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;
81 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
82 static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
83 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
84 static __read_mostly bool nf_conntrack_locks_all;
86 #define GC_MAX_BUCKETS_DIV 64u
87 #define GC_MAX_BUCKETS 8192u
88 #define GC_INTERVAL (5 * HZ)
89 #define GC_MAX_EVICTS 256u
91 static struct conntrack_gc_work conntrack_gc_work;
93 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
96 while (unlikely(nf_conntrack_locks_all)) {
100 * Order the 'nf_conntrack_locks_all' load vs. the
101 * spin_unlock_wait() loads below, to ensure
102 * that 'nf_conntrack_locks_all_lock' is indeed held:
104 smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
105 spin_unlock_wait(&nf_conntrack_locks_all_lock);
109 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
111 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
113 h1 %= CONNTRACK_LOCKS;
114 h2 %= CONNTRACK_LOCKS;
115 spin_unlock(&nf_conntrack_locks[h1]);
117 spin_unlock(&nf_conntrack_locks[h2]);
120 /* return true if we need to recompute hashes (in case hash table was resized) */
121 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
122 unsigned int h2, unsigned int sequence)
124 h1 %= CONNTRACK_LOCKS;
125 h2 %= CONNTRACK_LOCKS;
127 nf_conntrack_lock(&nf_conntrack_locks[h1]);
129 spin_lock_nested(&nf_conntrack_locks[h2],
130 SINGLE_DEPTH_NESTING);
132 nf_conntrack_lock(&nf_conntrack_locks[h2]);
133 spin_lock_nested(&nf_conntrack_locks[h1],
134 SINGLE_DEPTH_NESTING);
136 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
137 nf_conntrack_double_unlock(h1, h2);
143 static void nf_conntrack_all_lock(void)
147 spin_lock(&nf_conntrack_locks_all_lock);
148 nf_conntrack_locks_all = true;
151 * Order the above store of 'nf_conntrack_locks_all' against
152 * the spin_unlock_wait() loads below, such that if
153 * nf_conntrack_lock() observes 'nf_conntrack_locks_all'
154 * we must observe nf_conntrack_locks[] held:
156 smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
158 for (i = 0; i < CONNTRACK_LOCKS; i++) {
159 spin_unlock_wait(&nf_conntrack_locks[i]);
163 static void nf_conntrack_all_unlock(void)
166 * All prior stores must be complete before we clear
167 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
168 * might observe the false value but not the entire
171 smp_store_release(&nf_conntrack_locks_all, false);
172 spin_unlock(&nf_conntrack_locks_all_lock);
175 unsigned int nf_conntrack_htable_size __read_mostly;
176 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
178 unsigned int nf_conntrack_max __read_mostly;
179 seqcount_t nf_conntrack_generation __read_mostly;
181 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
182 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
184 static unsigned int nf_conntrack_hash_rnd __read_mostly;
186 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
187 const struct net *net)
192 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
194 /* The direction must be ignored, so we hash everything up to the
195 * destination ports (which is a multiple of 4) and treat the last
196 * three bytes manually.
