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 128u
89 /* upper bound of full table scan */
90 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
91 /* desired ratio of entries found to be expired */
92 #define GC_EVICT_RATIO 50u
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 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
355 /* Released via destroy_conntrack() */
356 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
357 const struct nf_conntrack_zone *zone,
360 struct nf_conn *tmpl, *p;
362 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
363 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
368 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
370 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
371 tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
374 tmpl = kzalloc(sizeof(*tmpl), flags);
379 tmpl->status = IPS_TEMPLATE;
380 write_pnet(&tmpl->ct_net, net);
381 nf_ct_zone_add(tmpl, zone);
382 atomic_set(&tmpl->ct_general.use, 0);
386 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
388 void nf_ct_tmpl_free(struct nf_conn *tmpl)
390 nf_ct_ext_destroy(tmpl);
391 nf_ct_ext_free(tmpl);
393 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
394 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
398 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
401 destroy_conntrack(struct nf_conntrack *nfct)
403 struct nf_conn *ct = (struct nf_conn *)nfct;
404 struct nf_conntrack_l4proto *l4proto;
406 pr_debug("destroy_conntrack(%p)\n", ct);
407 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
409 if (unlikely(nf_ct_is_template(ct))) {
414 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
415 if (l4proto->destroy)
416 l4proto->destroy(ct);
421 /* Expectations will have been removed in clean_from_lists,
422 * except TFTP can create an expectation on the first packet,
423 * before connection is in the list, so we need to clean here,
426 nf_ct_remove_expectations(ct);
428 nf_ct_del_from_dying_or_unconfirmed_list(ct);
433 nf_ct_put(ct->master);
435 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
436 nf_conntrack_free(ct);
439 static void nf_ct_delete_from_lists(struct nf_conn *ct)
441 struct net *net = nf_ct_net(ct);
442 unsigned int hash, reply_hash;
443 unsigned int sequence;
445 nf_ct_helper_destroy(ct);
449 sequence = read_seqcount_begin(&nf_conntrack_generation);
450 hash = hash_conntrack(net,
451 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
452 reply_hash = hash_conntrack(net,
453 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
454 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
456 clean_from_lists(ct);
457 nf_conntrack_double_unlock(hash, reply_hash);
459 nf_ct_add_to_dying_list(ct);
464 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
466 struct nf_conn_tstamp *tstamp;
468 if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
471 tstamp = nf_conn_tstamp_find(ct);
472 if (tstamp && tstamp->stop == 0)
473 tstamp->stop = ktime_get_real_ns();
475 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
476 portid, report) < 0) {
477 /* destroy event was not delivered. nf_ct_put will
478 * be done by event cache worker on redelivery.
480 nf_ct_delete_from_lists(ct);
481 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
485 nf_conntrack_ecache_work(nf_ct_net(ct));
486 nf_ct_delete_from_lists(ct);
490 EXPORT_SYMBOL_GPL(nf_ct_delete);
493 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
494 const struct nf_conntrack_tuple *tuple,
495 const struct nf_conntrack_zone *zone,
496 const struct net *net)
498 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
500 /* A conntrack can be recreated with the equal tuple,
501 * so we need to check that the conntrack is confirmed
503 return nf_ct_tuple_equal(tuple, &h->tuple) &&
504 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
505 nf_ct_is_confirmed(ct) &&
506 net_eq(net, nf_ct_net(ct));
509 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
510 static void nf_ct_gc_expired(struct nf_conn *ct)
512 if (!atomic_inc_not_zero(&ct->ct_general.use))
515 if (nf_ct_should_gc(ct))
523 * - Caller must take a reference on returned object
524 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
526 static struct nf_conntrack_tuple_hash *
527 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
528 const struct nf_conntrack_tuple *tuple, u32 hash)
530 struct nf_conntrack_tuple_hash *h;
531 struct hlist_nulls_head *ct_hash;
532 struct hlist_nulls_node *n;
533 unsigned int bucket, hsize;
536 nf_conntrack_get_ht(&ct_hash, &hsize);
537 bucket = reciprocal_scale(hash, hsize);
539 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
542 ct = nf_ct_tuplehash_to_ctrack(h);
543 if (nf_ct_is_expired(ct)) {
544 nf_ct_gc_expired(ct);
548 if (nf_ct_is_dying(ct))
551 if (nf_ct_key_equal(h, tuple, zone, net))
555 * if the nulls value we got at the end of this lookup is
556 * not the expected one, we must restart lookup.
557 * We probably met an item that was moved to another chain.
