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);
374 NF_CT_ASSERT(!timer_pending(&ct->timeout));
376 if (unlikely(nf_ct_is_template(ct))) {
381 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
382 if (l4proto->destroy)
383 l4proto->destroy(ct);
388 /* Expectations will have been removed in clean_from_lists,
389 * except TFTP can create an expectation on the first packet,
390 * before connection is in the list, so we need to clean here,
393 nf_ct_remove_expectations(ct);
395 nf_ct_del_from_dying_or_unconfirmed_list(ct);
397 NF_CT_STAT_INC(net, delete);
401 nf_ct_put(ct->master);
403 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
404 nf_conntrack_free(ct);
407 static void nf_ct_delete_from_lists(struct nf_conn *ct)
409 struct net *net = nf_ct_net(ct);
410 unsigned int hash, reply_hash;
411 unsigned int sequence;
413 nf_ct_helper_destroy(ct);
417 sequence = read_seqcount_begin(&nf_conntrack_generation);
418 hash = hash_conntrack(net,
419 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
420 reply_hash = hash_conntrack(net,
421 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
422 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
424 clean_from_lists(ct);
425 nf_conntrack_double_unlock(hash, reply_hash);
427 nf_ct_add_to_dying_list(ct);
429 NF_CT_STAT_INC(net, delete_list);
433 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
435 struct nf_conn_tstamp *tstamp;
437 tstamp = nf_conn_tstamp_find(ct);
438 if (tstamp && tstamp->stop == 0)
439 tstamp->stop = ktime_get_real_ns();
441 if (nf_ct_is_dying(ct))
444 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
445 portid, report) < 0) {
446 /* destroy event was not delivered */
447 nf_ct_delete_from_lists(ct);
448 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
452 nf_conntrack_ecache_work(nf_ct_net(ct));
453 set_bit(IPS_DYING_BIT, &ct->status);
455 nf_ct_delete_from_lists(ct);
459 EXPORT_SYMBOL_GPL(nf_ct_delete);
461 static void death_by_timeout(unsigned long ul_conntrack)
463 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
467 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
468 const struct nf_conntrack_tuple *tuple,
469 const struct nf_conntrack_zone *zone,
470 const struct net *net)
472 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
474 /* A conntrack can be recreated with the equal tuple,
475 * so we need to check that the conntrack is confirmed
477 return nf_ct_tuple_equal(tuple, &h->tuple) &&
478 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
479 nf_ct_is_confirmed(ct) &&
480 net_eq(net, nf_ct_net(ct));
485 * - Caller must take a reference on returned object
486 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
488 static struct nf_conntrack_tuple_hash *
489 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
490 const struct nf_conntrack_tuple *tuple, u32 hash)
492 struct nf_conntrack_tuple_hash *h;
493 struct hlist_nulls_head *ct_hash;
494 struct hlist_nulls_node *n;
495 unsigned int bucket, hsize;
498 nf_conntrack_get_ht(&ct_hash, &hsize);
499 bucket = reciprocal_scale(hash, hsize);
501 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
502 if (nf_ct_key_equal(h, tuple, zone, net)) {
503 NF_CT_STAT_INC_ATOMIC(net, found);
506 NF_CT_STAT_INC_ATOMIC(net, searched);
509 * if the nulls value we got at the end of this lookup is
510 * not the expected one, we must restart lookup.
511 * We probably met an item that was moved to another chain.