198 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
199 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
200 return jhash2((u32 *)tuple, n, seed ^
201 (((__force __u16)tuple->dst.u.all << 16) |
202 tuple->dst.protonum));
205 static u32 scale_hash(u32 hash)
207 return reciprocal_scale(hash, nf_conntrack_htable_size);
210 static u32 __hash_conntrack(const struct net *net,
211 const struct nf_conntrack_tuple *tuple,
214 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
217 static u32 hash_conntrack(const struct net *net,
218 const struct nf_conntrack_tuple *tuple)
220 return scale_hash(hash_conntrack_raw(tuple, net));
224 nf_ct_get_tuple(const struct sk_buff *skb,
226 unsigned int dataoff,
230 struct nf_conntrack_tuple *tuple,
231 const struct nf_conntrack_l3proto *l3proto,
232 const struct nf_conntrack_l4proto *l4proto)
234 memset(tuple, 0, sizeof(*tuple));
236 tuple->src.l3num = l3num;
237 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
240 tuple->dst.protonum = protonum;
241 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
243 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
245 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
247 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
249 struct net *net, struct nf_conntrack_tuple *tuple)
251 struct nf_conntrack_l3proto *l3proto;
252 struct nf_conntrack_l4proto *l4proto;
253 unsigned int protoff;
259 l3proto = __nf_ct_l3proto_find(l3num);
260 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
261 if (ret != NF_ACCEPT) {
266 l4proto = __nf_ct_l4proto_find(l3num, protonum);
268 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
274 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
277 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
278 const struct nf_conntrack_tuple *orig,
279 const struct nf_conntrack_l3proto *l3proto,
280 const struct nf_conntrack_l4proto *l4proto)
282 memset(inverse, 0, sizeof(*inverse));
284 inverse->src.l3num = orig->src.l3num;
285 if (l3proto->invert_tuple(inverse, orig) == 0)
288 inverse->dst.dir = !orig->dst.dir;
290 inverse->dst.protonum = orig->dst.protonum;
291 return l4proto->invert_tuple(inverse, orig);
293 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
296 clean_from_lists(struct nf_conn *ct)
298 pr_debug("clean_from_lists(%p)\n", ct);
299 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
300 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
302 /* Destroy all pending expectations */
303 nf_ct_remove_expectations(ct);
306 /* must be called with local_bh_disable */
307 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
309 struct ct_pcpu *pcpu;
311 /* add this conntrack to the (per cpu) dying list */
312 ct->cpu = smp_processor_id();
313 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
315 spin_lock(&pcpu->lock);
316 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
318 spin_unlock(&pcpu->lock);
321 /* must be called with local_bh_disable */
322 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
324 struct ct_pcpu *pcpu;
326 /* add this conntrack to the (per cpu) unconfirmed list */
327 ct->cpu = smp_processor_id();
328 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
330 spin_lock(&pcpu->lock);
331 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
333 spin_unlock(&pcpu->lock);
336 /* must be called with local_bh_disable */
337 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
339 struct ct_pcpu *pcpu;
341 /* We overload first tuple to link into unconfirmed or dying list.*/
342 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
344 spin_lock(&pcpu->lock);
345 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
346 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
347 spin_unlock(&pcpu->lock);
350 /* Released via destroy_conntrack() */
351 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
352 const struct nf_conntrack_zone *zone,
355 struct nf_conn *tmpl;
357 tmpl = kzalloc(sizeof(*tmpl), flags);
361 tmpl->status = IPS_TEMPLATE;
362 write_pnet(&tmpl->ct_net, net);
363 nf_ct_zone_add(tmpl, zone);
364 atomic_set(&tmpl->ct_general.use, 0);
368 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
370 void nf_ct_tmpl_free(struct nf_conn *tmpl)
372 nf_ct_ext_destroy(tmpl);
373 nf_ct_ext_free(tmpl);
376 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
379 destroy_conntrack(struct nf_conntrack *nfct)
381 struct nf_conn *ct = (struct nf_conn *)nfct;
382 struct net *net = nf_ct_net(ct);
383 struct nf_conntrack_l4proto *l4proto;
385 pr_debug("destroy_conntrack(%p)\n", ct);
386 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
388 if (unlikely(nf_ct_is_template(ct))) {
393 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
394 if (l4proto->destroy)
395 l4proto->destroy(ct);
400 /* Expectations will have been removed in clean_from_lists,
401 * except TFTP can create an expectation on the first packet,
402 * before connection is in the list, so we need to clean here,
405 nf_ct_remove_expectations(ct);
407 nf_ct_del_from_dying_or_unconfirmed_list(ct);
409 NF_CT_STAT_INC(net, delete);
413 nf_ct_put(ct->master);
415 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
416 nf_conntrack_free(ct);
419 static void nf_ct_delete_from_lists(struct nf_conn *ct)
421 struct net *net = nf_ct_net(ct);
422 unsigned int hash, reply_hash;
423 unsigned int sequence;
425 nf_ct_helper_destroy(ct);
429 sequence = read_seqcount_begin(&nf_conntrack_generation);
430 hash = hash_conntrack(net,
431 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
432 reply_hash = hash_conntrack(net,
433 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
434 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
436 clean_from_lists(ct);
437 nf_conntrack_double_unlock(hash, reply_hash);
439 nf_ct_add_to_dying_list(ct);
441 NF_CT_STAT_INC(net, delete_list);
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)) {
533 NF_CT_STAT_INC_ATOMIC(net, found);
536 NF_CT_STAT_INC_ATOMIC(net, searched);
539 * if the nulls value we got at the end of this lookup is
540 * not the expected one, we must restart lookup.