559 if (get_nulls_value(n) != bucket) {
560 NF_CT_STAT_INC_ATOMIC(net, search_restart);
567 /* Find a connection corresponding to a tuple. */
568 static struct nf_conntrack_tuple_hash *
569 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
570 const struct nf_conntrack_tuple *tuple, u32 hash)
572 struct nf_conntrack_tuple_hash *h;
577 h = ____nf_conntrack_find(net, zone, tuple, hash);
579 ct = nf_ct_tuplehash_to_ctrack(h);
580 if (unlikely(nf_ct_is_dying(ct) ||
581 !atomic_inc_not_zero(&ct->ct_general.use)))
584 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
595 struct nf_conntrack_tuple_hash *
596 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
597 const struct nf_conntrack_tuple *tuple)
599 return __nf_conntrack_find_get(net, zone, tuple,
600 hash_conntrack_raw(tuple, net));
602 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
604 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
606 unsigned int reply_hash)
608 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
609 &nf_conntrack_hash[hash]);
610 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
611 &nf_conntrack_hash[reply_hash]);
615 nf_conntrack_hash_check_insert(struct nf_conn *ct)
617 const struct nf_conntrack_zone *zone;
618 struct net *net = nf_ct_net(ct);
619 unsigned int hash, reply_hash;
620 struct nf_conntrack_tuple_hash *h;
621 struct hlist_nulls_node *n;
622 unsigned int sequence;
624 zone = nf_ct_zone(ct);
628 sequence = read_seqcount_begin(&nf_conntrack_generation);
629 hash = hash_conntrack(net,
630 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
631 reply_hash = hash_conntrack(net,
632 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
633 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
635 /* See if there's one in the list already, including reverse */
636 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
637 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
641 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
642 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
647 /* The caller holds a reference to this object */
648 atomic_set(&ct->ct_general.use, 2);
649 __nf_conntrack_hash_insert(ct, hash, reply_hash);
650 nf_conntrack_double_unlock(hash, reply_hash);
651 NF_CT_STAT_INC(net, insert);
656 nf_conntrack_double_unlock(hash, reply_hash);
657 NF_CT_STAT_INC(net, insert_failed);
661 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
663 static inline void nf_ct_acct_update(struct nf_conn *ct,
664 enum ip_conntrack_info ctinfo,
667 struct nf_conn_acct *acct;
669 acct = nf_conn_acct_find(ct);
671 struct nf_conn_counter *counter = acct->counter;
673 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
674 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
678 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
679 const struct nf_conn *loser_ct)
681 struct nf_conn_acct *acct;
683 acct = nf_conn_acct_find(loser_ct);
685 struct nf_conn_counter *counter = acct->counter;
688 /* u32 should be fine since we must have seen one packet. */
689 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
690 nf_ct_acct_update(ct, ctinfo, bytes);
694 /* Resolve race on insertion if this protocol allows this. */
695 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
696 enum ip_conntrack_info ctinfo,
697 struct nf_conntrack_tuple_hash *h)
699 /* This is the conntrack entry already in hashes that won race. */
700 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
701 struct nf_conntrack_l4proto *l4proto;
703 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
704 if (l4proto->allow_clash &&
706 !nf_ct_is_dying(ct) &&
707 atomic_inc_not_zero(&ct->ct_general.use)) {
708 enum ip_conntrack_info oldinfo;
709 struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo);
711 nf_ct_acct_merge(ct, ctinfo, loser_ct);
712 nf_conntrack_put(&loser_ct->ct_general);
713 nf_ct_set(skb, ct, oldinfo);
716 NF_CT_STAT_INC(net, drop);
720 /* Confirm a connection given skb; places it in hash table */
722 __nf_conntrack_confirm(struct sk_buff *skb)
724 const struct nf_conntrack_zone *zone;
725 unsigned int hash, reply_hash;
726 struct nf_conntrack_tuple_hash *h;
728 struct nf_conn_help *help;
729 struct nf_conn_tstamp *tstamp;
730 struct hlist_nulls_node *n;
731 enum ip_conntrack_info ctinfo;
733 unsigned int sequence;
736 ct = nf_ct_get(skb, &ctinfo);
739 /* ipt_REJECT uses nf_conntrack_attach to attach related
740 ICMP/TCP RST packets in other direction. Actual packet
741 which created connection will be IP_CT_NEW or for an
742 expected connection, IP_CT_RELATED. */
743 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
746 zone = nf_ct_zone(ct);
750 sequence = read_seqcount_begin(&nf_conntrack_generation);
751 /* reuse the hash saved before */
752 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
753 hash = scale_hash(hash);
754 reply_hash = hash_conntrack(net,
755 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
757 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
759 /* We're not in hash table, and we refuse to set up related
760 * connections for unconfirmed conns. But packet copies and
761 * REJECT will give spurious warnings here.
763 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
765 /* No external references means no one else could have
768 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
769 pr_debug("Confirming conntrack %p\n", ct);
770 /* We have to check the DYING flag after unlink to prevent
771 * a race against nf_ct_get_next_corpse() possibly called from
772 * user context, else we insert an already 'dead' hash, blocking
773 * further use of that particular connection -JM.
