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 seqcount_t nf_conntrack_generation;
78 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
79 static __read_mostly bool nf_conntrack_locks_all;
81 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
84 while (unlikely(nf_conntrack_locks_all)) {
86 spin_unlock_wait(&nf_conntrack_locks_all_lock);
90 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
92 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
94 h1 %= CONNTRACK_LOCKS;
95 h2 %= CONNTRACK_LOCKS;
96 spin_unlock(&nf_conntrack_locks[h1]);
98 spin_unlock(&nf_conntrack_locks[h2]);
101 /* return true if we need to recompute hashes (in case hash table was resized) */
102 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
103 unsigned int h2, unsigned int sequence)
105 h1 %= CONNTRACK_LOCKS;
106 h2 %= CONNTRACK_LOCKS;
108 nf_conntrack_lock(&nf_conntrack_locks[h1]);
110 spin_lock_nested(&nf_conntrack_locks[h2],
111 SINGLE_DEPTH_NESTING);
113 nf_conntrack_lock(&nf_conntrack_locks[h2]);
114 spin_lock_nested(&nf_conntrack_locks[h1],
115 SINGLE_DEPTH_NESTING);
117 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
118 nf_conntrack_double_unlock(h1, h2);
124 static void nf_conntrack_all_lock(void)
128 spin_lock(&nf_conntrack_locks_all_lock);
129 nf_conntrack_locks_all = true;
131 for (i = 0; i < CONNTRACK_LOCKS; i++) {
132 spin_unlock_wait(&nf_conntrack_locks[i]);
136 static void nf_conntrack_all_unlock(void)
138 nf_conntrack_locks_all = false;
139 spin_unlock(&nf_conntrack_locks_all_lock);
142 unsigned int nf_conntrack_htable_size __read_mostly;
143 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
145 unsigned int nf_conntrack_max __read_mostly;
146 EXPORT_SYMBOL_GPL(nf_conntrack_max);
148 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
149 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
151 static unsigned int nf_conntrack_hash_rnd __read_mostly;
153 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
154 const struct net *net)
159 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
161 /* The direction must be ignored, so we hash everything up to the
162 * destination ports (which is a multiple of 4) and treat the last
163 * three bytes manually.
165 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
166 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
167 return jhash2((u32 *)tuple, n, seed ^
168 (((__force __u16)tuple->dst.u.all << 16) |
169 tuple->dst.protonum));
172 static u32 scale_hash(u32 hash)
174 return reciprocal_scale(hash, nf_conntrack_htable_size);
177 static u32 __hash_conntrack(const struct net *net,
178 const struct nf_conntrack_tuple *tuple,
181 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
184 static u32 hash_conntrack(const struct net *net,
185 const struct nf_conntrack_tuple *tuple)
187 return scale_hash(hash_conntrack_raw(tuple, net));
191 nf_ct_get_tuple(const struct sk_buff *skb,
193 unsigned int dataoff,
197 struct nf_conntrack_tuple *tuple,
198 const struct nf_conntrack_l3proto *l3proto,
199 const struct nf_conntrack_l4proto *l4proto)
201 memset(tuple, 0, sizeof(*tuple));
203 tuple->src.l3num = l3num;
204 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
207 tuple->dst.protonum = protonum;
208 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
210 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
212 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
214 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
216 struct net *net, struct nf_conntrack_tuple *tuple)
218 struct nf_conntrack_l3proto *l3proto;
219 struct nf_conntrack_l4proto *l4proto;
220 unsigned int protoff;
226 l3proto = __nf_ct_l3proto_find(l3num);
227 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
228 if (ret != NF_ACCEPT) {
233 l4proto = __nf_ct_l4proto_find(l3num, protonum);
235 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
241 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
244 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
245 const struct nf_conntrack_tuple *orig,
246 const struct nf_conntrack_l3proto *l3proto,
247 const struct nf_conntrack_l4proto *l4proto)
249 memset(inverse, 0, sizeof(*inverse));
251 inverse->src.l3num = orig->src.l3num;
252 if (l3proto->invert_tuple(inverse, orig) == 0)
255 inverse->dst.dir = !orig->dst.dir;
257 inverse->dst.protonum = orig->dst.protonum;
258 return l4proto->invert_tuple(inverse, orig);