541 * We probably met an item that was moved to another chain.
543 if (get_nulls_value(n) != bucket) {
544 NF_CT_STAT_INC_ATOMIC(net, search_restart);
551 /* Find a connection corresponding to a tuple. */
552 static struct nf_conntrack_tuple_hash *
553 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
554 const struct nf_conntrack_tuple *tuple, u32 hash)
556 struct nf_conntrack_tuple_hash *h;
561 h = ____nf_conntrack_find(net, zone, tuple, hash);
563 ct = nf_ct_tuplehash_to_ctrack(h);
564 if (unlikely(nf_ct_is_dying(ct) ||
565 !atomic_inc_not_zero(&ct->ct_general.use)))
568 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
579 struct nf_conntrack_tuple_hash *
580 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
581 const struct nf_conntrack_tuple *tuple)
583 return __nf_conntrack_find_get(net, zone, tuple,
584 hash_conntrack_raw(tuple, net));
586 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
588 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
590 unsigned int reply_hash)
592 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
593 &nf_conntrack_hash[hash]);
594 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
595 &nf_conntrack_hash[reply_hash]);
599 nf_conntrack_hash_check_insert(struct nf_conn *ct)
601 const struct nf_conntrack_zone *zone;
602 struct net *net = nf_ct_net(ct);
603 unsigned int hash, reply_hash;
604 struct nf_conntrack_tuple_hash *h;
605 struct hlist_nulls_node *n;
606 unsigned int sequence;
608 zone = nf_ct_zone(ct);
612 sequence = read_seqcount_begin(&nf_conntrack_generation);
613 hash = hash_conntrack(net,
614 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
615 reply_hash = hash_conntrack(net,
616 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
617 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
619 /* See if there's one in the list already, including reverse */
620 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
621 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
625 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
626 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
631 /* The caller holds a reference to this object */
632 atomic_set(&ct->ct_general.use, 2);
633 __nf_conntrack_hash_insert(ct, hash, reply_hash);
634 nf_conntrack_double_unlock(hash, reply_hash);
635 NF_CT_STAT_INC(net, insert);
640 nf_conntrack_double_unlock(hash, reply_hash);
641 NF_CT_STAT_INC(net, insert_failed);
645 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
647 static inline void nf_ct_acct_update(struct nf_conn *ct,
648 enum ip_conntrack_info ctinfo,
651 struct nf_conn_acct *acct;
653 acct = nf_conn_acct_find(ct);
655 struct nf_conn_counter *counter = acct->counter;
657 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
658 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
662 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
663 const struct nf_conn *loser_ct)
665 struct nf_conn_acct *acct;
667 acct = nf_conn_acct_find(loser_ct);
669 struct nf_conn_counter *counter = acct->counter;
672 /* u32 should be fine since we must have seen one packet. */
673 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
674 nf_ct_acct_update(ct, ctinfo, bytes);
678 /* Resolve race on insertion if this protocol allows this. */
679 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
680 enum ip_conntrack_info ctinfo,
681 struct nf_conntrack_tuple_hash *h)
683 /* This is the conntrack entry already in hashes that won race. */
684 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
685 struct nf_conntrack_l4proto *l4proto;
687 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
688 if (l4proto->allow_clash &&
690 !nf_ct_is_dying(ct) &&
691 atomic_inc_not_zero(&ct->ct_general.use)) {
692 nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct);
693 nf_conntrack_put(skb->nfct);
694 /* Assign conntrack already in hashes to this skbuff. Don't
695 * modify skb->nfctinfo to ensure consistent stateful filtering.