775 nf_ct_del_from_dying_or_unconfirmed_list(ct);
777 if (unlikely(nf_ct_is_dying(ct))) {
778 nf_ct_add_to_dying_list(ct);
782 /* See if there's one in the list already, including reverse:
783 NAT could have grabbed it without realizing, since we're
784 not in the hash. If there is, we lost race. */
785 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
786 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
790 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
791 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
795 /* Timer relative to confirmation time, not original
796 setting time, otherwise we'd get timer wrap in
797 weird delay cases. */
798 ct->timeout += nfct_time_stamp;
799 atomic_inc(&ct->ct_general.use);
800 ct->status |= IPS_CONFIRMED;
802 /* set conntrack timestamp, if enabled. */
803 tstamp = nf_conn_tstamp_find(ct);
805 if (skb->tstamp == 0)
806 __net_timestamp(skb);
808 tstamp->start = ktime_to_ns(skb->tstamp);
810 /* Since the lookup is lockless, hash insertion must be done after
811 * starting the timer and setting the CONFIRMED bit. The RCU barriers
812 * guarantee that no other CPU can find the conntrack before the above
813 * stores are visible.
815 __nf_conntrack_hash_insert(ct, hash, reply_hash);
816 nf_conntrack_double_unlock(hash, reply_hash);
819 help = nfct_help(ct);
820 if (help && help->helper)
821 nf_conntrack_event_cache(IPCT_HELPER, ct);
823 nf_conntrack_event_cache(master_ct(ct) ?
824 IPCT_RELATED : IPCT_NEW, ct);
828 nf_ct_add_to_dying_list(ct);
829 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
831 nf_conntrack_double_unlock(hash, reply_hash);
832 NF_CT_STAT_INC(net, insert_failed);
836 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
838 /* Returns true if a connection correspondings to the tuple (required
841 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
842 const struct nf_conn *ignored_conntrack)
844 struct net *net = nf_ct_net(ignored_conntrack);
845 const struct nf_conntrack_zone *zone;
846 struct nf_conntrack_tuple_hash *h;
847 struct hlist_nulls_head *ct_hash;
848 unsigned int hash, hsize;
849 struct hlist_nulls_node *n;
852 zone = nf_ct_zone(ignored_conntrack);
856 nf_conntrack_get_ht(&ct_hash, &hsize);
857 hash = __hash_conntrack(net, tuple, hsize);
859 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
860 ct = nf_ct_tuplehash_to_ctrack(h);
862 if (ct == ignored_conntrack)
865 if (nf_ct_is_expired(ct)) {
866 nf_ct_gc_expired(ct);
870 if (nf_ct_key_equal(h, tuple, zone, net)) {
871 NF_CT_STAT_INC_ATOMIC(net, found);
877 if (get_nulls_value(n) != hash) {
878 NF_CT_STAT_INC_ATOMIC(net, search_restart);
886 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
888 #define NF_CT_EVICTION_RANGE 8
890 /* There's a small race here where we may free a just-assured
891 connection. Too bad: we're in trouble anyway. */
892 static unsigned int early_drop_list(struct net *net,
893 struct hlist_nulls_head *head)
895 struct nf_conntrack_tuple_hash *h;
896 struct hlist_nulls_node *n;
897 unsigned int drops = 0;
900 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
901 tmp = nf_ct_tuplehash_to_ctrack(h);
903 if (nf_ct_is_expired(tmp)) {
904 nf_ct_gc_expired(tmp);
908 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
909 !net_eq(nf_ct_net(tmp), net) ||
913 if (!atomic_inc_not_zero(&tmp->ct_general.use))
916 /* kill only if still in same netns -- might have moved due to
917 * SLAB_DESTROY_BY_RCU rules.
919 * We steal the timer reference. If that fails timer has
920 * already fired or someone else deleted it. Just drop ref
921 * and move to next entry.
923 if (net_eq(nf_ct_net(tmp), net) &&
924 nf_ct_is_confirmed(tmp) &&
925 nf_ct_delete(tmp, 0, 0))
934 static noinline int early_drop(struct net *net, unsigned int _hash)
938 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
939 struct hlist_nulls_head *ct_hash;
940 unsigned int hash, hsize, drops;
943 nf_conntrack_get_ht(&ct_hash, &hsize);
944 hash = reciprocal_scale(_hash++, hsize);
946 drops = early_drop_list(net, &ct_hash[hash]);
950 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
958 static void gc_worker(struct work_struct *work)
960 unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
961 unsigned int i, goal, buckets = 0, expired_count = 0;
962 struct conntrack_gc_work *gc_work;
963 unsigned int ratio, scanned = 0;
964 unsigned long next_run;
966 gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
968 goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
969 i = gc_work->last_bucket;
972 struct nf_conntrack_tuple_hash *h;
973 struct hlist_nulls_head *ct_hash;
974 struct hlist_nulls_node *n;
981 nf_conntrack_get_ht(&ct_hash, &hashsz);
985 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
986 tmp = nf_ct_tuplehash_to_ctrack(h);
989 if (nf_ct_is_expired(tmp)) {
990 nf_ct_gc_expired(tmp);
996 /* could check get_nulls_value() here and restart if ct
997 * was moved to another chain. But given gc is best-effort
998 * we will just continue with next hash slot.
