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 struct conntrack_gc_work {
76 struct delayed_work dwork;
82 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
83 static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
84 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
85 static __read_mostly bool nf_conntrack_locks_all;
87 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
88 #define GC_MAX_BUCKETS_DIV 64u
89 /* upper bound of scan intervals */
90 #define GC_INTERVAL_MAX (2 * HZ)
91 /* maximum conntracks to evict per gc run */
92 #define GC_MAX_EVICTS 256u
94 static struct conntrack_gc_work conntrack_gc_work;
96 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
99 while (unlikely(nf_conntrack_locks_all)) {
103 * Order the 'nf_conntrack_locks_all' load vs. the
104 * spin_unlock_wait() loads below, to ensure
105 * that 'nf_conntrack_locks_all_lock' is indeed held:
107 smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
108 spin_unlock_wait(&nf_conntrack_locks_all_lock);
112 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
114 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
116 h1 %= CONNTRACK_LOCKS;
117 h2 %= CONNTRACK_LOCKS;
118 spin_unlock(&nf_conntrack_locks[h1]);
120 spin_unlock(&nf_conntrack_locks[h2]);
123 /* return true if we need to recompute hashes (in case hash table was resized) */
124 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
125 unsigned int h2, unsigned int sequence)
127 h1 %= CONNTRACK_LOCKS;
128 h2 %= CONNTRACK_LOCKS;
130 nf_conntrack_lock(&nf_conntrack_locks[h1]);
132 spin_lock_nested(&nf_conntrack_locks[h2],
133 SINGLE_DEPTH_NESTING);
135 nf_conntrack_lock(&nf_conntrack_locks[h2]);
136 spin_lock_nested(&nf_conntrack_locks[h1],
137 SINGLE_DEPTH_NESTING);
139 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
140 nf_conntrack_double_unlock(h1, h2);
146 static void nf_conntrack_all_lock(void)
150 spin_lock(&nf_conntrack_locks_all_lock);
151 nf_conntrack_locks_all = true;
154 * Order the above store of 'nf_conntrack_locks_all' against
155 * the spin_unlock_wait() loads below, such that if
156 * nf_conntrack_lock() observes 'nf_conntrack_locks_all'
157 * we must observe nf_conntrack_locks[] held:
159 smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
161 for (i = 0; i < CONNTRACK_LOCKS; i++) {
162 spin_unlock_wait(&nf_conntrack_locks[i]);
166 static void nf_conntrack_all_unlock(void)
169 * All prior stores must be complete before we clear
170 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
171 * might observe the false value but not the entire
174 smp_store_release(&nf_conntrack_locks_all, false);
175 spin_unlock(&nf_conntrack_locks_all_lock);
178 unsigned int nf_conntrack_htable_size __read_mostly;
179 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
181 unsigned int nf_conntrack_max __read_mostly;
182 seqcount_t nf_conntrack_generation __read_mostly;
184 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
185 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
187 static unsigned int nf_conntrack_hash_rnd __read_mostly;
189 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
190 const struct net *net)
195 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
197 /* The direction must be ignored, so we hash everything up to the
198 * destination ports (which is a multiple of 4) and treat the last
199 * three bytes manually.
201 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
202 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
203 return jhash2((u32 *)tuple, n, seed ^
204 (((__force __u16)tuple->dst.u.all << 16) |
205 tuple->dst.protonum));
208 static u32 scale_hash(u32 hash)
210 return reciprocal_scale(hash, nf_conntrack_htable_size);
213 static u32 __hash_conntrack(const struct net *net,
214 const struct nf_conntrack_tuple *tuple,
217 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
220 static u32 hash_conntrack(const struct net *net,
221 const struct nf_conntrack_tuple *tuple)
223 return scale_hash(hash_conntrack_raw(tuple, net));
227 nf_ct_get_tuple(const struct sk_buff *skb,
229 unsigned int dataoff,
233 struct nf_conntrack_tuple *tuple,
234 const struct nf_conntrack_l3proto *l3proto,
235 const struct nf_conntrack_l4proto *l4proto)
237 memset(tuple, 0, sizeof(*tuple));
239 tuple->src.l3num = l3num;
240 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
243 tuple->dst.protonum = protonum;
244 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
246 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
248 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
250 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
252 struct net *net, struct nf_conntrack_tuple *tuple)
254 struct nf_conntrack_l3proto *l3proto;
255 struct nf_conntrack_l4proto *l4proto;
256 unsigned int protoff;
262 l3proto = __nf_ct_l3proto_find(l3num);
263 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
264 if (ret != NF_ACCEPT) {
269 l4proto = __nf_ct_l4proto_find(l3num, protonum);
271 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
277 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
280 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
281 const struct nf_conntrack_tuple *orig,
282 const struct nf_conntrack_l3proto *l3proto,
283 const struct nf_conntrack_l4proto *l4proto)
285 memset(inverse, 0, sizeof(*inverse));
287 inverse->src.l3num = orig->src.l3num;
288 if (l3proto->invert_tuple(inverse, orig) == 0)
291 inverse->dst.dir = !orig->dst.dir;
293 inverse->dst.protonum = orig->dst.protonum;
294 return l4proto->invert_tuple(inverse, orig);
296 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
299 clean_from_lists(struct nf_conn *ct)
301 pr_debug("clean_from_lists(%p)\n", ct);