260 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
263 clean_from_lists(struct nf_conn *ct)
265 pr_debug("clean_from_lists(%p)\n", ct);
266 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
267 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
269 /* Destroy all pending expectations */
270 nf_ct_remove_expectations(ct);
273 /* must be called with local_bh_disable */
274 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
276 struct ct_pcpu *pcpu;
278 /* add this conntrack to the (per cpu) dying list */
279 ct->cpu = smp_processor_id();
280 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
282 spin_lock(&pcpu->lock);
283 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
285 spin_unlock(&pcpu->lock);
288 /* must be called with local_bh_disable */
289 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
291 struct ct_pcpu *pcpu;
293 /* add this conntrack to the (per cpu) unconfirmed list */
294 ct->cpu = smp_processor_id();
295 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
297 spin_lock(&pcpu->lock);
298 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
300 spin_unlock(&pcpu->lock);
303 /* must be called with local_bh_disable */
304 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
306 struct ct_pcpu *pcpu;
308 /* We overload first tuple to link into unconfirmed or dying list.*/
309 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
311 spin_lock(&pcpu->lock);
312 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
313 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
314 spin_unlock(&pcpu->lock);
317 /* Released via destroy_conntrack() */
318 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
319 const struct nf_conntrack_zone *zone,
322 struct nf_conn *tmpl;
324 tmpl = kzalloc(sizeof(*tmpl), flags);
328 tmpl->status = IPS_TEMPLATE;
329 write_pnet(&tmpl->ct_net, net);
330 nf_ct_zone_add(tmpl, zone);
331 atomic_set(&tmpl->ct_general.use, 0);
335 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
337 void nf_ct_tmpl_free(struct nf_conn *tmpl)
339 nf_ct_ext_destroy(tmpl);
340 nf_ct_ext_free(tmpl);
343 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
346 destroy_conntrack(struct nf_conntrack *nfct)
348 struct nf_conn *ct = (struct nf_conn *)nfct;
349 struct net *net = nf_ct_net(ct);
350 struct nf_conntrack_l4proto *l4proto;
352 pr_debug("destroy_conntrack(%p)\n", ct);
353 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
354 NF_CT_ASSERT(!timer_pending(&ct->timeout));
356 if (unlikely(nf_ct_is_template(ct))) {
361 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
362 if (l4proto->destroy)
363 l4proto->destroy(ct);
368 /* Expectations will have been removed in clean_from_lists,
369 * except TFTP can create an expectation on the first packet,
370 * before connection is in the list, so we need to clean here,
373 nf_ct_remove_expectations(ct);
375 nf_ct_del_from_dying_or_unconfirmed_list(ct);
377 NF_CT_STAT_INC(net, delete);
381 nf_ct_put(ct->master);
383 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
384 nf_conntrack_free(ct);
387 static void nf_ct_delete_from_lists(struct nf_conn *ct)
389 struct net *net = nf_ct_net(ct);
390 unsigned int hash, reply_hash;
391 unsigned int sequence;
393 nf_ct_helper_destroy(ct);
397 sequence = read_seqcount_begin(&nf_conntrack_generation);
398 hash = hash_conntrack(net,
399 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
400 reply_hash = hash_conntrack(net,
401 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
402 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
404 clean_from_lists(ct);
405 nf_conntrack_double_unlock(hash, reply_hash);
407 nf_ct_add_to_dying_list(ct);
409 NF_CT_STAT_INC(net, delete_list);
413 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
415 struct nf_conn_tstamp *tstamp;
417 tstamp = nf_conn_tstamp_find(ct);
418 if (tstamp && tstamp->stop == 0)
419 tstamp->stop = ktime_get_real_ns();
421 if (nf_ct_is_dying(ct))
424 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
425 portid, report) < 0) {
426 /* destroy event was not delivered */
427 nf_ct_delete_from_lists(ct);
428 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
432 nf_conntrack_ecache_work(nf_ct_net(ct));
433 set_bit(IPS_DYING_BIT, &ct->status);
435 nf_ct_delete_from_lists(ct);
439 EXPORT_SYMBOL_GPL(nf_ct_delete);
441 static void death_by_timeout(unsigned long ul_conntrack)