1001 cond_resched_rcu_qs();
1002 } while (++buckets < goal);
1004 if (gc_work->exiting)
1008 * Eviction will normally happen from the packet path, and not
1009 * from this gc worker.
1011 * This worker is only here to reap expired entries when system went
1012 * idle after a busy period.
1014 * The heuristics below are supposed to balance conflicting goals:
1016 * 1. Minimize time until we notice a stale entry
1017 * 2. Maximize scan intervals to not waste cycles
1019 * Normally, expire ratio will be close to 0.
1021 * As soon as a sizeable fraction of the entries have expired
1022 * increase scan frequency.
1024 ratio = scanned ? expired_count * 100 / scanned : 0;
1025 if (ratio > GC_EVICT_RATIO) {
1026 gc_work->next_gc_run = min_interval;
1028 unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV;
1030 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0);
1032 gc_work->next_gc_run += min_interval;
1033 if (gc_work->next_gc_run > max)
1034 gc_work->next_gc_run = max;
1037 next_run = gc_work->next_gc_run;
1038 gc_work->last_bucket = i;
1039 queue_delayed_work(system_long_wq, &gc_work->dwork, next_run);
1042 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1044 INIT_DELAYED_WORK(&gc_work->dwork, gc_worker);
1045 gc_work->next_gc_run = HZ;
1046 gc_work->exiting = false;
1049 static struct nf_conn *
1050 __nf_conntrack_alloc(struct net *net,
1051 const struct nf_conntrack_zone *zone,
1052 const struct nf_conntrack_tuple *orig,
1053 const struct nf_conntrack_tuple *repl,
1054 gfp_t gfp, u32 hash)
1058 /* We don't want any race condition at early drop stage */
1059 atomic_inc(&net->ct.count);
1061 if (nf_conntrack_max &&
1062 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1063 if (!early_drop(net, hash)) {
1064 atomic_dec(&net->ct.count);
1065 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1066 return ERR_PTR(-ENOMEM);
1071 * Do not use kmem_cache_zalloc(), as this cache uses
1072 * SLAB_DESTROY_BY_RCU.
1074 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1078 spin_lock_init(&ct->lock);
1079 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1080 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1081 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1082 /* save hash for reusing when confirming */
1083 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1085 write_pnet(&ct->ct_net, net);
1086 memset(&ct->__nfct_init_offset[0], 0,
1087 offsetof(struct nf_conn, proto) -
1088 offsetof(struct nf_conn, __nfct_init_offset[0]));
1090 nf_ct_zone_add(ct, zone);
1092 /* Because we use RCU lookups, we set ct_general.use to zero before
1093 * this is inserted in any list.
1095 atomic_set(&ct->ct_general.use, 0);
1098 atomic_dec(&net->ct.count);
1099 return ERR_PTR(-ENOMEM);
1102 struct nf_conn *nf_conntrack_alloc(struct net *net,
1103 const struct nf_conntrack_zone *zone,
1104 const struct nf_conntrack_tuple *orig,
1105 const struct nf_conntrack_tuple *repl,
1108 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1110 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1112 void nf_conntrack_free(struct nf_conn *ct)
1114 struct net *net = nf_ct_net(ct);
1116 /* A freed object has refcnt == 0, that's
1117 * the golden rule for SLAB_DESTROY_BY_RCU
1119 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
1121 nf_ct_ext_destroy(ct);
1123 kmem_cache_free(nf_conntrack_cachep, ct);
1124 smp_mb__before_atomic();
1125 atomic_dec(&net->ct.count);
1127 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1130 /* Allocate a new conntrack: we return -ENOMEM if classification
1131 failed due to stress. Otherwise it really is unclassifiable. */
1132 static struct nf_conntrack_tuple_hash *
1133 init_conntrack(struct net *net, struct nf_conn *tmpl,
1134 const struct nf_conntrack_tuple *tuple,
1135 struct nf_conntrack_l3proto *l3proto,
1136 struct nf_conntrack_l4proto *l4proto,
1137 struct sk_buff *skb,
1138 unsigned int dataoff, u32 hash)
1141 struct nf_conn_help *help;
1142 struct nf_conntrack_tuple repl_tuple;
1143 struct nf_conntrack_ecache *ecache;
1144 struct nf_conntrack_expect *exp = NULL;
1145 const struct nf_conntrack_zone *zone;
1146 struct nf_conn_timeout *timeout_ext;
1147 struct nf_conntrack_zone tmp;
1148 unsigned int *timeouts;
1150 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1151 pr_debug("Can't invert tuple.\n");
1155 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1156 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1159 return (struct nf_conntrack_tuple_hash *)ct;