697 skb->nfct = &ct->ct_general;
700 NF_CT_STAT_INC(net, drop);
704 /* Confirm a connection given skb; places it in hash table */
706 __nf_conntrack_confirm(struct sk_buff *skb)
708 const struct nf_conntrack_zone *zone;
709 unsigned int hash, reply_hash;
710 struct nf_conntrack_tuple_hash *h;
712 struct nf_conn_help *help;
713 struct nf_conn_tstamp *tstamp;
714 struct hlist_nulls_node *n;
715 enum ip_conntrack_info ctinfo;
717 unsigned int sequence;
720 ct = nf_ct_get(skb, &ctinfo);
723 /* ipt_REJECT uses nf_conntrack_attach to attach related
724 ICMP/TCP RST packets in other direction. Actual packet
725 which created connection will be IP_CT_NEW or for an
726 expected connection, IP_CT_RELATED. */
727 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
730 zone = nf_ct_zone(ct);
734 sequence = read_seqcount_begin(&nf_conntrack_generation);
735 /* reuse the hash saved before */
736 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
737 hash = scale_hash(hash);
738 reply_hash = hash_conntrack(net,
739 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
741 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
743 /* We're not in hash table, and we refuse to set up related
744 * connections for unconfirmed conns. But packet copies and
745 * REJECT will give spurious warnings here.
747 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
749 /* No external references means no one else could have
752 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
753 pr_debug("Confirming conntrack %p\n", ct);
754 /* We have to check the DYING flag after unlink to prevent
755 * a race against nf_ct_get_next_corpse() possibly called from
756 * user context, else we insert an already 'dead' hash, blocking
757 * further use of that particular connection -JM.
759 nf_ct_del_from_dying_or_unconfirmed_list(ct);
761 if (unlikely(nf_ct_is_dying(ct))) {
762 nf_ct_add_to_dying_list(ct);
766 /* See if there's one in the list already, including reverse:
767 NAT could have grabbed it without realizing, since we're
768 not in the hash. If there is, we lost race. */
769 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
770 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
774 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
775 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
779 /* Timer relative to confirmation time, not original
780 setting time, otherwise we'd get timer wrap in
781 weird delay cases. */
782 ct->timeout += nfct_time_stamp;
783 atomic_inc(&ct->ct_general.use);
784 ct->status |= IPS_CONFIRMED;
786 /* set conntrack timestamp, if enabled. */
787 tstamp = nf_conn_tstamp_find(ct);
789 if (skb->tstamp.tv64 == 0)
790 __net_timestamp(skb);
792 tstamp->start = ktime_to_ns(skb->tstamp);
794 /* Since the lookup is lockless, hash insertion must be done after
795 * starting the timer and setting the CONFIRMED bit. The RCU barriers
796 * guarantee that no other CPU can find the conntrack before the above
797 * stores are visible.
799 __nf_conntrack_hash_insert(ct, hash, reply_hash);
800 nf_conntrack_double_unlock(hash, reply_hash);
801 NF_CT_STAT_INC(net, insert);
804 help = nfct_help(ct);
805 if (help && help->helper)
806 nf_conntrack_event_cache(IPCT_HELPER, ct);
808 nf_conntrack_event_cache(master_ct(ct) ?