302 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
303 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
305 /* Destroy all pending expectations */
306 nf_ct_remove_expectations(ct);
309 /* must be called with local_bh_disable */
310 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
312 struct ct_pcpu *pcpu;
314 /* add this conntrack to the (per cpu) dying list */
315 ct->cpu = smp_processor_id();
316 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
318 spin_lock(&pcpu->lock);
319 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
321 spin_unlock(&pcpu->lock);
324 /* must be called with local_bh_disable */
325 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
327 struct ct_pcpu *pcpu;
329 /* add this conntrack to the (per cpu) unconfirmed list */
330 ct->cpu = smp_processor_id();
331 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
333 spin_lock(&pcpu->lock);
334 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
336 spin_unlock(&pcpu->lock);
339 /* must be called with local_bh_disable */
340 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
342 struct ct_pcpu *pcpu;
344 /* We overload first tuple to link into unconfirmed or dying list.*/
345 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
347 spin_lock(&pcpu->lock);
348 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
349 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
350 spin_unlock(&pcpu->lock);
353 /* Released via destroy_conntrack() */
354 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
355 const struct nf_conntrack_zone *zone,
358 struct nf_conn *tmpl;
360 tmpl = kzalloc(sizeof(*tmpl), flags);
364 tmpl->status = IPS_TEMPLATE;
365 write_pnet(&tmpl->ct_net, net);
366 nf_ct_zone_add(tmpl, zone);
367 atomic_set(&tmpl->ct_general.use, 0);
371 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
373 void nf_ct_tmpl_free(struct nf_conn *tmpl)
375 nf_ct_ext_destroy(tmpl);
376 nf_ct_ext_free(tmpl);
379 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
382 destroy_conntrack(struct nf_conntrack *nfct)
384 struct nf_conn *ct = (struct nf_conn *)nfct;
385 struct nf_conntrack_l4proto *l4proto;
387 pr_debug("destroy_conntrack(%p)\n", ct);
388 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
390 if (unlikely(nf_ct_is_template(ct))) {
395 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
396 if (l4proto->destroy)
397 l4proto->destroy(ct);
402 /* Expectations will have been removed in clean_from_lists,
403 * except TFTP can create an expectation on the first packet,
404 * before connection is in the list, so we need to clean here,
407 nf_ct_remove_expectations(ct);
409 nf_ct_del_from_dying_or_unconfirmed_list(ct);
414 nf_ct_put(ct->master);
416 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
417 nf_conntrack_free(ct);
420 static void nf_ct_delete_from_lists(struct nf_conn *ct)
422 struct net *net = nf_ct_net(ct);
423 unsigned int hash, reply_hash;
424 unsigned int sequence;
426 nf_ct_helper_destroy(ct);
430 sequence = read_seqcount_begin(&nf_conntrack_generation);
431 hash = hash_conntrack(net,
432 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
433 reply_hash = hash_conntrack(net,
434 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
435 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
437 clean_from_lists(ct);
438 nf_conntrack_double_unlock(hash, reply_hash);
440 nf_ct_add_to_dying_list(ct);
445 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
447 struct nf_conn_tstamp *tstamp;
449 if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
452 tstamp = nf_conn_tstamp_find(ct);
453 if (tstamp && tstamp->stop == 0)
454 tstamp->stop = ktime_get_real_ns();
456 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
457 portid, report) < 0) {
458 /* destroy event was not delivered. nf_ct_put will
459 * be done by event cache worker on redelivery.
461 nf_ct_delete_from_lists(ct);
462 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
466 nf_conntrack_ecache_work(nf_ct_net(ct));
467 nf_ct_delete_from_lists(ct);
471 EXPORT_SYMBOL_GPL(nf_ct_delete);
474 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
475 const struct nf_conntrack_tuple *tuple,
476 const struct nf_conntrack_zone *zone,
477 const struct net *net)
479 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
481 /* A conntrack can be recreated with the equal tuple,
482 * so we need to check that the conntrack is confirmed
484 return nf_ct_tuple_equal(tuple, &h->tuple) &&
485 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
486 nf_ct_is_confirmed(ct) &&
487 net_eq(net, nf_ct_net(ct));
490 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
491 static void nf_ct_gc_expired(struct nf_conn *ct)
493 if (!atomic_inc_not_zero(&ct->ct_general.use))
496 if (nf_ct_should_gc(ct))
504 * - Caller must take a reference on returned object
505 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
507 static struct nf_conntrack_tuple_hash *
508 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
509 const struct nf_conntrack_tuple *tuple, u32 hash)
511 struct nf_conntrack_tuple_hash *h;
512 struct hlist_nulls_head *ct_hash;
513 struct hlist_nulls_node *n;
514 unsigned int bucket, hsize;
517 nf_conntrack_get_ht(&ct_hash, &hsize);
518 bucket = reciprocal_scale(hash, hsize);
520 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
523 ct = nf_ct_tuplehash_to_ctrack(h);
524 if (nf_ct_is_expired(ct)) {
525 nf_ct_gc_expired(ct);
529 if (nf_ct_is_dying(ct))
532 if (nf_ct_key_equal(h, tuple, zone, net))
536 * if the nulls value we got at the end of this lookup is
537 * not the expected one, we must restart lookup.
538 * We probably met an item that was moved to another chain.