1161 if (!nf_ct_add_synproxy(ct, tmpl)) {
1162 nf_conntrack_free(ct);
1163 return ERR_PTR(-ENOMEM);
1166 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1168 timeouts = nf_ct_timeout_data(timeout_ext);
1169 if (unlikely(!timeouts))
1170 timeouts = l4proto->get_timeouts(net);
1172 timeouts = l4proto->get_timeouts(net);
1175 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1176 nf_conntrack_free(ct);
1177 pr_debug("can't track with proto module\n");
1182 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1185 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1186 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1187 nf_ct_labels_ext_add(ct);
1189 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1190 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1191 ecache ? ecache->expmask : 0,
1195 if (net->ct.expect_count) {
1196 spin_lock(&nf_conntrack_expect_lock);
1197 exp = nf_ct_find_expectation(net, zone, tuple);
1199 pr_debug("expectation arrives ct=%p exp=%p\n",
1201 /* Welcome, Mr. Bond. We've been expecting you... */
1202 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1203 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1204 ct->master = exp->master;
1206 help = nf_ct_helper_ext_add(ct, exp->helper,
1209 rcu_assign_pointer(help->helper, exp->helper);
1212 #ifdef CONFIG_NF_CONNTRACK_MARK
1213 ct->mark = exp->master->mark;
1215 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1216 ct->secmark = exp->master->secmark;
1218 NF_CT_STAT_INC(net, expect_new);
1220 spin_unlock(&nf_conntrack_expect_lock);
1223 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1225 /* Now it is inserted into the unconfirmed list, bump refcount */
1226 nf_conntrack_get(&ct->ct_general);
1227 nf_ct_add_to_unconfirmed_list(ct);
1233 exp->expectfn(ct, exp);
1234 nf_ct_expect_put(exp);
1237 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1240 /* On success, returns conntrack ptr, sets skb->_nfct | ctinfo */
1241 static inline struct nf_conn *
1242 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1243 struct sk_buff *skb,
1244 unsigned int dataoff,
1247 struct nf_conntrack_l3proto *l3proto,
1248 struct nf_conntrack_l4proto *l4proto,
1250 enum ip_conntrack_info *ctinfo)
1252 const struct nf_conntrack_zone *zone;
1253 struct nf_conntrack_tuple tuple;
1254 struct nf_conntrack_tuple_hash *h;
1255 struct nf_conntrack_zone tmp;
1259 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1260 dataoff, l3num, protonum, net, &tuple, l3proto,
1262 pr_debug("Can't get tuple\n");
1266 /* look for tuple match */
1267 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1268 hash = hash_conntrack_raw(&tuple, net);
1269 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1271 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1272 skb, dataoff, hash);
1278 ct = nf_ct_tuplehash_to_ctrack(h);
1280 /* It exists; we have (non-exclusive) reference. */
1281 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1282 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1283 /* Please set reply bit if this packet OK */
1286 /* Once we've had two way comms, always ESTABLISHED. */
1287 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1288 pr_debug("normal packet for %p\n", ct);
1289 *ctinfo = IP_CT_ESTABLISHED;
1290 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1291 pr_debug("related packet for %p\n", ct);
1292 *ctinfo = IP_CT_RELATED;
1294 pr_debug("new packet for %p\n", ct);
1295 *ctinfo = IP_CT_NEW;
1299 nf_ct_set(skb, ct, *ctinfo);
1304 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1305 struct sk_buff *skb)
1307 struct nf_conn *ct, *tmpl;
1308 enum ip_conntrack_info ctinfo;
1309 struct nf_conntrack_l3proto *l3proto;
1310 struct nf_conntrack_l4proto *l4proto;
1311 unsigned int *timeouts;
1312 unsigned int dataoff;
1317 tmpl = nf_ct_get(skb, &ctinfo);
1319 /* Previously seen (loopback or untracked)? Ignore. */
1320 if (!nf_ct_is_template(tmpl)) {
1321 NF_CT_STAT_INC_ATOMIC(net, ignore);
1327 /* rcu_read_lock()ed by nf_hook_thresh */
1328 l3proto = __nf_ct_l3proto_find(pf);
1329 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1330 &dataoff, &protonum);
1332 pr_debug("not prepared to track yet or error occurred\n");
1333 NF_CT_STAT_INC_ATOMIC(net, error);
1334 NF_CT_STAT_INC_ATOMIC(net, invalid);
1339 l4proto = __nf_ct_l4proto_find(pf, protonum);
1341 /* It may be an special packet, error, unclean...