443 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
447 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
448 const struct nf_conntrack_tuple *tuple,
449 const struct nf_conntrack_zone *zone,
450 const struct net *net)
452 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
454 /* A conntrack can be recreated with the equal tuple,
455 * so we need to check that the conntrack is confirmed
457 return nf_ct_tuple_equal(tuple, &h->tuple) &&
458 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
459 nf_ct_is_confirmed(ct) &&
460 net_eq(net, nf_ct_net(ct));
463 /* must be called with rcu read lock held */
464 void nf_conntrack_get_ht(struct hlist_nulls_head **hash, unsigned int *hsize)
466 struct hlist_nulls_head *hptr;
467 unsigned int sequence, hsz;
470 sequence = read_seqcount_begin(&nf_conntrack_generation);
471 hsz = nf_conntrack_htable_size;
472 hptr = nf_conntrack_hash;
473 } while (read_seqcount_retry(&nf_conntrack_generation, sequence));
478 EXPORT_SYMBOL_GPL(nf_conntrack_get_ht);
482 * - Caller must take a reference on returned object
483 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
485 static struct nf_conntrack_tuple_hash *
486 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
487 const struct nf_conntrack_tuple *tuple, u32 hash)
489 struct nf_conntrack_tuple_hash *h;
490 struct hlist_nulls_head *ct_hash;
491 struct hlist_nulls_node *n;
492 unsigned int bucket, sequence;
496 sequence = read_seqcount_begin(&nf_conntrack_generation);
497 bucket = scale_hash(hash);
498 ct_hash = nf_conntrack_hash;
499 } while (read_seqcount_retry(&nf_conntrack_generation, sequence));
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, sequence;
806 struct hlist_nulls_node *n;
809 zone = nf_ct_zone(ignored_conntrack);
813 sequence = read_seqcount_begin(&nf_conntrack_generation);
814 hash = hash_conntrack(net, tuple);
815 ct_hash = nf_conntrack_hash;
816 } while (read_seqcount_retry(&nf_conntrack_generation, sequence));
818 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
819 ct = nf_ct_tuplehash_to_ctrack(h);
820 if (ct != ignored_conntrack &&
821 nf_ct_key_equal(h, tuple, zone, net)) {
822 NF_CT_STAT_INC_ATOMIC(net, found);
826 NF_CT_STAT_INC_ATOMIC(net, searched);
832 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
834 #define NF_CT_EVICTION_RANGE 8
836 /* There's a small race here where we may free a just-assured
837 connection. Too bad: we're in trouble anyway. */
838 static unsigned int early_drop_list(struct net *net,
839 struct hlist_nulls_head *head)
841 struct nf_conntrack_tuple_hash *h;
842 struct hlist_nulls_node *n;
843 unsigned int drops = 0;
846 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
847 tmp = nf_ct_tuplehash_to_ctrack(h);
849 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
850 !net_eq(nf_ct_net(tmp), net) ||
854 if (!atomic_inc_not_zero(&tmp->ct_general.use))
857 /* kill only if still in same netns -- might have moved due to
858 * SLAB_DESTROY_BY_RCU rules.
860 * We steal the timer reference. If that fails timer has
861 * already fired or someone else deleted it. Just drop ref
862 * and move to next entry.
864 if (net_eq(nf_ct_net(tmp), net) &&
865 nf_ct_is_confirmed(tmp) &&
866 del_timer(&tmp->timeout) &&
867 nf_ct_delete(tmp, 0, 0))
876 static noinline int early_drop(struct net *net, unsigned int _hash)
880 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
881 struct hlist_nulls_head *ct_hash;
882 unsigned hash, sequence, drops;
886 sequence = read_seqcount_begin(&nf_conntrack_generation);
887 hash = scale_hash(_hash++);
888 ct_hash = nf_conntrack_hash;
889 } while (read_seqcount_retry(&nf_conntrack_generation, sequence));
891 drops = early_drop_list(net, &ct_hash[hash]);
895 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
903 static struct nf_conn *
904 __nf_conntrack_alloc(struct net *net,
905 const struct nf_conntrack_zone *zone,
906 const struct nf_conntrack_tuple *orig,
907 const struct nf_conntrack_tuple *repl,
912 /* We don't want any race condition at early drop stage */
913 atomic_inc(&net->ct.count);
915 if (nf_conntrack_max &&
916 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
917 if (!early_drop(net, hash)) {
918 atomic_dec(&net->ct.count);
919 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
920 return ERR_PTR(-ENOMEM);
925 * Do not use kmem_cache_zalloc(), as this cache uses
926 * SLAB_DESTROY_BY_RCU.
