &ct->ct_general;
687 NF_CT_STAT_INC(net, drop);
691 /* Confirm a connection given skb; places it in hash table */
693 __nf_conntrack_confirm(struct sk_buff *skb)
695 const struct nf_conntrack_zone *zone;
696 unsigned int hash, reply_hash;
697 struct nf_conntrack_tuple_hash *h;
699 struct nf_conn_help *help;
700 struct nf_conn_tstamp *tstamp;
701 struct hlist_nulls_node *n;
702 enum ip_conntrack_info ctinfo;
704 unsigned int sequence;
707 ct = nf_ct_get(skb, &ctinfo);
710 /* ipt_REJECT uses nf_conntrack_attach to attach related
711 ICMP/TCP RST packets in other direction. Actual packet
712 which created connection will be IP_CT_NEW or for an
713 expected connection, IP_CT_RELATED. */
714 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
717 zone = nf_ct_zone(ct);
721 sequence = read_seqcount_begin(&nf_conntrack_generation);
722 /* reuse the hash saved before */
723 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
724 hash = scale_hash(hash);
725 reply_hash = hash_conntrack(net,
726 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
728 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
730 /* We're not in hash table, and we refuse to set up related
731 * connections for unconfirmed conns. But packet copies and
732 * REJECT will give spurious warnings here.
734 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
736 /* No external references means no one else could have
739 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
740 pr_debug("Confirming conntrack %p\n", ct);
741 /* We have to check the DYING flag after unlink to prevent
742 * a race against nf_ct_get_next_corpse() possibly called from
743 * user context, else we insert an already 'dead' hash, blocking
744 * further use of that particular connection -JM.
746 nf_ct_del_from_dying_or_unconfirmed_list(ct);
748 if (unlikely(nf_ct_is_dying(ct))) {
749 nf_ct_add_to_dying_list(ct);
753 /* See if there's one in the list already, including reverse:
754 NAT could have grabbed it without realizing, since we're
755 not in the hash. If there is, we lost race. */
756 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
757 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
761 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
762 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
766 /* Timer relative to confirmation time, not original
767 setting time, otherwise we'd get timer wrap in
768 weird delay cases. */
769 ct->timeout += nfct_time_stamp;
770 atomic_inc(&ct->ct_general.use);
771 ct->status |= IPS_CONFIRMED;
773 /* set conntrack timestamp, if enabled. */
774 tstamp = nf_conn_tstamp_find(ct);
776 if (skb->tstamp.tv64 == 0)
777 __net_timestamp(skb);
779 tstamp->start = ktime_to_ns(skb->tstamp);
781 /* Since the lookup is lockless, hash insertion must be done after
782 * starting the timer and setting the CONFIRMED bit. The RCU barriers
783 * guarantee that no other CPU can find the conntrack before the above
784 * stores are visible.
786 __nf_conntrack_hash_insert(ct, hash, reply_hash);
787 nf_conntrack_double_unlock(hash, reply_hash);
788 NF_CT_STAT_INC(net, insert);
791 help = nfct_help(ct);
792 if (help && help->helper)
793 nf_conntrack_event_cache(IPCT_HELPER, ct);
795 nf_conntrack_event_cache(master_ct(ct) ?