540 if (get_nulls_value(n) != bucket) {
541 NF_CT_STAT_INC_ATOMIC(net, search_restart);
548 /* Find a connection corresponding to a tuple. */
549 static 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, u32 hash)
553 struct nf_conntrack_tuple_hash *h;
558 h = ____nf_conntrack_find(net, zone, tuple, hash);
560 ct = nf_ct_tuplehash_to_ctrack(h);
561 if (unlikely(nf_ct_is_dying(ct) ||
562 !atomic_inc_not_zero(&ct->ct_general.use)))
565 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
576 struct nf_conntrack_tuple_hash *
577 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
578 const struct nf_conntrack_tuple *tuple)
580 return __nf_conntrack_find_get(net, zone, tuple,
581 hash_conntrack_raw(tuple, net));
583 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
585 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
587 unsigned int reply_hash)
589 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
590 &nf_conntrack_hash[hash]);
591 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
592 &nf_conntrack_hash[reply_hash]);
596 nf_conntrack_hash_check_insert(struct nf_conn *ct)
598 const struct nf_conntrack_zone *zone;
599 struct net *net = nf_ct_net(ct);
600 unsigned int hash, reply_hash;
601 struct nf_conntrack_tuple_hash *h;
602 struct hlist_nulls_node *n;
603 unsigned int sequence;
605 zone = nf_ct_zone(ct);
609 sequence = read_seqcount_begin(&nf_conntrack_generation);
610 hash = hash_conntrack(net,
611 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
612 reply_hash = hash_conntrack(net,
613 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
614 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
616 /* See if there's one in the list already, including reverse */
617 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
618 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
622 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
623 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
628 /* The caller holds a reference to this object */
629 atomic_set(&ct->ct_general.use, 2);
630 __nf_conntrack_hash_insert(ct, hash, reply_hash);
631 nf_conntrack_double_unlock(hash, reply_hash);
632 NF_CT_STAT_INC(net, insert);
637 nf_conntrack_double_unlock(hash, reply_hash);
638 NF_CT_STAT_INC(net, insert_failed);
642 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
644 static inline void nf_ct_acct_update(struct nf_conn *ct,
645 enum ip_conntrack_info ctinfo,
648 struct nf_conn_acct *acct;
650 acct = nf_conn_acct_find(ct);
652 struct nf_conn_counter *counter = acct->counter;
654 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
655 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
659 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
660 const struct nf_conn *loser_ct)
662 struct nf_conn_acct *acct;
664 acct = nf_conn_acct_find(loser_ct);
666 struct nf_conn_counter *counter = acct->counter;
669 /* u32 should be fine since we must have seen one packet. */
670 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
671 nf_ct_acct_update(ct, ctinfo, bytes);
675 /* Resolve race on insertion if this protocol allows this. */
676 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
677 enum ip_conntrack_info ctinfo,
678 struct nf_conntrack_tuple_hash *h)
680 /* This is the conntrack entry already in hashes that won race. */
681 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
682 struct nf_conntrack_l4proto *l4proto;
684 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
685 if (l4proto->allow_clash &&
687 !nf_ct_is_dying(ct) &&
688 atomic_inc_not_zero(&ct->ct_general.use)) {
689 nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct);
690 nf_conntrack_put(skb->nfct);
691 /* Assign conntrack already in hashes to this skbuff. Don't
692 * modify skb->nfctinfo to ensure consistent stateful filtering.
694 skb->nfct = &ct->ct_general;
697 NF_CT_STAT_INC(net, drop);
701 /* Confirm a connection given skb; places it in hash table */
703 __nf_conntrack_confirm(struct sk_buff *skb)
705 const struct nf_conntrack_zone *zone;
706 unsigned int hash, reply_hash;
707 struct nf_conntrack_tuple_hash *h;
709 struct nf_conn_help *help;
710 struct nf_conn_tstamp *tstamp;
711 struct hlist_nulls_node *n;
712 enum ip_conntrack_info ctinfo;
714 unsigned int sequence;
717 ct = nf_ct_get(skb, &ctinfo);
720 /* ipt_REJECT uses nf_conntrack_attach to attach related
721 ICMP/TCP RST packets in other direction. Actual packet
722 which created connection will be IP_CT_NEW or for an
723 expected connection, IP_CT_RELATED. */
724 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
727 zone = nf_ct_zone(ct);
731 sequence = read_seqcount_begin(&nf_conntrack_generation);
732 /* reuse the hash saved before */
733 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
734 hash = scale_hash(hash);
735 reply_hash = hash_conntrack(net,
736 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
738 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
740 /* We're not in hash table, and we refuse to set up related
741 * connections for unconfirmed conns. But packet copies and
742 * REJECT will give spurious warnings here.
744 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
746 /* No external references means no one else could have
749 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
750 pr_debug("Confirming conntrack %p\n", ct);
751 /* We have to check the DYING flag after unlink to prevent
752 * a race against nf_ct_get_next_corpse() possibly called from
753 * user context, else we insert an already 'dead' hash, blocking
754 * further use of that particular connection -JM.
756 nf_ct_del_from_dying_or_unconfirmed_list(ct);
758 if (unlikely(nf_ct_is_dying(ct))) {
759 nf_ct_add_to_dying_list(ct);
763 /* See if there's one in the list already, including reverse:
764 NAT could have grabbed it without realizing, since we're
765 not in the hash. If there is, we lost race. */
766 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
767 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
771 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
772 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
776 /* Timer relative to confirmation time, not original
777 setting time, otherwise we'd get timer wrap in
778 weird delay cases. */
779 ct->timeout += nfct_time_stamp;
780 atomic_inc(&ct->ct_general.use);
781 ct->status |= IPS_CONFIRMED;
783 /* set conntrack timestamp, if enabled. */
784 tstamp = nf_conn_tstamp_find(ct);
786 if (skb->tstamp.tv64 == 0)
787 __net_timestamp(skb);
789 tstamp->start = ktime_to_ns(skb->tstamp);
791 /* Since the lookup is lockless, hash insertion must be done after
792 * starting the timer and setting the CONFIRMED bit. The RCU barriers
793 * guarantee that no other CPU can find the conntrack before the above
794 * stores are visible.