928 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
932 spin_lock_init(&ct->lock);
933 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
934 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
935 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
936 /* save hash for reusing when confirming */
937 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
939 /* Don't set timer yet: wait for confirmation */
940 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
941 write_pnet(&ct->ct_net, net);
942 memset(&ct->__nfct_init_offset[0], 0,
943 offsetof(struct nf_conn, proto) -
944 offsetof(struct nf_conn, __nfct_init_offset[0]));
946 nf_ct_zone_add(ct, zone);
948 /* Because we use RCU lookups, we set ct_general.use to zero before
949 * this is inserted in any list.
951 atomic_set(&ct->ct_general.use, 0);
954 atomic_dec(&net->ct.count);
955 return ERR_PTR(-ENOMEM);
958 struct nf_conn *nf_conntrack_alloc(struct net *net,
959 const struct nf_conntrack_zone *zone,
960 const struct nf_conntrack_tuple *orig,
961 const struct nf_conntrack_tuple *repl,
964 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
966 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
968 void nf_conntrack_free(struct nf_conn *ct)
970 struct net *net = nf_ct_net(ct);
972 /* A freed object has refcnt == 0, that's
973 * the golden rule for SLAB_DESTROY_BY_RCU
975 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
977 nf_ct_ext_destroy(ct);
979 kmem_cache_free(nf_conntrack_cachep, ct);
980 smp_mb__before_atomic();
981 atomic_dec(&net->ct.count);
983 EXPORT_SYMBOL_GPL(nf_conntrack_free);
986 /* Allocate a new conntrack: we return -ENOMEM if classification
987 failed due to stress. Otherwise it really is unclassifiable. */
988 static struct nf_conntrack_tuple_hash *
989 init_conntrack(struct net *net, struct nf_conn *tmpl,
990 const struct nf_conntrack_tuple *tuple,
991 struct nf_conntrack_l3proto *l3proto,
992 struct nf_conntrack_l4proto *l4proto,
994 unsigned int dataoff, u32 hash)
997 struct nf_conn_help *help;
998 struct nf_conntrack_tuple repl_tuple;
999 struct nf_conntrack_ecache *ecache;
1000 struct nf_conntrack_expect *exp = NULL;
1001 const struct nf_conntrack_zone *zone;
1002 struct nf_conn_timeout *timeout_ext;
1003 struct nf_conntrack_zone tmp;
1004 unsigned int *timeouts;
1006 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1007 pr_debug("Can't invert tuple.\n");
1011 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1012 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1015 return (struct nf_conntrack_tuple_hash *)ct;
1017 if (tmpl && nfct_synproxy(tmpl)) {
1018 nfct_seqadj_ext_add(ct);
1019 nfct_synproxy_ext_add(ct);
1022 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1024 timeouts = nf_ct_timeout_data(timeout_ext);
1025 if (unlikely(!timeouts))
1026 timeouts = l4proto->get_timeouts(net);
1028 timeouts = l4proto->get_timeouts(net);
1031 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1032 nf_conntrack_free(ct);
1033 pr_debug("can't track with proto module\n");
1038 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1041 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1042 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1043 nf_ct_labels_ext_add(ct);
1045 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1046 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1047 ecache ? ecache->expmask : 0,
1051 if (net->ct.expect_count) {
1052 spin_lock(&nf_conntrack_expect_lock);
1053 exp = nf_ct_find_expectation(net, zone, tuple);
1055 pr_debug("expectation arrives ct=%p exp=%p\n",
1057 /* Welcome, Mr. Bond. We've been expecting you... */
1058 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1059 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1060 ct->master = exp->master;
1062 help = nf_ct_helper_ext_add(ct, exp->helper,
1065 rcu_assign_pointer(help->helper, exp->helper);
1068 #ifdef CONFIG_NF_CONNTRACK_MARK
1069 ct->mark = exp->master->mark;
1071 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1072 ct->secmark = exp->master->secmark;
1074 NF_CT_STAT_INC(net, expect_new);
1076 spin_unlock(&nf_conntrack_expect_lock);
1079 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1080 NF_CT_STAT_INC(net, new);
1083 /* Now it is inserted into the unconfirmed list, bump refcount */
1084 nf_conntrack_get(&ct->ct_general);
1085 nf_ct_add_to_unconfirmed_list(ct);
1091 exp->expectfn(ct, exp);
1092 nf_ct_expect_put(exp);
1095 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1098 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1099 static inline struct nf_conn *
1100 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1101 struct sk_buff *skb,
1102 unsigned int dataoff,
1105 struct nf_conntrack_l3proto *l3proto,
1106 struct nf_conntrack_l4proto *l4proto,
1108 enum ip_conntrack_info *ctinfo)