1342 * inverse of the return code tells to the netfilter
1343 * core what to do with the packet. */
1344 if (l4proto->error != NULL) {
1345 ret = l4proto->error(net, tmpl, skb, dataoff, pf, hooknum);
1347 NF_CT_STAT_INC_ATOMIC(net, error);
1348 NF_CT_STAT_INC_ATOMIC(net, invalid);
1352 /* ICMP[v6] protocol trackers may assign one conntrack. */
1357 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1358 l3proto, l4proto, &set_reply, &ctinfo);
1360 /* Not valid part of a connection */
1361 NF_CT_STAT_INC_ATOMIC(net, invalid);
1367 /* Too stressed to deal. */
1368 NF_CT_STAT_INC_ATOMIC(net, drop);
1373 NF_CT_ASSERT(skb_nfct(skb));
1375 /* Decide what timeout policy we want to apply to this flow. */
1376 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1378 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1380 /* Invalid: inverse of the return code tells
1381 * the netfilter core what to do */
1382 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1383 nf_conntrack_put(&ct->ct_general);
1385 NF_CT_STAT_INC_ATOMIC(net, invalid);
1386 if (ret == -NF_DROP)
1387 NF_CT_STAT_INC_ATOMIC(net, drop);
1388 /* Special case: TCP tracker reports an attempt to reopen a
1389 * closed/aborted connection. We have to go back and create a
1392 if (ret == -NF_REPEAT)
1398 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1399 nf_conntrack_event_cache(IPCT_REPLY, ct);
1406 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1408 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1409 const struct nf_conntrack_tuple *orig)
1414 ret = nf_ct_invert_tuple(inverse, orig,
1415 __nf_ct_l3proto_find(orig->src.l3num),
1416 __nf_ct_l4proto_find(orig->src.l3num,
1417 orig->dst.protonum));
1421 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1423 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1424 implicitly racy: see __nf_conntrack_confirm */
1425 void nf_conntrack_alter_reply(struct nf_conn *ct,
1426 const struct nf_conntrack_tuple *newreply)
1428 struct nf_conn_help *help = nfct_help(ct);
1430 /* Should be unconfirmed, so not in hash table yet */
1431 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1433 pr_debug("Altering reply tuple of %p to ", ct);
1434 nf_ct_dump_tuple(newreply);
1436 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1437 if (ct->master || (help && !hlist_empty(&help->expectations)))
1441 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1444 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1446 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1447 void __nf_ct_refresh_acct(struct nf_conn *ct,
1448 enum ip_conntrack_info ctinfo,
1449 const struct sk_buff *skb,
1450 unsigned long extra_jiffies,
1455 /* Only update if this is not a fixed timeout */
1456 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1459 /* If not in hash table, timer will not be active yet */
1460 if (nf_ct_is_confirmed(ct))
1461 extra_jiffies += nfct_time_stamp;
1463 ct->timeout = extra_jiffies;
1466 nf_ct_acct_update(ct, ctinfo, skb->len);
1468 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1470 bool nf_ct_kill_acct(struct nf_conn *ct,
1471 enum ip_conntrack_info ctinfo,
1472 const struct sk_buff *skb)
1474 nf_ct_acct_update(ct, ctinfo, skb->len);
1476 return nf_ct_delete(ct, 0, 0);
1478 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1480 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1482 #include <linux/netfilter/nfnetlink.h>
1483 #include <linux/netfilter/nfnetlink_conntrack.h>
1484 #include <linux/mutex.h>
1486 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1487 * in ip_conntrack_core, since we don't want the protocols to autoload
1488 * or depend on ctnetlink */
1489 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1490 const struct nf_conntrack_tuple *tuple)
1492 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1493 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1494 goto nla_put_failure;
1500 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1502 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1503 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1504 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1506 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1508 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1509 struct nf_conntrack_tuple *t)
1511 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1514 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1515 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1519 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1521 int nf_ct_port_nlattr_tuple_size(void)
1523 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1525 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1528 /* Used by ipt_REJECT and ip6t_REJECT. */
1529 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1532 enum ip_conntrack_info ctinfo;
1534 /* This ICMP is in reverse direction to the packet which caused it */
1535 ct = nf_ct_get(skb, &ctinfo);
1536 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1537 ctinfo = IP_CT_RELATED_REPLY;