796 __nf_conntrack_hash_insert(ct, hash, reply_hash);
797 nf_conntrack_double_unlock(hash, reply_hash);
800 help = nfct_help(ct);
801 if (help && help->helper)
802 nf_conntrack_event_cache(IPCT_HELPER, ct);
804 nf_conntrack_event_cache(master_ct(ct) ?
805 IPCT_RELATED : IPCT_NEW, ct);
809 nf_ct_add_to_dying_list(ct);
810 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
812 nf_conntrack_double_unlock(hash, reply_hash);
813 NF_CT_STAT_INC(net, insert_failed);
817 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
819 /* Returns true if a connection correspondings to the tuple (required
822 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
823 const struct nf_conn *ignored_conntrack)
825 struct net *net = nf_ct_net(ignored_conntrack);
826 const struct nf_conntrack_zone *zone;
827 struct nf_conntrack_tuple_hash *h;
828 struct hlist_nulls_head *ct_hash;
829 unsigned int hash, hsize;
830 struct hlist_nulls_node *n;
833 zone = nf_ct_zone(ignored_conntrack);
837 nf_conntrack_get_ht(&ct_hash, &hsize);
838 hash = __hash_conntrack(net, tuple, hsize);
840 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
841 ct = nf_ct_tuplehash_to_ctrack(h);
843 if (ct == ignored_conntrack)
846 if (nf_ct_is_expired(ct)) {
847 nf_ct_gc_expired(ct);
851 if (nf_ct_key_equal(h, tuple, zone, net)) {
852 NF_CT_STAT_INC_ATOMIC(net, found);
858 if (get_nulls_value(n) != hash) {
859 NF_CT_STAT_INC_ATOMIC(net, search_restart);
867 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
869 #define NF_CT_EVICTION_RANGE 8
871 /* There's a small race here where we may free a just-assured
872 connection. Too bad: we're in trouble anyway. */
873 static unsigned int early_drop_list(struct net *net,
874 struct hlist_nulls_head *head)
876 struct nf_conntrack_tuple_hash *h;
877 struct hlist_nulls_node *n;
878 unsigned int drops = 0;
881 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
882 tmp = nf_ct_tuplehash_to_ctrack(h);
884 if (nf_ct_is_expired(tmp)) {
885 nf_ct_gc_expired(tmp);
889 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
890 !net_eq(nf_ct_net(tmp), net) ||
894 if (!atomic_inc_not_zero(&tmp->ct_general.use))
897 /* kill only if still in same netns -- might have moved due to
898 * SLAB_DESTROY_BY_RCU rules.
900 * We steal the timer reference. If that fails timer has
901 * already fired or someone else deleted it. Just drop ref
902 * and move to next entry.
904 if (net_eq(nf_ct_net(tmp), net) &&
905 nf_ct_is_confirmed(tmp) &&
906 nf_ct_delete(tmp, 0, 0))
915 static noinline int early_drop(struct net *net, unsigned int _hash)
919 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
920 struct hlist_nulls_head *ct_hash;
921 unsigned int hash, hsize, drops;
924 nf_conntrack_get_ht(&ct_hash, &hsize);
925 hash = reciprocal_scale(_hash++, hsize);
927 drops = early_drop_list(net, &ct_hash[hash]);
931 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
939 static void gc_worker(struct work_struct *work)
941 unsigned int i, goal, buckets = 0, expired_count = 0;
942 struct conntrack_gc_work *gc_work;
943 unsigned int ratio, scanned = 0;
944 unsigned long next_run;
946 gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
948 goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
949 i = gc_work->last_bucket;
952 struct nf_conntrack_tuple_hash *h;
953 struct hlist_nulls_head *ct_hash;
954 struct hlist_nulls_node *n;
961 nf_conntrack_get_ht(&ct_hash, &hashsz);
965 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
966 tmp = nf_ct_tuplehash_to_ctrack(h);
969 if (nf_ct_is_expired(tmp)) {
970 nf_ct_gc_expired(tmp);
976 /* could check get_nulls_value() here and restart if ct
977 * was moved to another chain. But given gc is best-effort
978 * we will just continue with next hash slot.
981 cond_resched_rcu_qs();
982 } while (++buckets < goal &&
983 expired_count < GC_MAX_EVICTS);
985 if (gc_work->exiting)
989 * Eviction will normally happen from the packet path, and not
990 * from this gc worker.
992 * This worker is only here to reap expired entries when system went
993 * idle after a busy period.
995 * The heuristics below are supposed to balance conflicting goals:
997 * 1. Minimize time until we notice a stale entry
998 * 2. Maximize scan intervals to not waste cycles
1000 * Normally, expired_count will be 0, this increases the next_run time
1001 * to priorize 2) above.
1003 * As soon as a timed-out entry is found, move towards 1) and increase
1004 * the scan frequency.
1005 * In case we have lots of evictions next scan is done immediately.