1110 const struct nf_conntrack_zone *zone;
1111 struct nf_conntrack_tuple tuple;
1112 struct nf_conntrack_tuple_hash *h;
1113 struct nf_conntrack_zone tmp;
1117 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1118 dataoff, l3num, protonum, net, &tuple, l3proto,
1120 pr_debug("Can't get tuple\n");
1124 /* look for tuple match */
1125 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1126 hash = hash_conntrack_raw(&tuple, net);
1127 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1129 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1130 skb, dataoff, hash);
1136 ct = nf_ct_tuplehash_to_ctrack(h);
1138 /* It exists; we have (non-exclusive) reference. */
1139 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1140 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1141 /* Please set reply bit if this packet OK */
1144 /* Once we've had two way comms, always ESTABLISHED. */
1145 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1146 pr_debug("normal packet for %p\n", ct);
1147 *ctinfo = IP_CT_ESTABLISHED;
1148 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1149 pr_debug("related packet for %p\n", ct);
1150 *ctinfo = IP_CT_RELATED;
1152 pr_debug("new packet for %p\n", ct);
1153 *ctinfo = IP_CT_NEW;
1157 skb->nfct = &ct->ct_general;
1158 skb->nfctinfo = *ctinfo;
1163 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1164 struct sk_buff *skb)
1166 struct nf_conn *ct, *tmpl = NULL;
1167 enum ip_conntrack_info ctinfo;
1168 struct nf_conntrack_l3proto *l3proto;
1169 struct nf_conntrack_l4proto *l4proto;
1170 unsigned int *timeouts;
1171 unsigned int dataoff;
1177 /* Previously seen (loopback or untracked)? Ignore. */
1178 tmpl = (struct nf_conn *)skb->nfct;
1179 if (!nf_ct_is_template(tmpl)) {
1180 NF_CT_STAT_INC_ATOMIC(net, ignore);
1186 /* rcu_read_lock()ed by nf_hook_slow */
1187 l3proto = __nf_ct_l3proto_find(pf);
1188 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1189 &dataoff, &protonum);
1191 pr_debug("not prepared to track yet or error occurred\n");
1192 NF_CT_STAT_INC_ATOMIC(net, error);
1193 NF_CT_STAT_INC_ATOMIC(net, invalid);
1198 l4proto = __nf_ct_l4proto_find(pf, protonum);
1200 /* It may be an special packet, error, unclean...
1201 * inverse of the return code tells to the netfilter
1202 * core what to do with the packet. */
1203 if (l4proto->error != NULL) {
1204 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1207 NF_CT_STAT_INC_ATOMIC(net, error);
1208 NF_CT_STAT_INC_ATOMIC(net, invalid);
1212 /* ICMP[v6] protocol trackers may assign one conntrack. */
1217 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1218 l3proto, l4proto, &set_reply, &ctinfo);
1220 /* Not valid part of a connection */
1221 NF_CT_STAT_INC_ATOMIC(net, invalid);
1227 /* Too stressed to deal. */
1228 NF_CT_STAT_INC_ATOMIC(net, drop);
1233 NF_CT_ASSERT(skb->nfct);
1235 /* Decide what timeout policy we want to apply to this flow. */
1236 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1238 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1240 /* Invalid: inverse of the return code tells
1241 * the netfilter core what to do */
1242 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1243 nf_conntrack_put(skb->nfct);
1245 NF_CT_STAT_INC_ATOMIC(net, invalid);
1246 if (ret == -NF_DROP)
1247 NF_CT_STAT_INC_ATOMIC(net, drop);
1252 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1253 nf_conntrack_event_cache(IPCT_REPLY, ct);
1256 /* Special case: we have to repeat this hook, assign the
1257 * template again to this packet. We assume that this packet
1258 * has no conntrack assigned. This is used by nf_ct_tcp. */
1259 if (ret == NF_REPEAT)
1260 skb->nfct = (struct nf_conntrack *)tmpl;
1267 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1269 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1270 const struct nf_conntrack_tuple *orig)
1275 ret = nf_ct_invert_tuple(inverse, orig,
1276 __nf_ct_l3proto_find(orig->src.l3num),
1277 __nf_ct_l4proto_find(orig->src.l3num,
1278 orig->dst.protonum));
1282 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1284 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1285 implicitly racy: see __nf_conntrack_confirm */
1286 void nf_conntrack_alter_reply(struct nf_conn *ct,
1287 const struct nf_conntrack_tuple *newreply)
1289 struct nf_conn_help *help = nfct_help(ct);
1291 /* Should be unconfirmed, so not in hash table yet */
1292 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1294 pr_debug("Altering reply tuple of %p to ", ct);
1295 nf_ct_dump_tuple(newreply);
1297 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1298 if (ct->master || (help && !hlist_empty(&help->expectations)))
1302 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1305 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1307 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1308 void __nf_ct_refresh_acct(struct nf_conn *ct,