796 IPCT_RELATED : IPCT_NEW, ct);
800 nf_ct_add_to_dying_list(ct);
801 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
803 nf_conntrack_double_unlock(hash, reply_hash);
804 NF_CT_STAT_INC(net, insert_failed);
808 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
810 /* Returns true if a connection correspondings to the tuple (required
813 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
814 const struct nf_conn *ignored_conntrack)
816 struct net *net = nf_ct_net(ignored_conntrack);
817 const struct nf_conntrack_zone *zone;
818 struct nf_conntrack_tuple_hash *h;
819 struct hlist_nulls_head *ct_hash;
820 unsigned int hash, hsize;
821 struct hlist_nulls_node *n;
824 zone = nf_ct_zone(ignored_conntrack);
828 nf_conntrack_get_ht(&ct_hash, &hsize);
829 hash = __hash_conntrack(net, tuple, hsize);
831 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
832 ct = nf_ct_tuplehash_to_ctrack(h);
834 if (ct == ignored_conntrack)
837 if (nf_ct_is_expired(ct)) {
838 nf_ct_gc_expired(ct);
842 if (nf_ct_key_equal(h, tuple, zone, net)) {
843 NF_CT_STAT_INC_ATOMIC(net, found);
847 NF_CT_STAT_INC_ATOMIC(net, searched);
850 if (get_nulls_value(n) != hash) {
851 NF_CT_STAT_INC_ATOMIC(net, search_restart);
859 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
861 #define NF_CT_EVICTION_RANGE 8
863 /* There's a small race here where we may free a just-assured
864 connection. Too bad: we're in trouble anyway. */
865 static unsigned int early_drop_list(struct net *net,
866 struct hlist_nulls_head *head)
868 struct nf_conntrack_tuple_hash *h;
869 struct hlist_nulls_node *n;
870 unsigned int drops = 0;
873 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
874 tmp = nf_ct_tuplehash_to_ctrack(h);
876 if (nf_ct_is_expired(tmp)) {
877 nf_ct_gc_expired(tmp);
881 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
882 !net_eq(nf_ct_net(tmp), net) ||
886 if (!atomic_inc_not_zero(&tmp->ct_general.use))
889 /* kill only if still in same netns -- might have moved due to
890 * SLAB_DESTROY_BY_RCU rules.
892 * We steal the timer reference. If that fails timer has
893 * already fired or someone else deleted it. Just drop ref
894 * and move to next entry.
896 if (net_eq(nf_ct_net(tmp), net) &&
897 nf_ct_is_confirmed(tmp) &&
898 nf_ct_delete(tmp, 0, 0))
907 static noinline int early_drop(struct net *net, unsigned int _hash)
911 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
912 struct hlist_nulls_head *ct_hash;
913 unsigned int hash, hsize, drops;
916 nf_conntrack_get_ht(&ct_hash, &hsize);
917 hash = reciprocal_scale(_hash++, hsize);
919 drops = early_drop_list(net, &ct_hash[hash]);
923 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
931 static struct nf_conn *
932 __nf_conntrack_alloc(struct net *net,
933 const struct nf_conntrack_zone *zone,
934 const struct nf_conntrack_tuple *orig,
935 const struct nf_conntrack_tuple *repl,
940 /* We don't want any race condition at early drop stage */
941 atomic_inc(&net->ct.count);
943 if (nf_conntrack_max &&
944 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
945 if (!early_drop(net, hash)) {
946 atomic_dec(&net->ct.count);
947 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
948 return ERR_PTR(-ENOMEM);
953 * Do not use kmem_cache_zalloc(), as this cache uses
954 * SLAB_DESTROY_BY_RCU.
956 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
960 spin_lock_init(&ct->lock);
961 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
962 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
963 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
964 /* save hash for reusing when confirming */
965 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
967 write_pnet(&ct->ct_net, net);
968 memset(&ct->__nfct_init_offset[0], 0,
969 offsetof(struct nf_conn, proto) -
970 offsetof(struct nf_conn, __nfct_init_offset[0]));
972 nf_ct_zone_add(ct, zone);
974 /* Because we use RCU lookups, we set ct_general.use to zero before
975 * this is inserted in any list.