1007 ratio = scanned ? expired_count * 100 / scanned : 0;
1008 if (ratio >= 90 || expired_count == GC_MAX_EVICTS) {
1009 gc_work->next_gc_run = 0;
1011 } else if (expired_count) {
1012 gc_work->next_gc_run /= 2U;
1013 next_run = msecs_to_jiffies(1);
1015 if (gc_work->next_gc_run < GC_INTERVAL_MAX)
1016 gc_work->next_gc_run += msecs_to_jiffies(1);
1018 next_run = gc_work->next_gc_run;
1021 gc_work->last_bucket = i;
1022 queue_delayed_work(system_long_wq, &gc_work->dwork, next_run);
1025 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1027 INIT_DELAYED_WORK(&gc_work->dwork, gc_worker);
1028 gc_work->next_gc_run = GC_INTERVAL_MAX;
1029 gc_work->exiting = false;
1032 static struct nf_conn *
1033 __nf_conntrack_alloc(struct net *net,
1034 const struct nf_conntrack_zone *zone,
1035 const struct nf_conntrack_tuple *orig,
1036 const struct nf_conntrack_tuple *repl,
1037 gfp_t gfp, u32 hash)
1041 /* We don't want any race condition at early drop stage */
1042 atomic_inc(&net->ct.count);
1044 if (nf_conntrack_max &&
1045 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1046 if (!early_drop(net, hash)) {
1047 atomic_dec(&net->ct.count);
1048 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1049 return ERR_PTR(-ENOMEM);
1054 * Do not use kmem_cache_zalloc(), as this cache uses
1055 * SLAB_DESTROY_BY_RCU.
1057 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1061 spin_lock_init(&ct->lock);
1062 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1063 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1064 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1065 /* save hash for reusing when confirming */
1066 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1068 write_pnet(&ct->ct_net, net);
1069 memset(&ct->__nfct_init_offset[0], 0,
1070 offsetof(struct nf_conn, proto) -
1071 offsetof(struct nf_conn, __nfct_init_offset[0]));
1073 nf_ct_zone_add(ct, zone);
1075 /* Because we use RCU lookups, we set ct_general.use to zero before
1076 * this is inserted in any list.
1078 atomic_set(&ct->ct_general.use, 0);
1081 atomic_dec(&net->ct.count);
1082 return ERR_PTR(-ENOMEM);
1085 struct nf_conn *nf_conntrack_alloc(struct net *net,
1086 const struct nf_conntrack_zone *zone,
1087 const struct nf_conntrack_tuple *orig,
1088 const struct nf_conntrack_tuple *repl,
1091 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1093 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1095 void nf_conntrack_free(struct nf_conn *ct)
1097 struct net *net = nf_ct_net(ct);
1099 /* A freed object has refcnt == 0, that's
1100 * the golden rule for SLAB_DESTROY_BY_RCU
1102 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
1104 nf_ct_ext_destroy(ct);
1106 kmem_cache_free(nf_conntrack_cachep, ct);
1107 smp_mb__before_atomic();
1108 atomic_dec(&net->ct.count);
1110 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1113 /* Allocate a new conntrack: we return -ENOMEM if classification
1114 failed due to stress. Otherwise it really is unclassifiable. */
1115 static struct nf_conntrack_tuple_hash *
1116 init_conntrack(struct net *net, struct nf_conn *tmpl,
1117 const struct nf_conntrack_tuple *tuple,
1118 struct nf_conntrack_l3proto *l3proto,
1119 struct nf_conntrack_l4proto *l4proto,
1120 struct sk_buff *skb,
1121 unsigned int dataoff, u32 hash)
1124 struct nf_conn_help *help;
1125 struct nf_conntrack_tuple repl_tuple;
1126 struct nf_conntrack_ecache *ecache;
1127 struct nf_conntrack_expect *exp = NULL;
1128 const struct nf_conntrack_zone *zone;
1129 struct nf_conn_timeout *timeout_ext;
1130 struct nf_conntrack_zone tmp;
1131 unsigned int *timeouts;
1133 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1134 pr_debug("Can't invert tuple.\n");
1138 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1139 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1142 return (struct nf_conntrack_tuple_hash *)ct;
1144 if (!nf_ct_add_synproxy(ct, tmpl)) {
1145 nf_conntrack_free(ct);
1146 return ERR_PTR(-ENOMEM);
1149 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1151 timeouts = nf_ct_timeout_data(timeout_ext);
1152 if (unlikely(!timeouts))
1153 timeouts = l4proto->get_timeouts(net);
1155 timeouts = l4proto->get_timeouts(net);
1158 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1159 nf_conntrack_free(ct);
1160 pr_debug("can't track with proto module\n");
1165 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1168 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1169 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1170 nf_ct_labels_ext_add(ct);
1172 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1173 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1174 ecache ? ecache->expmask : 0,
1178 if (net->ct.expect_count) {
1179 spin_lock(&nf_conntrack_expect_lock);
1180 exp = nf_ct_find_expectation(net, zone, tuple);
1182 pr_debug("expectation arrives ct=%p exp=%p\n",
1184 /* Welcome, Mr. Bond. We've been expecting you... */
1185 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1186 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1187 ct->master = exp->master;
1189 help = nf_ct_helper_ext_add(ct, exp->helper,
1192 rcu_assign_pointer(help->helper, exp->helper);
1195 #ifdef CONFIG_NF_CONNTRACK_MARK
1196 ct->mark = exp->master->mark;
1198 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1199 ct->secmark = exp->master->secmark;
1201 NF_CT_STAT_INC(net, expect_new);
1203 spin_unlock(&nf_conntrack_expect_lock);
1206 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1208 /* Now it is inserted into the unconfirmed list, bump refcount */
1209 nf_conntrack_get(&ct->ct_general);
1210 nf_ct_add_to_unconfirmed_list(ct);
1216 exp->expectfn(ct, exp);
1217 nf_ct_expect_put(exp);
1220 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1223 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1224 static inline struct nf_conn *
1225 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1226 struct sk_buff *skb,
1227 unsigned int dataoff,
1230 struct nf_conntrack_l3proto *l3proto,
1231 struct nf_conntrack_l4proto *l4proto,
1233 enum ip_conntrack_info *ctinfo)
1235 const struct nf_conntrack_zone *zone;
1236 struct nf_conntrack_tuple tuple;