1309 enum ip_conntrack_info ctinfo,
1310 const struct sk_buff *skb,
1311 unsigned long extra_jiffies,
1314 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1317 /* Only update if this is not a fixed timeout */
1318 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1321 /* If not in hash table, timer will not be active yet */
1322 if (!nf_ct_is_confirmed(ct)) {
1323 ct->timeout.expires = extra_jiffies;
1325 unsigned long newtime = jiffies + extra_jiffies;
1327 /* Only update the timeout if the new timeout is at least
1328 HZ jiffies from the old timeout. Need del_timer for race
1329 avoidance (may already be dying). */
1330 if (newtime - ct->timeout.expires >= HZ)
1331 mod_timer_pending(&ct->timeout, newtime);
1336 nf_ct_acct_update(ct, ctinfo, skb->len);
1338 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1340 bool __nf_ct_kill_acct(struct nf_conn *ct,
1341 enum ip_conntrack_info ctinfo,
1342 const struct sk_buff *skb,
1346 nf_ct_acct_update(ct, ctinfo, skb->len);
1348 if (del_timer(&ct->timeout)) {
1349 ct->timeout.function((unsigned long)ct);
1354 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1356 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1358 #include <linux/netfilter/nfnetlink.h>
1359 #include <linux/netfilter/nfnetlink_conntrack.h>
1360 #include <linux/mutex.h>
1362 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1363 * in ip_conntrack_core, since we don't want the protocols to autoload
1364 * or depend on ctnetlink */
1365 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1366 const struct nf_conntrack_tuple *tuple)
1368 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1369 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1370 goto nla_put_failure;
1376 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1378 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1379 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1380 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1382 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1384 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1385 struct nf_conntrack_tuple *t)
1387 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1390 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1391 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1395 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1397 int nf_ct_port_nlattr_tuple_size(void)
1399 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1401 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1404 /* Used by ipt_REJECT and ip6t_REJECT. */
1405 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1408 enum ip_conntrack_info ctinfo;
1410 /* This ICMP is in reverse direction to the packet which caused it */
1411 ct = nf_ct_get(skb, &ctinfo);
1412 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1413 ctinfo = IP_CT_RELATED_REPLY;
1415 ctinfo = IP_CT_RELATED;
1417 /* Attach to new skbuff, and increment count */
1418 nskb->nfct = &ct->ct_general;
1419 nskb->nfctinfo = ctinfo;
1420 nf_conntrack_get(nskb->nfct);
1423 /* Bring out ya dead! */
1424 static struct nf_conn *
1425 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1426 void *data, unsigned int *bucket)
1428 struct nf_conntrack_tuple_hash *h;
1430 struct hlist_nulls_node *n;
1434 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1435 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1437 nf_conntrack_lock(lockp);
1438 if (*bucket < nf_conntrack_htable_size) {
1439 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1440 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1442 ct = nf_ct_tuplehash_to_ctrack(h);
1443 if (net_eq(nf_ct_net(ct), net) &&
1453 for_each_possible_cpu(cpu) {
1454 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1456 spin_lock_bh(&pcpu->lock);
1457 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1458 ct = nf_ct_tuplehash_to_ctrack(h);
1460 set_bit(IPS_DYING_BIT, &ct->status);
1462 spin_unlock_bh(&pcpu->lock);
1467 atomic_inc(&ct->ct_general.use);
1473 void nf_ct_iterate_cleanup(struct net *net,
1474 int (*iter)(struct nf_conn *i, void *data),
1475 void *data, u32 portid, int report)
1478 unsigned int bucket = 0;
1482 if (atomic_read(&net->ct.count) == 0)
1485 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1486 /* Time to push up daises... */
1487 if (del_timer(&ct->timeout))
1488 nf_ct_delete(ct, portid, report);
1490 /* ... else the timer will get him soon. */
1496 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1498 static int kill_all(struct nf_conn *i, void *data)
1503 void nf_ct_free_hashtable(void *hash, unsigned int size)
1505 if (is_vmalloc_addr(hash))
1508 free_pages((unsigned long)hash,
1509 get_order(sizeof(struct hlist_head) * size));