809 IPCT_RELATED : IPCT_NEW, ct);
813 nf_ct_add_to_dying_list(ct);
814 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
816 nf_conntrack_double_unlock(hash, reply_hash);
817 NF_CT_STAT_INC(net, insert_failed);
821 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
823 /* Returns true if a connection correspondings to the tuple (required
826 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
827 const struct nf_conn *ignored_conntrack)
829 struct net *net = nf_ct_net(ignored_conntrack);
830 const struct nf_conntrack_zone *zone;
831 struct nf_conntrack_tuple_hash *h;
832 struct hlist_nulls_head *ct_hash;
833 unsigned int hash, hsize;
834 struct hlist_nulls_node *n;
837 zone = nf_ct_zone(ignored_conntrack);
841 nf_conntrack_get_ht(&ct_hash, &hsize);
842 hash = __hash_conntrack(net, tuple, hsize);
844 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
845 ct = nf_ct_tuplehash_to_ctrack(h);
847 if (ct == ignored_conntrack)
850 if (nf_ct_is_expired(ct)) {
851 nf_ct_gc_expired(ct);
855 if (nf_ct_key_equal(h, tuple, zone, net)) {
856 NF_CT_STAT_INC_ATOMIC(net, found);
860 NF_CT_STAT_INC_ATOMIC(net, searched);
863 if (get_nulls_value(n) != hash) {
864 NF_CT_STAT_INC_ATOMIC(net, search_restart);
872 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
874 #define NF_CT_EVICTION_RANGE 8
876 /* There's a small race here where we may free a just-assured
877 connection. Too bad: we're in trouble anyway. */
878 static unsigned int early_drop_list(struct net *net,
879 struct hlist_nulls_head *head)
881 struct nf_conntrack_tuple_hash *h;
882 struct hlist_nulls_node *n;
883 unsigned int drops = 0;
886 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
887 tmp = nf_ct_tuplehash_to_ctrack(h);
889 if (nf_ct_is_expired(tmp)) {
890 nf_ct_gc_expired(tmp);
894 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
895 !net_eq(nf_ct_net(tmp), net) ||
899 if (!atomic_inc_not_zero(&tmp->ct_general.use))
902 /* kill only if still in same netns -- might have moved due to
903 * SLAB_DESTROY_BY_RCU rules.
905 * We steal the timer reference. If that fails timer has
906 * already fired or someone else deleted it. Just drop ref
907 * and move to next entry.
909 if (net_eq(nf_ct_net(tmp), net) &&
910 nf_ct_is_confirmed(tmp) &&
911 nf_ct_delete(tmp, 0, 0))
920 static noinline int early_drop(struct net *net, unsigned int _hash)
924 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
925 struct hlist_nulls_head *ct_hash;
926 unsigned int hash, hsize, drops;
929 nf_conntrack_get_ht(&ct_hash, &hsize);
930 hash = reciprocal_scale(_hash++, hsize);
932 drops = early_drop_list(net, &ct_hash[hash]);
936 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
944 static void gc_worker(struct work_struct *work)
946 unsigned int i, goal, buckets = 0, expired_count = 0;
947 unsigned long next_run = GC_INTERVAL;
948 unsigned int ratio, scanned = 0;
949 struct conntrack_gc_work *gc_work;
951 gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
953 goal = min(nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV, GC_MAX_BUCKETS);
954 i = gc_work->last_bucket;
957 struct nf_conntrack_tuple_hash *h;
958 struct hlist_nulls_head *ct_hash;
959 struct hlist_nulls_node *n;
966 nf_conntrack_get_ht(&ct_hash, &hashsz);
970 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
971 tmp = nf_ct_tuplehash_to_ctrack(h);
974 if (nf_ct_is_expired(tmp)) {
975 nf_ct_gc_expired(tmp);
981 /* could check get_nulls_value() here and restart if ct
982 * was moved to another chain. But given gc is best-effort
983 * we will just continue with next hash slot.