513 if (get_nulls_value(n) != bucket) {
514 NF_CT_STAT_INC_ATOMIC(net, search_restart);
521 /* Find a connection corresponding to a tuple. */
522 static struct nf_conntrack_tuple_hash *
523 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
524 const struct nf_conntrack_tuple *tuple, u32 hash)
526 struct nf_conntrack_tuple_hash *h;
531 h = ____nf_conntrack_find(net, zone, tuple, hash);
533 ct = nf_ct_tuplehash_to_ctrack(h);
534 if (unlikely(nf_ct_is_dying(ct) ||
535 !atomic_inc_not_zero(&ct->ct_general.use)))
538 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
549 struct nf_conntrack_tuple_hash *
550 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
551 const struct nf_conntrack_tuple *tuple)
553 return __nf_conntrack_find_get(net, zone, tuple,
554 hash_conntrack_raw(tuple, net));
556 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
558 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
560 unsigned int reply_hash)
562 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
563 &nf_conntrack_hash[hash]);
564 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
565 &nf_conntrack_hash[reply_hash]);
569 nf_conntrack_hash_check_insert(struct nf_conn *ct)
571 const struct nf_conntrack_zone *zone;
572 struct net *net = nf_ct_net(ct);
573 unsigned int hash, reply_hash;
574 struct nf_conntrack_tuple_hash *h;
575 struct hlist_nulls_node *n;
576 unsigned int sequence;
578 zone = nf_ct_zone(ct);
582 sequence = read_seqcount_begin(&nf_conntrack_generation);
583 hash = hash_conntrack(net,
584 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
585 reply_hash = hash_conntrack(net,
586 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
587 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
589 /* See if there's one in the list already, including reverse */
590 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
591 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
595 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
596 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
600 add_timer(&ct->timeout);
602 /* The caller holds a reference to this object */
603 atomic_set(&ct->ct_general.use, 2);
604 __nf_conntrack_hash_insert(ct, hash, reply_hash);
605 nf_conntrack_double_unlock(hash, reply_hash);
606 NF_CT_STAT_INC(net, insert);
611 nf_conntrack_double_unlock(hash, reply_hash);
612 NF_CT_STAT_INC(net, insert_failed);
616 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
618 static inline void nf_ct_acct_update(struct nf_conn *ct,
619 enum ip_conntrack_info ctinfo,
622 struct nf_conn_acct *acct;
624 acct = nf_conn_acct_find(ct);
626 struct nf_conn_counter *counter = acct->counter;
628 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
629 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
633 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
634 const struct nf_conn *loser_ct)
636 struct nf_conn_acct *acct;
638 acct = nf_conn_acct_find(loser_ct);
640 struct nf_conn_counter *counter = acct->counter;
643 /* u32 should be fine since we must have seen one packet. */
644 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
645 nf_ct_acct_update(ct, ctinfo, bytes);
649 /* Resolve race on insertion if this protocol allows this. */
650 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
651 enum ip_conntrack_info ctinfo,
652 struct nf_conntrack_tuple_hash *h)
654 /* This is the conntrack entry already in hashes that won race. */
655 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
656 struct nf_conntrack_l4proto *l4proto;
658 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
659 if (l4proto->allow_clash &&
661 !nf_ct_is_dying(ct) &&
662 atomic_inc_not_zero(&ct->ct_general.use)) {
663 nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct);
664 nf_conntrack_put(skb->nfct);
665 /* Assign conntrack already in hashes to this skbuff. Don't
666 * modify skb->nfctinfo to ensure consistent stateful filtering.
668 skb->nfct = &ct->ct_general;
671 NF_CT_STAT_INC(net, drop);
675 /* Confirm a connection given skb; places it in hash table */
677 __nf_conntrack_confirm(struct sk_buff *skb)
679 const struct nf_conntrack_zone *zone;
680 unsigned int hash, reply_hash;
681 struct nf_conntrack_tuple_hash *h;
683 struct nf_conn_help *help;
684 struct nf_conn_tstamp *tstamp;
685 struct hlist_nulls_node *n;
686 enum ip_conntrack_info ctinfo;
688 unsigned int sequence;
691 ct = nf_ct_get(skb, &ctinfo);
694 /* ipt_REJECT uses nf_conntrack_attach to attach related
695 ICMP/TCP RST packets in other direction. Actual packet
696 which created connection will be IP_CT_NEW or for an
697 expected connection, IP_CT_RELATED. */
698 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
701 zone = nf_ct_zone(ct);
705 sequence = read_seqcount_begin(&nf_conntrack_generation);
706 /* reuse the hash saved before */
707 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
708 hash = scale_hash(hash);
709 reply_hash = hash_conntrack(net,
710 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
712 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
714 /* We're not in hash table, and we refuse to set up related
715 * connections for unconfirmed conns. But packet copies and
716 * REJECT will give spurious warnings here.
718 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
720 /* No external references means no one else could have
723 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
724 pr_debug("Confirming conntrack %p\n", ct);
725 /* We have to check the DYING flag after unlink to prevent
726 * a race against nf_ct_get_next_corpse() possibly called from
727 * user context, else we insert an already 'dead' hash, blocking
728 * further use of that particular connection -JM.