1237 struct nf_conntrack_tuple_hash *h;
1238 struct nf_conntrack_zone tmp;
1242 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1243 dataoff, l3num, protonum, net, &tuple, l3proto,
1245 pr_debug("Can't get tuple\n");
1249 /* look for tuple match */
1250 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1251 hash = hash_conntrack_raw(&tuple, net);
1252 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1254 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1255 skb, dataoff, hash);
1261 ct = nf_ct_tuplehash_to_ctrack(h);
1263 /* It exists; we have (non-exclusive) reference. */
1264 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1265 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1266 /* Please set reply bit if this packet OK */
1269 /* Once we've had two way comms, always ESTABLISHED. */
1270 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1271 pr_debug("normal packet for %p\n", ct);
1272 *ctinfo = IP_CT_ESTABLISHED;
1273 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1274 pr_debug("related packet for %p\n", ct);
1275 *ctinfo = IP_CT_RELATED;
1277 pr_debug("new packet for %p\n", ct);
1278 *ctinfo = IP_CT_NEW;
1282 skb->nfct = &ct->ct_general;
1283 skb->nfctinfo = *ctinfo;
1288 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1289 struct sk_buff *skb)
1291 struct nf_conn *ct, *tmpl = NULL;
1292 enum ip_conntrack_info ctinfo;
1293 struct nf_conntrack_l3proto *l3proto;
1294 struct nf_conntrack_l4proto *l4proto;
1295 unsigned int *timeouts;
1296 unsigned int dataoff;
1302 /* Previously seen (loopback or untracked)? Ignore. */
1303 tmpl = (struct nf_conn *)skb->nfct;
1304 if (!nf_ct_is_template(tmpl)) {
1305 NF_CT_STAT_INC_ATOMIC(net, ignore);
1311 /* rcu_read_lock()ed by nf_hook_thresh */
1312 l3proto = __nf_ct_l3proto_find(pf);
1313 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1314 &dataoff, &protonum);
1316 pr_debug("not prepared to track yet or error occurred\n");
1317 NF_CT_STAT_INC_ATOMIC(net, error);
1318 NF_CT_STAT_INC_ATOMIC(net, invalid);
1323 l4proto = __nf_ct_l4proto_find(pf, protonum);
1325 /* It may be an special packet, error, unclean...
1326 * inverse of the return code tells to the netfilter
1327 * core what to do with the packet. */
1328 if (l4proto->error != NULL) {
1329 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1332 NF_CT_STAT_INC_ATOMIC(net, error);
1333 NF_CT_STAT_INC_ATOMIC(net, invalid);
1337 /* ICMP[v6] protocol trackers may assign one conntrack. */
1342 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1343 l3proto, l4proto, &set_reply, &ctinfo);
1345 /* Not valid part of a connection */
1346 NF_CT_STAT_INC_ATOMIC(net, invalid);
1352 /* Too stressed to deal. */
1353 NF_CT_STAT_INC_ATOMIC(net, drop);
1358 NF_CT_ASSERT(skb->nfct);
1360 /* Decide what timeout policy we want to apply to this flow. */
1361 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1363 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1365 /* Invalid: inverse of the return code tells
1366 * the netfilter core what to do */
1367 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1368 nf_conntrack_put(skb->nfct);
1370 NF_CT_STAT_INC_ATOMIC(net, invalid);
1371 if (ret == -NF_DROP)
1372 NF_CT_STAT_INC_ATOMIC(net, drop);
1377 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1378 nf_conntrack_event_cache(IPCT_REPLY, ct);
1381 /* Special case: we have to repeat this hook, assign the
1382 * template again to this packet. We assume that this packet
1383 * has no conntrack assigned. This is used by nf_ct_tcp. */
1384 if (ret == NF_REPEAT)
1385 skb->nfct = (struct nf_conntrack *)tmpl;
1392 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1394 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1395 const struct nf_conntrack_tuple *orig)
1400 ret = nf_ct_invert_tuple(inverse, orig,
1401 __nf_ct_l3proto_find(orig->src.l3num),
1402 __nf_ct_l4proto_find(orig->src.l3num,
1403 orig->dst.protonum));
1407 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1409 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1410 implicitly racy: see __nf_conntrack_confirm */
1411 void nf_conntrack_alter_reply(struct nf_conn *ct,
1412 const struct nf_conntrack_tuple *newreply)
1414 struct nf_conn_help *help = nfct_help(ct);
1416 /* Should be unconfirmed, so not in hash table yet */
1417 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1419 pr_debug("Altering reply tuple of %p to ", ct);
1420 nf_ct_dump_tuple(newreply);
1422 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1423 if (ct->master || (help && !hlist_empty(&help->expectations)))
1427 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1430 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1432 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1433 void __nf_ct_refresh_acct(struct nf_conn *ct,
1434 enum ip_conntrack_info ctinfo,
1435 const struct sk_buff *skb,
1436 unsigned long extra_jiffies,
1441 /* Only update if this is not a fixed timeout */
1442 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1445 /* If not in hash table, timer will not be active yet */
1446 if (nf_ct_is_confirmed(ct))
1447 extra_jiffies += nfct_time_stamp;
1449 ct->timeout = extra_jiffies;
1452 nf_ct_acct_update(ct, ctinfo, skb->len);
1454 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1456 bool nf_ct_kill_acct(struct nf_conn *ct,
1457 enum ip_conntrack_info ctinfo,
1458 const struct sk_buff *skb)
1460 nf_ct_acct_update(ct, ctinfo, skb->len);
1462 return nf_ct_delete(ct, 0, 0);
1464 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1466 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1468 #include <linux/netfilter/nfnetlink.h>
1469 #include <linux/netfilter/nfnetlink_conntrack.h>
1470 #include <linux/mutex.h>
1472 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1473 * in ip_conntrack_core, since we don't want the protocols to autoload
1474 * or depend on ctnetlink */
1475 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1476 const struct nf_conntrack_tuple *tuple)
1478 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1479 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1480 goto nla_put_failure;