986 cond_resched_rcu_qs();
987 } while (++buckets < goal &&
988 expired_count < GC_MAX_EVICTS);
990 if (gc_work->exiting)
993 ratio = scanned ? expired_count * 100 / scanned : 0;
997 gc_work->last_bucket = i;
998 schedule_delayed_work(&gc_work->dwork, next_run);
1001 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1003 INIT_DELAYED_WORK(&gc_work->dwork, gc_worker);
1004 gc_work->exiting = false;
1007 static struct nf_conn *
1008 __nf_conntrack_alloc(struct net *net,
1009 const struct nf_conntrack_zone *zone,
1010 const struct nf_conntrack_tuple *orig,
1011 const struct nf_conntrack_tuple *repl,
1012 gfp_t gfp, u32 hash)
1016 /* We don't want any race condition at early drop stage */
1017 atomic_inc(&net->ct.count);
1019 if (nf_conntrack_max &&
1020 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1021 if (!early_drop(net, hash)) {
1022 atomic_dec(&net->ct.count);
1023 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1024 return ERR_PTR(-ENOMEM);
1029 * Do not use kmem_cache_zalloc(), as this cache uses
1030 * SLAB_DESTROY_BY_RCU.
1032 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1036 spin_lock_init(&ct->lock);
1037 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1038 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1039 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1040 /* save hash for reusing when confirming */
1041 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1043 write_pnet(&ct->ct_net, net);
1044 memset(&ct->__nfct_init_offset[0], 0,
1045 offsetof(struct nf_conn, proto) -
1046 offsetof(struct nf_conn, __nfct_init_offset[0]));
1048 nf_ct_zone_add(ct, zone);
1050 /* Because we use RCU lookups, we set ct_general.use to zero before
1051 * this is inserted in any list.
1053 atomic_set(&ct->ct_general.use, 0);
1056 atomic_dec(&net->ct.count);
1057 return ERR_PTR(-ENOMEM);
1060 struct nf_conn *nf_conntrack_alloc(struct net *net,
1061 const struct nf_conntrack_zone *zone,
1062 const struct nf_conntrack_tuple *orig,
1063 const struct nf_conntrack_tuple *repl,
1066 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1068 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1070 void nf_conntrack_free(struct nf_conn *ct)
1072 struct net *net = nf_ct_net(ct);
1074 /* A freed object has refcnt == 0, that's
1075 * the golden rule for SLAB_DESTROY_BY_RCU
1077 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
1079 nf_ct_ext_destroy(ct);
1081 kmem_cache_free(nf_conntrack_cachep, ct);
1082 smp_mb__before_atomic();
1083 atomic_dec(&net->ct.count);
1085 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1088 /* Allocate a new conntrack: we return -ENOMEM if classification
1089 failed due to stress. Otherwise it really is unclassifiable. */
1090 static struct nf_conntrack_tuple_hash *
1091 init_conntrack(struct net *net, struct nf_conn *tmpl,
1092 const struct nf_conntrack_tuple *tuple,
1093 struct nf_conntrack_l3proto *l3proto,
1094 struct nf_conntrack_l4proto *l4proto,
1095 struct sk_buff *skb,
1096 unsigned int dataoff, u32 hash)
1099 struct nf_conn_help *help;
1100 struct nf_conntrack_tuple repl_tuple;
1101 struct nf_conntrack_ecache *ecache;
1102 struct nf_conntrack_expect *exp = NULL;
1103 const struct nf_conntrack_zone *zone;
1104 struct nf_conn_timeout *timeout_ext;
1105 struct nf_conntrack_zone tmp;
1106 unsigned int *timeouts;
1108 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1109 pr_debug("Can't invert tuple.\n");
1113 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1114 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1117 return (struct nf_conntrack_tuple_hash *)ct;
1119 if (!nf_ct_add_synproxy(ct, tmpl)) {
1120 nf_conntrack_free(ct);
1121 return ERR_PTR(-ENOMEM);
1124 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1126 timeouts = nf_ct_timeout_data(timeout_ext);