730 nf_ct_del_from_dying_or_unconfirmed_list(ct);
732 if (unlikely(nf_ct_is_dying(ct))) {
733 nf_ct_add_to_dying_list(ct);
737 /* See if there's one in the list already, including reverse:
738 NAT could have grabbed it without realizing, since we're
739 not in the hash. If there is, we lost race. */
740 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
741 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
745 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
746 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
750 /* Timer relative to confirmation time, not original
751 setting time, otherwise we'd get timer wrap in
752 weird delay cases. */
753 ct->timeout.expires += jiffies;
754 add_timer(&ct->timeout);
755 atomic_inc(&ct->ct_general.use);
756 ct->status |= IPS_CONFIRMED;
758 /* set conntrack timestamp, if enabled. */
759 tstamp = nf_conn_tstamp_find(ct);
761 if (skb->tstamp.tv64 == 0)
762 __net_timestamp(skb);
764 tstamp->start = ktime_to_ns(skb->tstamp);
766 /* Since the lookup is lockless, hash insertion must be done after
767 * starting the timer and setting the CONFIRMED bit. The RCU barriers
768 * guarantee that no other CPU can find the conntrack before the above
769 * stores are visible.
771 __nf_conntrack_hash_insert(ct, hash, reply_hash);
772 nf_conntrack_double_unlock(hash, reply_hash);
773 NF_CT_STAT_INC(net, insert);
776 help = nfct_help(ct);
777 if (help && help->helper)
778 nf_conntrack_event_cache(IPCT_HELPER, ct);
780 nf_conntrack_event_cache(master_ct(ct) ?
781 IPCT_RELATED : IPCT_NEW, ct);
785 nf_ct_add_to_dying_list(ct);
786 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
788 nf_conntrack_double_unlock(hash, reply_hash);
789 NF_CT_STAT_INC(net, insert_failed);
793 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
795 /* Returns true if a connection correspondings to the tuple (required
798 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
799 const struct nf_conn *ignored_conntrack)
801 struct net *net = nf_ct_net(ignored_conntrack);
802 const struct nf_conntrack_zone *zone;
803 struct nf_conntrack_tuple_hash *h;
804 struct hlist_nulls_head *ct_hash;
805 unsigned int hash, hsize;
806 struct hlist_nulls_node *n;
809 zone = nf_ct_zone(ignored_conntrack);
812 nf_conntrack_get_ht(&ct_hash, &hsize);
813 hash = __hash_conntrack(net, tuple, hsize);
815 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
816 ct = nf_ct_tuplehash_to_ctrack(h);
817 if (ct != ignored_conntrack &&
818 nf_ct_key_equal(h, tuple, zone, net)) {
819 NF_CT_STAT_INC_ATOMIC(net, found);
823 NF_CT_STAT_INC_ATOMIC(net, searched);
829 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
831 #define NF_CT_EVICTION_RANGE 8
833 /* There's a small race here where we may free a just-assured
834 connection. Too bad: we're in trouble anyway. */
835 static unsigned int early_drop_list(struct net *net,
836 struct hlist_nulls_head *head)
838 struct nf_conntrack_tuple_hash *h;
839 struct hlist_nulls_node *n;
840 unsigned int drops = 0;
843 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
844 tmp = nf_ct_tuplehash_to_ctrack(h);
846 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
847 !net_eq(nf_ct_net(tmp), net) ||
851 if (!atomic_inc_not_zero(&tmp->ct_general.use))
854 /* kill only if still in same netns -- might have moved due to
855 * SLAB_DESTROY_BY_RCU rules.
857 * We steal the timer reference. If that fails timer has
858 * already fired or someone else deleted it. Just drop ref
859 * and move to next entry.
861 if (net_eq(nf_ct_net(tmp), net) &&
862 nf_ct_is_confirmed(tmp) &&
863 del_timer(&tmp->timeout) &&
864 nf_ct_delete(tmp, 0, 0))
873 static noinline int early_drop(struct net *net, unsigned int _hash)
877 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
878 struct hlist_nulls_head *ct_hash;
879 unsigned int hash, hsize, drops;
882 nf_conntrack_get_ht(&ct_hash, &hsize);
883 hash = reciprocal_scale(_hash++, hsize);
885 drops = early_drop_list(net, &ct_hash[hash]);
889 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
897 static struct nf_conn *
898 __nf_conntrack_alloc(struct net *net,
899 const struct nf_conntrack_zone *zone,
900 const struct nf_conntrack_tuple *orig,
901 const struct nf_conntrack_tuple *repl,
906 /* We don't want any race condition at early drop stage */
907 atomic_inc(&net->ct.count);
909 if (nf_conntrack_max &&
910 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
911 if (!early_drop(net, hash)) {
912 atomic_dec(&net->ct.count);
913 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
914 return ERR_PTR(-ENOMEM);
919 * Do not use kmem_cache_zalloc(), as this cache uses
920 * SLAB_DESTROY_BY_RCU.