977 atomic_set(&ct->ct_general.use, 0);
980 atomic_dec(&net->ct.count);
981 return ERR_PTR(-ENOMEM);
984 struct nf_conn *nf_conntrack_alloc(struct net *net,
985 const struct nf_conntrack_zone *zone,
986 const struct nf_conntrack_tuple *orig,
987 const struct nf_conntrack_tuple *repl,
990 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
992 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
994 void nf_conntrack_free(struct nf_conn *ct)
996 struct net *net = nf_ct_net(ct);
998 /* A freed object has refcnt == 0, that's
999 * the golden rule for SLAB_DESTROY_BY_RCU
1001 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
1003 nf_ct_ext_destroy(ct);
1005 kmem_cache_free(nf_conntrack_cachep, ct);
1006 smp_mb__before_atomic();
1007 atomic_dec(&net->ct.count);
1009 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1012 /* Allocate a new conntrack: we return -ENOMEM if classification
1013 failed due to stress. Otherwise it really is unclassifiable. */
1014 static struct nf_conntrack_tuple_hash *
1015 init_conntrack(struct net *net, struct nf_conn *tmpl,
1016 const struct nf_conntrack_tuple *tuple,
1017 struct nf_conntrack_l3proto *l3proto,
1018 struct nf_conntrack_l4proto *l4proto,
1019 struct sk_buff *skb,
1020 unsigned int dataoff, u32 hash)
1023 struct nf_conn_help *help;
1024 struct nf_conntrack_tuple repl_tuple;
1025 struct nf_conntrack_ecache *ecache;
1026 struct nf_conntrack_expect *exp = NULL;
1027 const struct nf_conntrack_zone *zone;
1028 struct nf_conn_timeout *timeout_ext;
1029 struct nf_conntrack_zone tmp;
1030 unsigned int *timeouts;
1032 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1033 pr_debug("Can't invert tuple.\n");
1037 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1038 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1041 return (struct nf_conntrack_tuple_hash *)ct;
1043 if (tmpl && nfct_synproxy(tmpl)) {
1044 nfct_seqadj_ext_add(ct);
1045 nfct_synproxy_ext_add(ct);
1048 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1050 timeouts = nf_ct_timeout_data(timeout_ext);
1051 if (unlikely(!timeouts))
1052 timeouts = l4proto->get_timeouts(net);
1054 timeouts = l4proto->get_timeouts(net);
1057 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1058 nf_conntrack_free(ct);
1059 pr_debug("can't track with proto module\n");
1064 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1067 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1068 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1069 nf_ct_labels_ext_add(ct);
1071 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1072 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1073 ecache ? ecache->expmask : 0,
1077 if (net->ct.expect_count) {
1078 spin_lock(&nf_conntrack_expect_lock);
1079 exp = nf_ct_find_expectation(net, zone, tuple);
1081 pr_debug("expectation arrives ct=%p exp=%p\n",
1083 /* Welcome, Mr. Bond. We've been expecting you... */
1084 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1085 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1086 ct->master = exp->master;
1088 help = nf_ct_helper_ext_add(ct, exp->helper,
1091 rcu_assign_pointer(help->helper, exp->helper);
1094 #ifdef CONFIG_NF_CONNTRACK_MARK
1095 ct->mark = exp->master->mark;
1097 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1098 ct->secmark = exp->master->secmark;
1100 NF_CT_STAT_INC(net, expect_new);
1102 spin_unlock(&nf_conntrack_expect_lock);
1105 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1106 NF_CT_STAT_INC(net, new);
1109 /* Now it is inserted into the unconfirmed list, bump refcount */
1110 nf_conntrack_get(&ct->ct_general);
1111 nf_ct_add_to_unconfirmed_list(ct);
1117 exp->expectfn(ct, exp);
1118 nf_ct_expect_put(exp);
1121 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1124 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1125 static inline struct nf_conn *
1126 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1127 struct sk_buff *skb,
1128 unsigned int dataoff,
1131 struct nf_conntrack_l3proto *l3proto,
1132 struct nf_conntrack_l4proto *l4proto,
1134 enum ip_conntrack_info *ctinfo)
1136 const struct nf_conntrack_zone *zone;
1137 struct nf_conntrack_tuple tuple;
1138 struct nf_conntrack_tuple_hash *h;
1139 struct nf_conntrack_zone tmp;
1143 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1144 dataoff, l3num, protonum, net, &tuple, l3proto,
1146 pr_debug("Can't get tuple\n");
1150 /* look for tuple match */
1151 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1152 hash = hash_conntrack_raw(&tuple, net);
1153 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1155 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1156 skb, dataoff, hash);
1162 ct = nf_ct_tuplehash_to_ctrack(h);