1539 ctinfo = IP_CT_RELATED;
1541 /* Attach to new skbuff, and increment count */
1542 nf_ct_set(nskb, ct, ctinfo);
1543 nf_conntrack_get(skb_nfct(nskb));
1546 /* Bring out ya dead! */
1547 static struct nf_conn *
1548 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1549 void *data, unsigned int *bucket)
1551 struct nf_conntrack_tuple_hash *h;
1553 struct hlist_nulls_node *n;
1557 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1558 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1560 nf_conntrack_lock(lockp);
1561 if (*bucket < nf_conntrack_htable_size) {
1562 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1563 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1565 ct = nf_ct_tuplehash_to_ctrack(h);
1566 if (net_eq(nf_ct_net(ct), net) &&
1576 for_each_possible_cpu(cpu) {
1577 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1579 spin_lock_bh(&pcpu->lock);
1580 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1581 ct = nf_ct_tuplehash_to_ctrack(h);
1583 set_bit(IPS_DYING_BIT, &ct->status);
1585 spin_unlock_bh(&pcpu->lock);
1590 atomic_inc(&ct->ct_general.use);
1596 void nf_ct_iterate_cleanup(struct net *net,
1597 int (*iter)(struct nf_conn *i, void *data),
1598 void *data, u32 portid, int report)
1601 unsigned int bucket = 0;
1605 if (atomic_read(&net->ct.count) == 0)
1608 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1609 /* Time to push up daises... */
1611 nf_ct_delete(ct, portid, report);
1616 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1618 static int kill_all(struct nf_conn *i, void *data)
1623 void nf_ct_free_hashtable(void *hash, unsigned int size)
1625 if (is_vmalloc_addr(hash))
1628 free_pages((unsigned long)hash,
1629 get_order(sizeof(struct hlist_head) * size));
1631 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1633 static int untrack_refs(void)
1637 for_each_possible_cpu(cpu) {
1638 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1640 cnt += atomic_read(&ct->ct_general.use) - 1;
1645 void nf_conntrack_cleanup_start(void)
1647 conntrack_gc_work.exiting = true;
1648 RCU_INIT_POINTER(ip_ct_attach, NULL);
1651 void nf_conntrack_cleanup_end(void)
1653 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1654 while (untrack_refs() > 0)
1657 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
1658 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1660 nf_conntrack_proto_fini();
1661 nf_conntrack_seqadj_fini();
1662 nf_conntrack_labels_fini();
1663 nf_conntrack_helper_fini();
1664 nf_conntrack_timeout_fini();
1665 nf_conntrack_ecache_fini();
1666 nf_conntrack_tstamp_fini();
1667 nf_conntrack_acct_fini();
1668 nf_conntrack_expect_fini();
1670 kmem_cache_destroy(nf_conntrack_cachep);
1674 * Mishearing the voices in his head, our hero wonders how he's
1675 * supposed to kill the mall.
1677 void nf_conntrack_cleanup_net(struct net *net)
1681 list_add(&net->exit_list, &single);
1682 nf_conntrack_cleanup_net_list(&single);
1685 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1691 * This makes sure all current packets have passed through
1692 * netfilter framework. Roll on, two-stage module
1698 list_for_each_entry(net, net_exit_list, exit_list) {
1699 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1700 if (atomic_read(&net->ct.count) != 0)
1705 goto i_see_dead_people;
1708 list_for_each_entry(net, net_exit_list, exit_list) {
1709 nf_conntrack_proto_pernet_fini(net);
1710 nf_conntrack_helper_pernet_fini(net);
1711 nf_conntrack_ecache_pernet_fini(net);
1712 nf_conntrack_tstamp_pernet_fini(net);
1713 nf_conntrack_acct_pernet_fini(net);
1714 nf_conntrack_expect_pernet_fini(net);
1715 free_percpu(net->ct.stat);
1716 free_percpu(net->ct.pcpu_lists);
1720 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1722 struct hlist_nulls_head *hash;
1723 unsigned int nr_slots, i;
1726 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1729 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1730 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1732 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1735 sz = nr_slots * sizeof(struct hlist_nulls_head);
1736 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1742 for (i = 0; i < nr_slots; i++)
1743 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1747 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1749 int nf_conntrack_hash_resize(unsigned int hashsize)
1752 unsigned int old_size;
1753 struct hlist_nulls_head *hash, *old_hash;
1754 struct nf_conntrack_tuple_hash *h;
1760 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1764 old_size = nf_conntrack_htable_size;
1765 if (old_size == hashsize) {
1766 nf_ct_free_hashtable(hash, hashsize);
1771 nf_conntrack_all_lock();
1772 write_seqcount_begin(&nf_conntrack_generation);
1774 /* Lookups in the old hash might happen in parallel, which means we
1775 * might get false negatives during connection lookup. New connections
1776 * created because of a false negative won't make it into the hash
1777 * though since that required taking the locks.