922 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
926 spin_lock_init(&ct->lock);
927 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
928 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
929 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
930 /* save hash for reusing when confirming */
931 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
933 /* Don't set timer yet: wait for confirmation */
934 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
935 write_pnet(&ct->ct_net, net);
936 memset(&ct->__nfct_init_offset[0], 0,
937 offsetof(struct nf_conn, proto) -
938 offsetof(struct nf_conn, __nfct_init_offset[0]));
940 nf_ct_zone_add(ct, zone);
942 /* Because we use RCU lookups, we set ct_general.use to zero before
943 * this is inserted in any list.
945 atomic_set(&ct->ct_general.use, 0);
948 atomic_dec(&net->ct.count);
949 return ERR_PTR(-ENOMEM);
952 struct nf_conn *nf_conntrack_alloc(struct net *net,
953 const struct nf_conntrack_zone *zone,
954 const struct nf_conntrack_tuple *orig,
955 const struct nf_conntrack_tuple *repl,
958 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
960 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
962 void nf_conntrack_free(struct nf_conn *ct)
964 struct net *net = nf_ct_net(ct);
966 /* A freed object has refcnt == 0, that's
967 * the golden rule for SLAB_DESTROY_BY_RCU
969 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
971 nf_ct_ext_destroy(ct);
973 kmem_cache_free(nf_conntrack_cachep, ct);
974 smp_mb__before_atomic();
975 atomic_dec(&net->ct.count);
977 EXPORT_SYMBOL_GPL(nf_conntrack_free);
980 /* Allocate a new conntrack: we return -ENOMEM if classification
981 failed due to stress. Otherwise it really is unclassifiable. */
982 static struct nf_conntrack_tuple_hash *
983 init_conntrack(struct net *net, struct nf_conn *tmpl,
984 const struct nf_conntrack_tuple *tuple,
985 struct nf_conntrack_l3proto *l3proto,
986 struct nf_conntrack_l4proto *l4proto,
988 unsigned int dataoff, u32 hash)
991 struct nf_conn_help *help;
992 struct nf_conntrack_tuple repl_tuple;
993 struct nf_conntrack_ecache *ecache;
994 struct nf_conntrack_expect *exp = NULL;
995 const struct nf_conntrack_zone *zone;
996 struct nf_conn_timeout *timeout_ext;
997 struct nf_conntrack_zone tmp;
998 unsigned int *timeouts;
1000 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1001 pr_debug("Can't invert tuple.\n");
1005 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1006 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1009 return (struct nf_conntrack_tuple_hash *)ct;
1011 if (tmpl && nfct_synproxy(tmpl)) {
1012 nfct_seqadj_ext_add(ct);
1013 nfct_synproxy_ext_add(ct);
1016 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1018 timeouts = nf_ct_timeout_data(timeout_ext);
1019 if (unlikely(!timeouts))
1020 timeouts = l4proto->get_timeouts(net);
1022 timeouts = l4proto->get_timeouts(net);
1025 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1026 nf_conntrack_free(ct);
1027 pr_debug("can't track with proto module\n");
1032 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1035 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1036 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1037 nf_ct_labels_ext_add(ct);
1039 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1040 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1041 ecache ? ecache->expmask : 0,
1045 if (net->ct.expect_count) {
1046 spin_lock(&nf_conntrack_expect_lock);
1047 exp = nf_ct_find_expectation(net, zone, tuple);
1049 pr_debug("expectation arrives ct=%p exp=%p\n",
1051 /* Welcome, Mr. Bond. We've been expecting you... */
1052 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1053 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1054 ct->master = exp->master;
1056 help = nf_ct_helper_ext_add(ct, exp->helper,
1059 rcu_assign_pointer(help->helper, exp->helper);
1062 #ifdef CONFIG_NF_CONNTRACK_MARK
1063 ct->mark = exp->master->mark;
1065 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1066 ct->secmark = exp->master->secmark;
1068 NF_CT_STAT_INC(net, expect_new);
1070 spin_unlock(&nf_conntrack_expect_lock);
1073 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1074 NF_CT_STAT_INC(net, new);
1077 /* Now it is inserted into the unconfirmed list, bump refcount */
1078 nf_conntrack_get(&ct->ct_general);
1079 nf_ct_add_to_unconfirmed_list(ct);
1085 exp->expectfn(ct, exp);
1086 nf_ct_expect_put(exp);
1089 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1092 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1093 static inline struct nf_conn *
1094 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1095 struct sk_buff *skb,
1096 unsigned int dataoff,
1099 struct nf_conntrack_l3proto *l3proto,
1100 struct nf_conntrack_l4proto *l4proto,
1102 enum ip_conntrack_info *ctinfo)