1164 /* It exists; we have (non-exclusive) reference. */
1165 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1166 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1167 /* Please set reply bit if this packet OK */
1170 /* Once we've had two way comms, always ESTABLISHED. */
1171 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1172 pr_debug("normal packet for %p\n", ct);
1173 *ctinfo = IP_CT_ESTABLISHED;
1174 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1175 pr_debug("related packet for %p\n", ct);
1176 *ctinfo = IP_CT_RELATED;
1178 pr_debug("new packet for %p\n", ct);
1179 *ctinfo = IP_CT_NEW;
1183 skb->nfct = &ct->ct_general;
1184 skb->nfctinfo = *ctinfo;
1189 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1190 struct sk_buff *skb)
1192 struct nf_conn *ct, *tmpl = NULL;
1193 enum ip_conntrack_info ctinfo;
1194 struct nf_conntrack_l3proto *l3proto;
1195 struct nf_conntrack_l4proto *l4proto;
1196 unsigned int *timeouts;
1197 unsigned int dataoff;
1203 /* Previously seen (loopback or untracked)? Ignore. */
1204 tmpl = (struct nf_conn *)skb->nfct;
1205 if (!nf_ct_is_template(tmpl)) {
1206 NF_CT_STAT_INC_ATOMIC(net, ignore);
1212 /* rcu_read_lock()ed by nf_hook_slow */
1213 l3proto = __nf_ct_l3proto_find(pf);
1214 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1215 &dataoff, &protonum);
1217 pr_debug("not prepared to track yet or error occurred\n");
1218 NF_CT_STAT_INC_ATOMIC(net, error);
1219 NF_CT_STAT_INC_ATOMIC(net, invalid);
1224 l4proto = __nf_ct_l4proto_find(pf, protonum);
1226 /* It may be an special packet, error, unclean...
1227 * inverse of the return code tells to the netfilter
1228 * core what to do with the packet. */
1229 if (l4proto->error != NULL) {
1230 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1233 NF_CT_STAT_INC_ATOMIC(net, error);
1234 NF_CT_STAT_INC_ATOMIC(net, invalid);
1238 /* ICMP[v6] protocol trackers may assign one conntrack. */
1243 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1244 l3proto, l4proto, &set_reply, &ctinfo);
1246 /* Not valid part of a connection */
1247 NF_CT_STAT_INC_ATOMIC(net, invalid);
1253 /* Too stressed to deal. */
1254 NF_CT_STAT_INC_ATOMIC(net, drop);
1259 NF_CT_ASSERT(skb->nfct);
1261 /* Decide what timeout policy we want to apply to this flow. */
1262 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1264 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1266 /* Invalid: inverse of the return code tells
1267 * the netfilter core what to do */
1268 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1269 nf_conntrack_put(skb->nfct);
1271 NF_CT_STAT_INC_ATOMIC(net, invalid);
1272 if (ret == -NF_DROP)
1273 NF_CT_STAT_INC_ATOMIC(net, drop);
1278 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1279 nf_conntrack_event_cache(IPCT_REPLY, ct);
1282 /* Special case: we have to repeat this hook, assign the
1283 * template again to this packet. We assume that this packet
1284 * has no conntrack assigned. This is used by nf_ct_tcp. */
1285 if (ret == NF_REPEAT)
1286 skb->nfct = (struct nf_conntrack *)tmpl;
1293 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1295 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1296 const struct nf_conntrack_tuple *orig)
1301 ret = nf_ct_invert_tuple(inverse, orig,
1302 __nf_ct_l3proto_find(orig->src.l3num),
1303 __nf_ct_l4proto_find(orig->src.l3num,
1304 orig->dst.protonum));
1308 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1310 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1311 implicitly racy: see __nf_conntrack_confirm */
1312 void nf_conntrack_alter_reply(struct nf_conn *ct,
1313 const struct nf_conntrack_tuple *newreply)
1315 struct nf_conn_help *help = nfct_help(ct);
1317 /* Should be unconfirmed, so not in hash table yet */
1318 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1320 pr_debug("Altering reply tuple of %p to ", ct);
1321 nf_ct_dump_tuple(newreply);
1323 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1324 if (ct->master || (help && !hlist_empty(&help->expectations)))
1328 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1331 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1333 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1334 void __nf_ct_refresh_acct(struct nf_conn *ct,
1335 enum ip_conntrack_info ctinfo,
1336 const struct sk_buff *skb,
1337 unsigned long extra_jiffies,
1342 /* Only update if this is not a fixed timeout */
1343 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1346 /* If not in hash table, timer will not be active yet */
1347 if (nf_ct_is_confirmed(ct))
1348 extra_jiffies += nfct_time_stamp;
1350 ct->timeout = extra_jiffies;
1353 nf_ct_acct_update(ct, ctinfo, skb->len);