1486 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1488 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1489 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1490 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1492 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1494 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1495 struct nf_conntrack_tuple *t)
1497 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1500 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1501 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1505 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1507 int nf_ct_port_nlattr_tuple_size(void)
1509 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1511 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1514 /* Used by ipt_REJECT and ip6t_REJECT. */
1515 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1518 enum ip_conntrack_info ctinfo;
1520 /* This ICMP is in reverse direction to the packet which caused it */
1521 ct = nf_ct_get(skb, &ctinfo);
1522 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1523 ctinfo = IP_CT_RELATED_REPLY;
1525 ctinfo = IP_CT_RELATED;
1527 /* Attach to new skbuff, and increment count */
1528 nskb->nfct = &ct->ct_general;
1529 nskb->nfctinfo = ctinfo;
1530 nf_conntrack_get(nskb->nfct);
1533 /* Bring out ya dead! */
1534 static struct nf_conn *
1535 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1536 void *data, unsigned int *bucket)
1538 struct nf_conntrack_tuple_hash *h;
1540 struct hlist_nulls_node *n;
1544 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1545 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1547 nf_conntrack_lock(lockp);
1548 if (*bucket < nf_conntrack_htable_size) {
1549 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1550 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1552 ct = nf_ct_tuplehash_to_ctrack(h);
1553 if (net_eq(nf_ct_net(ct), net) &&
1563 for_each_possible_cpu(cpu) {
1564 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1566 spin_lock_bh(&pcpu->lock);
1567 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1568 ct = nf_ct_tuplehash_to_ctrack(h);
1570 set_bit(IPS_DYING_BIT, &ct->status);
1572 spin_unlock_bh(&pcpu->lock);
1577 atomic_inc(&ct->ct_general.use);
1583 void nf_ct_iterate_cleanup(struct net *net,
1584 int (*iter)(struct nf_conn *i, void *data),
1585 void *data, u32 portid, int report)
1588 unsigned int bucket = 0;
1592 if (atomic_read(&net->ct.count) == 0)
1595 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1596 /* Time to push up daises... */
1598 nf_ct_delete(ct, portid, report);
1603 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1605 static int kill_all(struct nf_conn *i, void *data)
1610 void nf_ct_free_hashtable(void *hash, unsigned int size)
1612 if (is_vmalloc_addr(hash))
1615 free_pages((unsigned long)hash,
1616 get_order(sizeof(struct hlist_head) * size));
1618 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1620 static int untrack_refs(void)
1624 for_each_possible_cpu(cpu) {
1625 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1627 cnt += atomic_read(&ct->ct_general.use) - 1;
1632 void nf_conntrack_cleanup_start(void)
1634 conntrack_gc_work.exiting = true;
1635 RCU_INIT_POINTER(ip_ct_attach, NULL);
1638 void nf_conntrack_cleanup_end(void)
1640 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1641 while (untrack_refs() > 0)
1644 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
1645 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1647 nf_conntrack_proto_fini();
1648 nf_conntrack_seqadj_fini();
1649 nf_conntrack_labels_fini();
1650 nf_conntrack_helper_fini();
1651 nf_conntrack_timeout_fini();
1652 nf_conntrack_ecache_fini();
1653 nf_conntrack_tstamp_fini();
1654 nf_conntrack_acct_fini();
1655 nf_conntrack_expect_fini();
1657 kmem_cache_destroy(nf_conntrack_cachep);
1661 * Mishearing the voices in his head, our hero wonders how he's
1662 * supposed to kill the mall.
1664 void nf_conntrack_cleanup_net(struct net *net)
1668 list_add(&net->exit_list, &single);
1669 nf_conntrack_cleanup_net_list(&single);
1672 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1678 * This makes sure all current packets have passed through
1679 * netfilter framework. Roll on, two-stage module
1685 list_for_each_entry(net, net_exit_list, exit_list) {
1686 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1687 if (atomic_read(&net->ct.count) != 0)
1692 goto i_see_dead_people;
1695 list_for_each_entry(net, net_exit_list, exit_list) {
1696 nf_conntrack_proto_pernet_fini(net);
1697 nf_conntrack_helper_pernet_fini(net);
1698 nf_conntrack_ecache_pernet_fini(net);
1699 nf_conntrack_tstamp_pernet_fini(net);
1700 nf_conntrack_acct_pernet_fini(net);
1701 nf_conntrack_expect_pernet_fini(net);
1702 free_percpu(net->ct.stat);
1703 free_percpu(net->ct.pcpu_lists);
1707 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1709 struct hlist_nulls_head *hash;
1710 unsigned int nr_slots, i;
1713 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1716 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1717 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1719 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1722 sz = nr_slots * sizeof(struct hlist_nulls_head);
1723 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1729 for (i = 0; i < nr_slots; i++)
1730 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1734 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1736 int nf_conntrack_hash_resize(unsigned int hashsize)
1739 unsigned int old_size;
1740 struct hlist_nulls_head *hash, *old_hash;
1741 struct nf_conntrack_tuple_hash *h;
1747 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1751 old_size = nf_conntrack_htable_size;
1752 if (old_size == hashsize) {
1753 nf_ct_free_hashtable(hash, hashsize);
1758 nf_conntrack_all_lock();
1759 write_seqcount_begin(&nf_conntrack_generation);
1761 /* Lookups in the old hash might happen in parallel, which means we
1762 * might get false negatives during connection lookup. New connections
1763 * created because of a false negative won't make it into the hash
1764 * though since that required taking the locks.