1511 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1513 static int untrack_refs(void)
1517 for_each_possible_cpu(cpu) {
1518 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1520 cnt += atomic_read(&ct->ct_general.use) - 1;
1525 void nf_conntrack_cleanup_start(void)
1527 RCU_INIT_POINTER(ip_ct_attach, NULL);
1530 void nf_conntrack_cleanup_end(void)
1532 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1533 while (untrack_refs() > 0)
1536 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1538 nf_conntrack_proto_fini();
1539 nf_conntrack_seqadj_fini();
1540 nf_conntrack_labels_fini();
1541 nf_conntrack_helper_fini();
1542 nf_conntrack_timeout_fini();
1543 nf_conntrack_ecache_fini();
1544 nf_conntrack_tstamp_fini();
1545 nf_conntrack_acct_fini();
1546 nf_conntrack_expect_fini();
1548 kmem_cache_destroy(nf_conntrack_cachep);
1552 * Mishearing the voices in his head, our hero wonders how he's
1553 * supposed to kill the mall.
1555 void nf_conntrack_cleanup_net(struct net *net)
1559 list_add(&net->exit_list, &single);
1560 nf_conntrack_cleanup_net_list(&single);
1563 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1569 * This makes sure all current packets have passed through
1570 * netfilter framework. Roll on, two-stage module
1576 list_for_each_entry(net, net_exit_list, exit_list) {
1577 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1578 if (atomic_read(&net->ct.count) != 0)
1583 goto i_see_dead_people;
1586 list_for_each_entry(net, net_exit_list, exit_list) {
1587 nf_conntrack_proto_pernet_fini(net);
1588 nf_conntrack_helper_pernet_fini(net);
1589 nf_conntrack_ecache_pernet_fini(net);
1590 nf_conntrack_tstamp_pernet_fini(net);
1591 nf_conntrack_acct_pernet_fini(net);
1592 nf_conntrack_expect_pernet_fini(net);
1593 free_percpu(net->ct.stat);
1594 free_percpu(net->ct.pcpu_lists);
1598 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1600 struct hlist_nulls_head *hash;
1601 unsigned int nr_slots, i;
1604 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1607 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1608 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1610 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1613 sz = nr_slots * sizeof(struct hlist_nulls_head);
1614 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1620 for (i = 0; i < nr_slots; i++)
1621 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1625 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1627 int nf_conntrack_hash_resize(unsigned int hashsize)
1630 unsigned int old_size;
1631 struct hlist_nulls_head *hash, *old_hash;
1632 struct nf_conntrack_tuple_hash *h;
1638 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1642 old_size = nf_conntrack_htable_size;
1643 if (old_size == hashsize) {
1644 nf_ct_free_hashtable(hash, hashsize);
1649 nf_conntrack_all_lock();
1650 write_seqcount_begin(&nf_conntrack_generation);
1652 /* Lookups in the old hash might happen in parallel, which means we
1653 * might get false negatives during connection lookup. New connections
1654 * created because of a false negative won't make it into the hash
1655 * though since that required taking the locks.
1658 for (i = 0; i < nf_conntrack_htable_size; i++) {
1659 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1660 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1661 struct nf_conntrack_tuple_hash, hnnode);
1662 ct = nf_ct_tuplehash_to_ctrack(h);
1663 hlist_nulls_del_rcu(&h->hnnode);
1664 bucket = __hash_conntrack(nf_ct_net(ct),
1665 &h->tuple, hashsize);
1666 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1669 old_size = nf_conntrack_htable_size;
1670 old_hash = nf_conntrack_hash;
1672 nf_conntrack_hash = hash;
1673 nf_conntrack_htable_size = hashsize;
1675 write_seqcount_end(&nf_conntrack_generation);
1676 nf_conntrack_all_unlock();
1680 nf_ct_free_hashtable(old_hash, old_size);
1684 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1686 unsigned int hashsize;
1689 if (current->nsproxy->net_ns != &init_net)
1692 /* On boot, we can set this without any fancy locking. */
1693 if (!nf_conntrack_htable_size)
1694 return param_set_uint(val, kp);
1696 rc = kstrtouint(val, 0, &hashsize);
1700 return nf_conntrack_hash_resize(hashsize);
1702 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1704 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1705 &nf_conntrack_htable_size, 0600);
1707 void nf_ct_untracked_status_or(unsigned long bits)
1711 for_each_possible_cpu(cpu)
1712 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1714 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1716 int nf_conntrack_init_start(void)
1722 seqcount_init(&nf_conntrack_generation);
1724 for (i = 0; i < CONNTRACK_LOCKS; i++)
1725 spin_lock_init(&nf_conntrack_locks[i]);
1727 if (!nf_conntrack_htable_size) {
1728 /* Idea from tcp.c: use 1/16384 of memory.