1355 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1357 bool __nf_ct_kill_acct(struct nf_conn *ct,
1358 enum ip_conntrack_info ctinfo,
1359 const struct sk_buff *skb,
1363 nf_ct_acct_update(ct, ctinfo, skb->len);
1365 return nf_ct_delete(ct, 0, 0);
1367 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1369 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1371 #include <linux/netfilter/nfnetlink.h>
1372 #include <linux/netfilter/nfnetlink_conntrack.h>
1373 #include <linux/mutex.h>
1375 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1376 * in ip_conntrack_core, since we don't want the protocols to autoload
1377 * or depend on ctnetlink */
1378 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1379 const struct nf_conntrack_tuple *tuple)
1381 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1382 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1383 goto nla_put_failure;
1389 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1391 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1392 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1393 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1395 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1397 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1398 struct nf_conntrack_tuple *t)
1400 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1403 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1404 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1408 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1410 int nf_ct_port_nlattr_tuple_size(void)
1412 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1414 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1417 /* Used by ipt_REJECT and ip6t_REJECT. */
1418 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1421 enum ip_conntrack_info ctinfo;
1423 /* This ICMP is in reverse direction to the packet which caused it */
1424 ct = nf_ct_get(skb, &ctinfo);
1425 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1426 ctinfo = IP_CT_RELATED_REPLY;
1428 ctinfo = IP_CT_RELATED;
1430 /* Attach to new skbuff, and increment count */
1431 nskb->nfct = &ct->ct_general;
1432 nskb->nfctinfo = ctinfo;
1433 nf_conntrack_get(nskb->nfct);
1436 /* Bring out ya dead! */
1437 static struct nf_conn *
1438 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1439 void *data, unsigned int *bucket)
1441 struct nf_conntrack_tuple_hash *h;
1443 struct hlist_nulls_node *n;
1447 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1448 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1450 nf_conntrack_lock(lockp);
1451 if (*bucket < nf_conntrack_htable_size) {
1452 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1453 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1455 ct = nf_ct_tuplehash_to_ctrack(h);
1456 if (net_eq(nf_ct_net(ct), net) &&
1466 for_each_possible_cpu(cpu) {
1467 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1469 spin_lock_bh(&pcpu->lock);
1470 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1471 ct = nf_ct_tuplehash_to_ctrack(h);
1473 set_bit(IPS_DYING_BIT, &ct->status);
1475 spin_unlock_bh(&pcpu->lock);
1480 atomic_inc(&ct->ct_general.use);
1486 void nf_ct_iterate_cleanup(struct net *net,
1487 int (*iter)(struct nf_conn *i, void *data),
1488 void *data, u32 portid, int report)
1491 unsigned int bucket = 0;
1495 if (atomic_read(&net->ct.count) == 0)
1498 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1499 /* Time to push up daises... */
1501 nf_ct_delete(ct, portid, report);
1506 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1508 static int kill_all(struct nf_conn *i, void *data)
1513 void nf_ct_free_hashtable(void *hash, unsigned int size)
1515 if (is_vmalloc_addr(hash))
1518 free_pages((unsigned long)hash,
1519 get_order(sizeof(struct hlist_head) * size));
1521 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1523 static int untrack_refs(void)
1527 for_each_possible_cpu(cpu) {
1528 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1530 cnt += atomic_read(&ct->ct_general.use) - 1;
1535 void nf_conntrack_cleanup_start(void)
1537 RCU_INIT_POINTER(ip_ct_attach, NULL);
1540 void nf_conntrack_cleanup_end(void)
1542 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1543 while (untrack_refs() > 0)
1546 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1548 nf_conntrack_proto_fini();
1549 nf_conntrack_seqadj_fini();
1550 nf_conntrack_labels_fini();
1551 nf_conntrack_helper_fini();
1552 nf_conntrack_timeout_fini();
1553 nf_conntrack_ecache_fini();
1554 nf_conntrack_tstamp_fini();
1555 nf_conntrack_acct_fini();
1556 nf_conntrack_expect_fini();
1558 kmem_cache_destroy(nf_conntrack_cachep);
1562 * Mishearing the voices in his head, our hero wonders how he's
1563 * supposed to kill the mall.