1780 for (i = 0; i < nf_conntrack_htable_size; i++) {
1781 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1782 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1783 struct nf_conntrack_tuple_hash, hnnode);
1784 ct = nf_ct_tuplehash_to_ctrack(h);
1785 hlist_nulls_del_rcu(&h->hnnode);
1786 bucket = __hash_conntrack(nf_ct_net(ct),
1787 &h->tuple, hashsize);
1788 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1791 old_size = nf_conntrack_htable_size;
1792 old_hash = nf_conntrack_hash;
1794 nf_conntrack_hash = hash;
1795 nf_conntrack_htable_size = hashsize;
1797 write_seqcount_end(&nf_conntrack_generation);
1798 nf_conntrack_all_unlock();
1802 nf_ct_free_hashtable(old_hash, old_size);
1806 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1808 unsigned int hashsize;
1811 if (current->nsproxy->net_ns != &init_net)
1814 /* On boot, we can set this without any fancy locking. */
1815 if (!nf_conntrack_htable_size)
1816 return param_set_uint(val, kp);
1818 rc = kstrtouint(val, 0, &hashsize);
1822 return nf_conntrack_hash_resize(hashsize);
1824 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1826 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1827 &nf_conntrack_htable_size, 0600);
1829 void nf_ct_untracked_status_or(unsigned long bits)
1833 for_each_possible_cpu(cpu)
1834 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1836 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1838 int nf_conntrack_init_start(void)
1844 seqcount_init(&nf_conntrack_generation);
1846 for (i = 0; i < CONNTRACK_LOCKS; i++)
1847 spin_lock_init(&nf_conntrack_locks[i]);
1849 if (!nf_conntrack_htable_size) {
1850 /* Idea from tcp.c: use 1/16384 of memory.
1851 * On i386: 32MB machine has 512 buckets.
1852 * >= 1GB machines have 16384 buckets.
1853 * >= 4GB machines have 65536 buckets.
1855 nf_conntrack_htable_size
1856 = (((totalram_pages << PAGE_SHIFT) / 16384)
1857 / sizeof(struct hlist_head));
1858 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1859 nf_conntrack_htable_size = 65536;
1860 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1861 nf_conntrack_htable_size = 16384;
1862 if (nf_conntrack_htable_size < 32)
1863 nf_conntrack_htable_size = 32;
1865 /* Use a max. factor of four by default to get the same max as
1866 * with the old struct list_heads. When a table size is given
1867 * we use the old value of 8 to avoid reducing the max.
1872 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1873 if (!nf_conntrack_hash)
1876 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1878 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1879 sizeof(struct nf_conn),
1881 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1882 if (!nf_conntrack_cachep)
1885 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1886 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1889 ret = nf_conntrack_expect_init();
1893 ret = nf_conntrack_acct_init();
1897 ret = nf_conntrack_tstamp_init();
1901 ret = nf_conntrack_ecache_init();
1905 ret = nf_conntrack_timeout_init();
1909 ret = nf_conntrack_helper_init();
1913 ret = nf_conntrack_labels_init();
1917 ret = nf_conntrack_seqadj_init();
1921 ret = nf_conntrack_proto_init();
1925 /* Set up fake conntrack: to never be deleted, not in any hashes */
1926 for_each_possible_cpu(cpu) {
1927 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1928 write_pnet(&ct->ct_net, &init_net);
1929 atomic_set(&ct->ct_general.use, 1);
1931 /* - and look it like as a confirmed connection */
1932 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1934 conntrack_gc_work_init(&conntrack_gc_work);
1935 queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, HZ);
1940 nf_conntrack_seqadj_fini();
1942 nf_conntrack_labels_fini();
1944 nf_conntrack_helper_fini();
1946 nf_conntrack_timeout_fini();
1948 nf_conntrack_ecache_fini();
1950 nf_conntrack_tstamp_fini();
1952 nf_conntrack_acct_fini();
1954 nf_conntrack_expect_fini();
1956 kmem_cache_destroy(nf_conntrack_cachep);
1958 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1962 void nf_conntrack_init_end(void)
1964 /* For use by REJECT target */
1965 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1966 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1970 * We need to use special "null" values, not used in hash table
1972 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1973 #define DYING_NULLS_VAL ((1<<30)+1)
1974 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1976 int nf_conntrack_init_net(struct net *net)
1981 atomic_set(&net->ct.count, 0);
1983 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1984 if (!net->ct.pcpu_lists)
1987 for_each_possible_cpu(cpu) {
1988 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1990 spin_lock_init(&pcpu->lock);
1991 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1992 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1995 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1997 goto err_pcpu_lists;
1999 ret = nf_conntrack_expect_pernet_init(net);
2002 ret = nf_conntrack_acct_pernet_init(net);
2005 ret = nf_conntrack_tstamp_pernet_init(net);
2008 ret = nf_conntrack_ecache_pernet_init(net);
2011 ret = nf_conntrack_helper_pernet_init(net);
2014 ret = nf_conntrack_proto_pernet_init(net);
2020 nf_conntrack_helper_pernet_fini(net);
2022 nf_conntrack_ecache_pernet_fini(net);
2024 nf_conntrack_tstamp_pernet_fini(net);
2026 nf_conntrack_acct_pernet_fini(net);
2028 nf_conntrack_expect_pernet_fini(net);
2030 free_percpu(net->ct.stat);
2032 free_percpu(net->ct.pcpu_lists);