1729 * On i386: 32MB machine has 512 buckets.
1730 * >= 1GB machines have 16384 buckets.
1731 * >= 4GB machines have 65536 buckets.
1733 nf_conntrack_htable_size
1734 = (((totalram_pages << PAGE_SHIFT) / 16384)
1735 / sizeof(struct hlist_head));
1736 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1737 nf_conntrack_htable_size = 65536;
1738 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1739 nf_conntrack_htable_size = 16384;
1740 if (nf_conntrack_htable_size < 32)
1741 nf_conntrack_htable_size = 32;
1743 /* Use a max. factor of four by default to get the same max as
1744 * with the old struct list_heads. When a table size is given
1745 * we use the old value of 8 to avoid reducing the max.
1750 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1751 if (!nf_conntrack_hash)
1754 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1756 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1757 sizeof(struct nf_conn), 0,
1758 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1759 if (!nf_conntrack_cachep)
1762 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1763 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1766 ret = nf_conntrack_expect_init();
1770 ret = nf_conntrack_acct_init();
1774 ret = nf_conntrack_tstamp_init();
1778 ret = nf_conntrack_ecache_init();
1782 ret = nf_conntrack_timeout_init();
1786 ret = nf_conntrack_helper_init();
1790 ret = nf_conntrack_labels_init();
1794 ret = nf_conntrack_seqadj_init();
1798 ret = nf_conntrack_proto_init();
1802 /* Set up fake conntrack: to never be deleted, not in any hashes */
1803 for_each_possible_cpu(cpu) {
1804 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1805 write_pnet(&ct->ct_net, &init_net);
1806 atomic_set(&ct->ct_general.use, 1);
1808 /* - and look it like as a confirmed connection */
1809 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1813 nf_conntrack_seqadj_fini();
1815 nf_conntrack_labels_fini();
1817 nf_conntrack_helper_fini();
1819 nf_conntrack_timeout_fini();
1821 nf_conntrack_ecache_fini();
1823 nf_conntrack_tstamp_fini();
1825 nf_conntrack_acct_fini();
1827 nf_conntrack_expect_fini();
1829 kmem_cache_destroy(nf_conntrack_cachep);
1831 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1835 void nf_conntrack_init_end(void)
1837 /* For use by REJECT target */
1838 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1839 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1843 * We need to use special "null" values, not used in hash table
1845 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1846 #define DYING_NULLS_VAL ((1<<30)+1)
1847 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1849 int nf_conntrack_init_net(struct net *net)
1854 atomic_set(&net->ct.count, 0);
1856 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1857 if (!net->ct.pcpu_lists)
1860 for_each_possible_cpu(cpu) {
1861 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1863 spin_lock_init(&pcpu->lock);
1864 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1865 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1868 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1870 goto err_pcpu_lists;
1872 ret = nf_conntrack_expect_pernet_init(net);
1875 ret = nf_conntrack_acct_pernet_init(net);
1878 ret = nf_conntrack_tstamp_pernet_init(net);
1881 ret = nf_conntrack_ecache_pernet_init(net);
1884 ret = nf_conntrack_helper_pernet_init(net);
1887 ret = nf_conntrack_proto_pernet_init(net);
1893 nf_conntrack_helper_pernet_fini(net);
1895 nf_conntrack_ecache_pernet_fini(net);
1897 nf_conntrack_tstamp_pernet_fini(net);
1899 nf_conntrack_acct_pernet_fini(net);
1901 nf_conntrack_expect_pernet_fini(net);
1903 free_percpu(net->ct.stat);
1905 free_percpu(net->ct.pcpu_lists);