1565 void nf_conntrack_cleanup_net(struct net *net)
1569 list_add(&net->exit_list, &single);
1570 nf_conntrack_cleanup_net_list(&single);
1573 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1579 * This makes sure all current packets have passed through
1580 * netfilter framework. Roll on, two-stage module
1586 list_for_each_entry(net, net_exit_list, exit_list) {
1587 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1588 if (atomic_read(&net->ct.count) != 0)
1593 goto i_see_dead_people;
1596 list_for_each_entry(net, net_exit_list, exit_list) {
1597 nf_conntrack_proto_pernet_fini(net);
1598 nf_conntrack_helper_pernet_fini(net);
1599 nf_conntrack_ecache_pernet_fini(net);
1600 nf_conntrack_tstamp_pernet_fini(net);
1601 nf_conntrack_acct_pernet_fini(net);
1602 nf_conntrack_expect_pernet_fini(net);
1603 free_percpu(net->ct.stat);
1604 free_percpu(net->ct.pcpu_lists);
1608 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1610 struct hlist_nulls_head *hash;
1611 unsigned int nr_slots, i;
1614 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1617 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1618 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1620 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1623 sz = nr_slots * sizeof(struct hlist_nulls_head);
1624 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1630 for (i = 0; i < nr_slots; i++)
1631 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1635 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1637 int nf_conntrack_hash_resize(unsigned int hashsize)
1640 unsigned int old_size;
1641 struct hlist_nulls_head *hash, *old_hash;
1642 struct nf_conntrack_tuple_hash *h;
1648 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1652 old_size = nf_conntrack_htable_size;
1653 if (old_size == hashsize) {
1654 nf_ct_free_hashtable(hash, hashsize);
1659 nf_conntrack_all_lock();
1660 write_seqcount_begin(&nf_conntrack_generation);
1662 /* Lookups in the old hash might happen in parallel, which means we
1663 * might get false negatives during connection lookup. New connections
1664 * created because of a false negative won't make it into the hash
1665 * though since that required taking the locks.
1668 for (i = 0; i < nf_conntrack_htable_size; i++) {
1669 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1670 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1671 struct nf_conntrack_tuple_hash, hnnode);
1672 ct = nf_ct_tuplehash_to_ctrack(h);
1673 hlist_nulls_del_rcu(&h->hnnode);
1674 bucket = __hash_conntrack(nf_ct_net(ct),
1675 &h->tuple, hashsize);
1676 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1679 old_size = nf_conntrack_htable_size;
1680 old_hash = nf_conntrack_hash;
1682 nf_conntrack_hash = hash;
1683 nf_conntrack_htable_size = hashsize;
1685 write_seqcount_end(&nf_conntrack_generation);
1686 nf_conntrack_all_unlock();
1690 nf_ct_free_hashtable(old_hash, old_size);
1694 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1696 unsigned int hashsize;
1699 if (current->nsproxy->net_ns != &init_net)
1702 /* On boot, we can set this without any fancy locking. */
1703 if (!nf_conntrack_htable_size)
1704 return param_set_uint(val, kp);
1706 rc = kstrtouint(val, 0, &hashsize);
1710 return nf_conntrack_hash_resize(hashsize);
1712 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1714 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1715 &nf_conntrack_htable_size, 0600);
1717 void nf_ct_untracked_status_or(unsigned long bits)
1721 for_each_possible_cpu(cpu)
1722 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1724 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1726 int nf_conntrack_init_start(void)
1732 seqcount_init(&nf_conntrack_generation);
1734 for (i = 0; i < CONNTRACK_LOCKS; i++)
1735 spin_lock_init(&nf_conntrack_locks[i]);
1737 if (!nf_conntrack_htable_size) {
1738 /* Idea from tcp.c: use 1/16384 of memory.