1104 const struct nf_conntrack_zone *zone;
1105 struct nf_conntrack_tuple tuple;
1106 struct nf_conntrack_tuple_hash *h;
1107 struct nf_conntrack_zone tmp;
1111 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1112 dataoff, l3num, protonum, net, &tuple, l3proto,
1114 pr_debug("Can't get tuple\n");
1118 /* look for tuple match */
1119 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1120 hash = hash_conntrack_raw(&tuple, net);
1121 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1123 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1124 skb, dataoff, hash);
1130 ct = nf_ct_tuplehash_to_ctrack(h);
1132 /* It exists; we have (non-exclusive) reference. */
1133 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1134 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1135 /* Please set reply bit if this packet OK */
1138 /* Once we've had two way comms, always ESTABLISHED. */
1139 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1140 pr_debug("normal packet for %p\n", ct);
1141 *ctinfo = IP_CT_ESTABLISHED;
1142 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1143 pr_debug("related packet for %p\n", ct);
1144 *ctinfo = IP_CT_RELATED;
1146 pr_debug("new packet for %p\n", ct);
1147 *ctinfo = IP_CT_NEW;
1151 skb->nfct = &ct->ct_general;
1152 skb->nfctinfo = *ctinfo;
1157 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1158 struct sk_buff *skb)
1160 struct nf_conn *ct, *tmpl = NULL;
1161 enum ip_conntrack_info ctinfo;
1162 struct nf_conntrack_l3proto *l3proto;
1163 struct nf_conntrack_l4proto *l4proto;
1164 unsigned int *timeouts;
1165 unsigned int dataoff;
1171 /* Previously seen (loopback or untracked)? Ignore. */
1172 tmpl = (struct nf_conn *)skb->nfct;
1173 if (!nf_ct_is_template(tmpl)) {
1174 NF_CT_STAT_INC_ATOMIC(net, ignore);
1180 /* rcu_read_lock()ed by nf_hook_slow */
1181 l3proto = __nf_ct_l3proto_find(pf);
1182 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1183 &dataoff, &protonum);
1185 pr_debug("not prepared to track yet or error occurred\n");
1186 NF_CT_STAT_INC_ATOMIC(net, error);
1187 NF_CT_STAT_INC_ATOMIC(net, invalid);
1192 l4proto = __nf_ct_l4proto_find(pf, protonum);
1194 /* It may be an special packet, error, unclean...
1195 * inverse of the return code tells to the netfilter
1196 * core what to do with the packet. */
1197 if (l4proto->error != NULL) {
1198 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1201 NF_CT_STAT_INC_ATOMIC(net, error);
1202 NF_CT_STAT_INC_ATOMIC(net, invalid);
1206 /* ICMP[v6] protocol trackers may assign one conntrack. */
1211 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1212 l3proto, l4proto, &set_reply, &ctinfo);
1214 /* Not valid part of a connection */
1215 NF_CT_STAT_INC_ATOMIC(net, invalid);
1221 /* Too stressed to deal. */
1222 NF_CT_STAT_INC_ATOMIC(net, drop);
1227 NF_CT_ASSERT(skb->nfct);
1229 /* Decide what timeout policy we want to apply to this flow. */
1230 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1232 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1234 /* Invalid: inverse of the return code tells
1235 * the netfilter core what to do */
1236 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1237 nf_conntrack_put(skb->nfct);
1239 NF_CT_STAT_INC_ATOMIC(net, invalid);
1240 if (ret == -NF_DROP)
1241 NF_CT_STAT_INC_ATOMIC(net, drop);
1246 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1247 nf_conntrack_event_cache(IPCT_REPLY, ct);
1250 /* Special case: we have to repeat this hook, assign the
1251 * template again to this packet. We assume that this packet
1252 * has no conntrack assigned. This is used by nf_ct_tcp. */
1253 if (ret == NF_REPEAT)
1254 skb->nfct = (struct nf_conntrack *)tmpl;
1261 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1263 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1264 const struct nf_conntrack_tuple *orig)
1269 ret = nf_ct_invert_tuple(inverse, orig,
1270 __nf_ct_l3proto_find(orig->src.l3num),
1271 __nf_ct_l4proto_find(orig->src.l3num,
1272 orig->dst.protonum));
1276 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1278 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1279 implicitly racy: see __nf_conntrack_confirm */
1280 void nf_conntrack_alter_reply(struct nf_conn *ct,
1281 const struct nf_conntrack_tuple *newreply)
1283 struct nf_conn_help *help = nfct_help(ct);
1285 /* Should be unconfirmed, so not in hash table yet */
1286 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1288 pr_debug("Altering reply tuple of %p to ", ct);
1289 nf_ct_dump_tuple(newreply);
1291 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1292 if (ct->master || (help && !hlist_empty(&help->expectations)))
1296 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1299 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1301 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1302 void __nf_ct_refresh_acct(struct nf_conn *ct,