1767 for (i = 0; i < nf_conntrack_htable_size; i++) {
1768 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1769 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1770 struct nf_conntrack_tuple_hash, hnnode);
1771 ct = nf_ct_tuplehash_to_ctrack(h);
1772 hlist_nulls_del_rcu(&h->hnnode);
1773 bucket = __hash_conntrack(nf_ct_net(ct),
1774 &h->tuple, hashsize);
1775 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1778 old_size = nf_conntrack_htable_size;
1779 old_hash = nf_conntrack_hash;
1781 nf_conntrack_hash = hash;
1782 nf_conntrack_htable_size = hashsize;
1784 write_seqcount_end(&nf_conntrack_generation);
1785 nf_conntrack_all_unlock();
1789 nf_ct_free_hashtable(old_hash, old_size);
1793 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1795 unsigned int hashsize;
1798 if (current->nsproxy->net_ns != &init_net)
1801 /* On boot, we can set this without any fancy locking. */
1802 if (!nf_conntrack_htable_size)
1803 return param_set_uint(val, kp);
1805 rc = kstrtouint(val, 0, &hashsize);
1809 return nf_conntrack_hash_resize(hashsize);
1811 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1813 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1814 &nf_conntrack_htable_size, 0600);
1816 void nf_ct_untracked_status_or(unsigned long bits)
1820 for_each_possible_cpu(cpu)
1821 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1823 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1825 int nf_conntrack_init_start(void)
1831 seqcount_init(&nf_conntrack_generation);
1833 for (i = 0; i < CONNTRACK_LOCKS; i++)
1834 spin_lock_init(&nf_conntrack_locks[i]);
1836 if (!nf_conntrack_htable_size) {
1837 /* Idea from tcp.c: use 1/16384 of memory.
1838 * On i386: 32MB machine has 512 buckets.
1839 * >= 1GB machines have 16384 buckets.
1840 * >= 4GB machines have 65536 buckets.
1842 nf_conntrack_htable_size
1843 = (((totalram_pages << PAGE_SHIFT) / 16384)
1844 / sizeof(struct hlist_head));
1845 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1846 nf_conntrack_htable_size = 65536;
1847 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1848 nf_conntrack_htable_size = 16384;
1849 if (nf_conntrack_htable_size < 32)
1850 nf_conntrack_htable_size = 32;
1852 /* Use a max. factor of four by default to get the same max as
1853 * with the old struct list_heads. When a table size is given
1854 * we use the old value of 8 to avoid reducing the max.
1859 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1860 if (!nf_conntrack_hash)
1863 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1865 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1866 sizeof(struct nf_conn), 0,
1867 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1868 if (!nf_conntrack_cachep)
1871 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1872 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1875 ret = nf_conntrack_expect_init();
1879 ret = nf_conntrack_acct_init();
1883 ret = nf_conntrack_tstamp_init();
1887 ret = nf_conntrack_ecache_init();
1891 ret = nf_conntrack_timeout_init();
1895 ret = nf_conntrack_helper_init();
1899 ret = nf_conntrack_labels_init();
1903 ret = nf_conntrack_seqadj_init();
1907 ret = nf_conntrack_proto_init();
1911 /* Set up fake conntrack: to never be deleted, not in any hashes */
1912 for_each_possible_cpu(cpu) {
1913 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1914 write_pnet(&ct->ct_net, &init_net);
1915 atomic_set(&ct->ct_general.use, 1);
1917 /* - and look it like as a confirmed connection */
1918 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1920 conntrack_gc_work_init(&conntrack_gc_work);
1921 queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, GC_INTERVAL_MAX);
1926 nf_conntrack_seqadj_fini();
1928 nf_conntrack_labels_fini();
1930 nf_conntrack_helper_fini();
1932 nf_conntrack_timeout_fini();
1934 nf_conntrack_ecache_fini();
1936 nf_conntrack_tstamp_fini();
1938 nf_conntrack_acct_fini();
1940 nf_conntrack_expect_fini();
1942 kmem_cache_destroy(nf_conntrack_cachep);
1944 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1948 void nf_conntrack_init_end(void)
1950 /* For use by REJECT target */
1951 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1952 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1956 * We need to use special "null" values, not used in hash table
1958 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1959 #define DYING_NULLS_VAL ((1<<30)+1)
1960 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1962 int nf_conntrack_init_net(struct net *net)
1967 atomic_set(&net->ct.count, 0);
1969 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1970 if (!net->ct.pcpu_lists)
1973 for_each_possible_cpu(cpu) {
1974 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1976 spin_lock_init(&pcpu->lock);
1977 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1978 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1981 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1983 goto err_pcpu_lists;
1985 ret = nf_conntrack_expect_pernet_init(net);
1988 ret = nf_conntrack_acct_pernet_init(net);
1991 ret = nf_conntrack_tstamp_pernet_init(net);
1994 ret = nf_conntrack_ecache_pernet_init(net);
1997 ret = nf_conntrack_helper_pernet_init(net);
2000 ret = nf_conntrack_proto_pernet_init(net);
2006 nf_conntrack_helper_pernet_fini(net);
2008 nf_conntrack_ecache_pernet_fini(net);
2010 nf_conntrack_tstamp_pernet_fini(net);
2012 nf_conntrack_acct_pernet_fini(net);
2014 nf_conntrack_expect_pernet_fini(net);
2016 free_percpu(net->ct.stat);
2018 free_percpu(net->ct.pcpu_lists);