1739 * On i386: 32MB machine has 512 buckets.
1740 * >= 1GB machines have 16384 buckets.
1741 * >= 4GB machines have 65536 buckets.
1743 nf_conntrack_htable_size
1744 = (((totalram_pages << PAGE_SHIFT) / 16384)
1745 / sizeof(struct hlist_head));
1746 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1747 nf_conntrack_htable_size = 65536;
1748 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1749 nf_conntrack_htable_size = 16384;
1750 if (nf_conntrack_htable_size < 32)
1751 nf_conntrack_htable_size = 32;
1753 /* Use a max. factor of four by default to get the same max as
1754 * with the old struct list_heads. When a table size is given
1755 * we use the old value of 8 to avoid reducing the max.
1760 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1761 if (!nf_conntrack_hash)
1764 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1766 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1767 sizeof(struct nf_conn), 0,
1768 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1769 if (!nf_conntrack_cachep)
1772 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1773 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1776 ret = nf_conntrack_expect_init();
1780 ret = nf_conntrack_acct_init();
1784 ret = nf_conntrack_tstamp_init();
1788 ret = nf_conntrack_ecache_init();
1792 ret = nf_conntrack_timeout_init();
1796 ret = nf_conntrack_helper_init();
1800 ret = nf_conntrack_labels_init();
1804 ret = nf_conntrack_seqadj_init();
1808 ret = nf_conntrack_proto_init();
1812 /* Set up fake conntrack: to never be deleted, not in any hashes */
1813 for_each_possible_cpu(cpu) {
1814 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1815 write_pnet(&ct->ct_net, &init_net);
1816 atomic_set(&ct->ct_general.use, 1);
1818 /* - and look it like as a confirmed connection */
1819 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1823 nf_conntrack_seqadj_fini();
1825 nf_conntrack_labels_fini();
1827 nf_conntrack_helper_fini();
1829 nf_conntrack_timeout_fini();
1831 nf_conntrack_ecache_fini();
1833 nf_conntrack_tstamp_fini();
1835 nf_conntrack_acct_fini();
1837 nf_conntrack_expect_fini();
1839 kmem_cache_destroy(nf_conntrack_cachep);
1841 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1845 void nf_conntrack_init_end(void)
1847 /* For use by REJECT target */
1848 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1849 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1853 * We need to use special "null" values, not used in hash table
1855 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1856 #define DYING_NULLS_VAL ((1<<30)+1)
1857 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1859 int nf_conntrack_init_net(struct net *net)
1864 atomic_set(&net->ct.count, 0);
1866 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1867 if (!net->ct.pcpu_lists)
1870 for_each_possible_cpu(cpu) {
1871 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1873 spin_lock_init(&pcpu->lock);
1874 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1875 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1878 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1880 goto err_pcpu_lists;
1882 ret = nf_conntrack_expect_pernet_init(net);
1885 ret = nf_conntrack_acct_pernet_init(net);
1888 ret = nf_conntrack_tstamp_pernet_init(net);
1891 ret = nf_conntrack_ecache_pernet_init(net);
1894 ret = nf_conntrack_helper_pernet_init(net);
1897 ret = nf_conntrack_proto_pernet_init(net);
1903 nf_conntrack_helper_pernet_fini(net);
1905 nf_conntrack_ecache_pernet_fini(net);
1907 nf_conntrack_tstamp_pernet_fini(net);
1909 nf_conntrack_acct_pernet_fini(net);
1911 nf_conntrack_expect_pernet_fini(net);
1913 free_percpu(net->ct.stat);
1915 free_percpu(net->ct.pcpu_lists);