1303 enum ip_conntrack_info ctinfo,
1304 const struct sk_buff *skb,
1305 unsigned long extra_jiffies,
1308 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1311 /* Only update if this is not a fixed timeout */
1312 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1315 /* If not in hash table, timer will not be active yet */
1316 if (!nf_ct_is_confirmed(ct)) {
1317 ct->timeout.expires = extra_jiffies;
1319 unsigned long newtime = jiffies + extra_jiffies;
1321 /* Only update the timeout if the new timeout is at least
1322 HZ jiffies from the old timeout. Need del_timer for race
1323 avoidance (may already be dying). */
1324 if (newtime - ct->timeout.expires >= HZ)
1325 mod_timer_pending(&ct->timeout, newtime);
1330 nf_ct_acct_update(ct, ctinfo, skb->len);
1332 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1334 bool __nf_ct_kill_acct(struct nf_conn *ct,
1335 enum ip_conntrack_info ctinfo,
1336 const struct sk_buff *skb,
1340 nf_ct_acct_update(ct, ctinfo, skb->len);
1342 if (del_timer(&ct->timeout)) {
1343 ct->timeout.function((unsigned long)ct);
1348 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1350 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1352 #include <linux/netfilter/nfnetlink.h>
1353 #include <linux/netfilter/nfnetlink_conntrack.h>
1354 #include <linux/mutex.h>
1356 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1357 * in ip_conntrack_core, since we don't want the protocols to autoload
1358 * or depend on ctnetlink */
1359 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1360 const struct nf_conntrack_tuple *tuple)
1362 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1363 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1364 goto nla_put_failure;
1370 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1372 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1373 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1374 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1376 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1378 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1379 struct nf_conntrack_tuple *t)
1381 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1384 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1385 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1389 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1391 int nf_ct_port_nlattr_tuple_size(void)
1393 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1395 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1398 /* Used by ipt_REJECT and ip6t_REJECT. */
1399 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1402 enum ip_conntrack_info ctinfo;
1404 /* This ICMP is in reverse direction to the packet which caused it */
1405 ct = nf_ct_get(skb, &ctinfo);
1406 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1407 ctinfo = IP_CT_RELATED_REPLY;
1409 ctinfo = IP_CT_RELATED;
1411 /* Attach to new skbuff, and increment count */
1412 nskb->nfct = &ct->ct_general;
1413 nskb->nfctinfo = ctinfo;
1414 nf_conntrack_get(nskb->nfct);
1417 /* Bring out ya dead! */
1418 static struct nf_conn *
1419 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1420 void *data, unsigned int *bucket)
1422 struct nf_conntrack_tuple_hash *h;
1424 struct hlist_nulls_node *n;
1428 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1429 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1431 nf_conntrack_lock(lockp);
1432 if (*bucket < nf_conntrack_htable_size) {
1433 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1434 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1436 ct = nf_ct_tuplehash_to_ctrack(h);
1437 if (net_eq(nf_ct_net(ct), net) &&
1447 for_each_possible_cpu(cpu) {
1448 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1450 spin_lock_bh(&pcpu->lock);
1451 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1452 ct = nf_ct_tuplehash_to_ctrack(h);
1454 set_bit(IPS_DYING_BIT, &ct->status);
1456 spin_unlock_bh(&pcpu->lock);
1461 atomic_inc(&ct->ct_general.use);
1467 void nf_ct_iterate_cleanup(struct net *net,
1468 int (*iter)(struct nf_conn *i, void *data),
1469 void *data, u32 portid, int report)
1472 unsigned int bucket = 0;
1476 if (atomic_read(&net->ct.count) == 0)
1479 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1480 /* Time to push up daises... */
1481 if (del_timer(&ct->timeout))
1482 nf_ct_delete(ct, portid, report);
1484 /* ... else the timer will get him soon. */
1490 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1492 static int kill_all(struct nf_conn *i, void *data)
1497 void nf_ct_free_hashtable(void *hash, unsigned int size)
1499 if (is_vmalloc_addr(hash))
1502 free_pages((unsigned long)hash,
1503 get_order(sizeof(struct hlist_head) * size));
1505 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1507 static int untrack_refs(void)
1511 for_each_possible_cpu(cpu) {
1512 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1514 cnt += atomic_read(&ct->ct_general.use) - 1;
1519 void nf_conntrack_cleanup_start(void)
1521 RCU_INIT_POINTER(ip_ct_attach, NULL);
1524 void nf_conntrack_cleanup_end(void)
1526 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1527 while (untrack_refs() > 0)
1530 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1532 nf_conntrack_proto_fini();
1533 nf_conntrack_seqadj_fini();
1534 nf_conntrack_labels_fini();
1535 nf_conntrack_helper_fini();
1536 nf_conntrack_timeout_fini();
1537 nf_conntrack_ecache_fini();
1538 nf_conntrack_tstamp_fini();
1539 nf_conntrack_acct_fini();
1540 nf_conntrack_expect_fini();
1542 kmem_cache_destroy(nf_conntrack_cachep);
1546 * Mishearing the voices in his head, our hero wonders how he's
1547 * supposed to kill the mall.
1549 void nf_conntrack_cleanup_net(struct net *net)
1553 list_add(&net->exit_list, &single);
1554 nf_conntrack_cleanup_net_list(&single);
1557 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1563 * This makes sure all current packets have passed through
1564 * netfilter framework. Roll on, two-stage module
1570 list_for_each_entry(net, net_exit_list, exit_list) {
1571 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1572 if (atomic_read(&net->ct.count) != 0)
1577 goto i_see_dead_people;
1580 list_for_each_entry(net, net_exit_list, exit_list) {
1581 nf_conntrack_proto_pernet_fini(net);
1582 nf_conntrack_helper_pernet_fini(net);
1583 nf_conntrack_ecache_pernet_fini(net);
1584 nf_conntrack_tstamp_pernet_fini(net);
1585 nf_conntrack_acct_pernet_fini(net);
1586 nf_conntrack_expect_pernet_fini(net);
1587 free_percpu(net->ct.stat);
1588 free_percpu(net->ct.pcpu_lists);
1592 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1594 struct hlist_nulls_head *hash;
1595 unsigned int nr_slots, i;
1598 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1601 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1602 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1604 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1607 sz = nr_slots * sizeof(struct hlist_nulls_head);
1608 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1614 for (i = 0; i < nr_slots; i++)
1615 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1619 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1621 int nf_conntrack_hash_resize(unsigned int hashsize)
1624 unsigned int old_size;
1625 struct hlist_nulls_head *hash, *old_hash;
1626 struct nf_conntrack_tuple_hash *h;
1632 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1636 old_size = nf_conntrack_htable_size;
1637 if (old_size == hashsize) {
1638 nf_ct_free_hashtable(hash, hashsize);
1643 nf_conntrack_all_lock();
1644 write_seqcount_begin(&nf_conntrack_generation);
1646 /* Lookups in the old hash might happen in parallel, which means we
1647 * might get false negatives during connection lookup. New connections
1648 * created because of a false negative won't make it into the hash
1649 * though since that required taking the locks.
1652 for (i = 0; i < nf_conntrack_htable_size; i++) {
1653 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1654 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1655 struct nf_conntrack_tuple_hash, hnnode);
1656 ct = nf_ct_tuplehash_to_ctrack(h);
1657 hlist_nulls_del_rcu(&h->hnnode);
1658 bucket = __hash_conntrack(nf_ct_net(ct),
1659 &h->tuple, hashsize);
1660 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1663 old_size = nf_conntrack_htable_size;
1664 old_hash = nf_conntrack_hash;
1666 nf_conntrack_hash = hash;
1667 nf_conntrack_htable_size = hashsize;
1669 write_seqcount_end(&nf_conntrack_generation);
1670 nf_conntrack_all_unlock();
1674 nf_ct_free_hashtable(old_hash, old_size);
1678 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1680 unsigned int hashsize;
1683 if (current->nsproxy->net_ns != &init_net)
1686 /* On boot, we can set this without any fancy locking. */
1687 if (!nf_conntrack_htable_size)
1688 return param_set_uint(val, kp);
1690 rc = kstrtouint(val, 0, &hashsize);
1694 return nf_conntrack_hash_resize(hashsize);
1696 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1698 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1699 &nf_conntrack_htable_size, 0600);
1701 void nf_ct_untracked_status_or(unsigned long bits)
1705 for_each_possible_cpu(cpu)
1706 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1708 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1710 int nf_conntrack_init_start(void)
1716 seqcount_init(&nf_conntrack_generation);
1718 for (i = 0; i < CONNTRACK_LOCKS; i++)
1719 spin_lock_init(&nf_conntrack_locks[i]);
1721 if (!nf_conntrack_htable_size) {
1722 /* Idea from tcp.c: use 1/16384 of memory.
1723 * On i386: 32MB machine has 512 buckets.
1724 * >= 1GB machines have 16384 buckets.
1725 * >= 4GB machines have 65536 buckets.
1727 nf_conntrack_htable_size
1728 = (((totalram_pages << PAGE_SHIFT) / 16384)
1729 / sizeof(struct hlist_head));
1730 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1731 nf_conntrack_htable_size = 65536;
1732 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1733 nf_conntrack_htable_size = 16384;
1734 if (nf_conntrack_htable_size < 32)
1735 nf_conntrack_htable_size = 32;
1737 /* Use a max. factor of four by default to get the same max as
1738 * with the old struct list_heads. When a table size is given
1739 * we use the old value of 8 to avoid reducing the max.
1744 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1745 if (!nf_conntrack_hash)
1748 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1750 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1751 sizeof(struct nf_conn), 0,
1752 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1753 if (!nf_conntrack_cachep)
1756 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1757 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1760 ret = nf_conntrack_expect_init();
1764 ret = nf_conntrack_acct_init();
1768 ret = nf_conntrack_tstamp_init();
1772 ret = nf_conntrack_ecache_init();
1776 ret = nf_conntrack_timeout_init();
1780 ret = nf_conntrack_helper_init();
1784 ret = nf_conntrack_labels_init();
1788 ret = nf_conntrack_seqadj_init();
1792 ret = nf_conntrack_proto_init();
1796 /* Set up fake conntrack: to never be deleted, not in any hashes */
1797 for_each_possible_cpu(cpu) {
1798 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1799 write_pnet(&ct->ct_net, &init_net);
1800 atomic_set(&ct->ct_general.use, 1);
1802 /* - and look it like as a confirmed connection */
1803 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1807 nf_conntrack_seqadj_fini();
1809 nf_conntrack_labels_fini();
1811 nf_conntrack_helper_fini();
1813 nf_conntrack_timeout_fini();
1815 nf_conntrack_ecache_fini();
1817 nf_conntrack_tstamp_fini();
1819 nf_conntrack_acct_fini();
1821 nf_conntrack_expect_fini();
1823 kmem_cache_destroy(nf_conntrack_cachep);
1825 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1829 void nf_conntrack_init_end(void)
1831 /* For use by REJECT target */
1832 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1833 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1837 * We need to use special "null" values, not used in hash table
1839 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1840 #define DYING_NULLS_VAL ((1<<30)+1)
1841 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1843 int nf_conntrack_init_net(struct net *net)
1848 atomic_set(&net->ct.count, 0);
1850 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1851 if (!net->ct.pcpu_lists)
1854 for_each_possible_cpu(cpu) {
1855 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1857 spin_lock_init(&pcpu->lock);
1858 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1859 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1862 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1864 goto err_pcpu_lists;
1866 ret = nf_conntrack_expect_pernet_init(net);
1869 ret = nf_conntrack_acct_pernet_init(net);
1872 ret = nf_conntrack_tstamp_pernet_init(net);
1875 ret = nf_conntrack_ecache_pernet_init(net);
1878 ret = nf_conntrack_helper_pernet_init(net);
1881 ret = nf_conntrack_proto_pernet_init(net);
1887 nf_conntrack_helper_pernet_fini(net);
1889 nf_conntrack_ecache_pernet_fini(net);
1891 nf_conntrack_tstamp_pernet_fini(net);
1893 nf_conntrack_acct_pernet_fini(net);
1895 nf_conntrack_expect_pernet_fini(net);
1897 free_percpu(net->ct.stat);
1899 free_percpu(net->ct.pcpu_lists);