1127 if (unlikely(!timeouts))
1128 timeouts = l4proto->get_timeouts(net);
1130 timeouts = l4proto->get_timeouts(net);
1133 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1134 nf_conntrack_free(ct);
1135 pr_debug("can't track with proto module\n");
1140 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1143 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1144 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1145 nf_ct_labels_ext_add(ct);
1147 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1148 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1149 ecache ? ecache->expmask : 0,
1153 if (net->ct.expect_count) {
1154 spin_lock(&nf_conntrack_expect_lock);
1155 exp = nf_ct_find_expectation(net, zone, tuple);
1157 pr_debug("expectation arrives ct=%p exp=%p\n",
1159 /* Welcome, Mr. Bond. We've been expecting you... */
1160 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1161 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1162 ct->master = exp->master;
1164 help = nf_ct_helper_ext_add(ct, exp->helper,
1167 rcu_assign_pointer(help->helper, exp->helper);
1170 #ifdef CONFIG_NF_CONNTRACK_MARK
1171 ct->mark = exp->master->mark;
1173 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1174 ct->secmark = exp->master->secmark;
1176 NF_CT_STAT_INC(net, expect_new);
1178 spin_unlock(&nf_conntrack_expect_lock);
1181 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1182 NF_CT_STAT_INC(net, new);
1185 /* Now it is inserted into the unconfirmed list, bump refcount */
1186 nf_conntrack_get(&ct->ct_general);
1187 nf_ct_add_to_unconfirmed_list(ct);
1193 exp->expectfn(ct, exp);
1194 nf_ct_expect_put(exp);
1197 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1200 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1201 static inline struct nf_conn *
1202 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1203 struct sk_buff *skb,
1204 unsigned int dataoff,
1207 struct nf_conntrack_l3proto *l3proto,
1208 struct nf_conntrack_l4proto *l4proto,
1210 enum ip_conntrack_info *ctinfo)
1212 const struct nf_conntrack_zone *zone;
1213 struct nf_conntrack_tuple tuple;
1214 struct nf_conntrack_tuple_hash *h;
1215 struct nf_conntrack_zone tmp;
1219 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1220 dataoff, l3num, protonum, net, &tuple, l3proto,
1222 pr_debug("Can't get tuple\n");
1226 /* look for tuple match */
1227 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1228 hash = hash_conntrack_raw(&tuple, net);
1229 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1231 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1232 skb, dataoff, hash);
1238 ct = nf_ct_tuplehash_to_ctrack(h);
1240 /* It exists; we have (non-exclusive) reference. */
1241 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1242 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1243 /* Please set reply bit if this packet OK */
1246 /* Once we've had two way comms, always ESTABLISHED. */
1247 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1248 pr_debug("normal packet for %p\n", ct);
1249 *ctinfo = IP_CT_ESTABLISHED;
1250 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1251 pr_debug("related packet for %p\n", ct);
1252 *ctinfo = IP_CT_RELATED;
1254 pr_debug("new packet for %p\n", ct);
1255 *ctinfo = IP_CT_NEW;
1259 skb->nfct = &ct->ct_general;
1260 skb->nfctinfo = *ctinfo;
1265 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1266 struct sk_buff *skb)
1268 struct nf_conn *ct, *tmpl = NULL;
1269 enum ip_conntrack_info ctinfo;
1270 struct nf_conntrack_l3proto *l3proto;
1271 struct nf_conntrack_l4proto *l4proto;
1272 unsigned int *timeouts;
1273 unsigned int dataoff;
1279 /* Previously seen (loopback or untracked)? Ignore. */
1280 tmpl = (struct nf_conn *)skb->nfct;
1281 if (!nf_ct_is_template(tmpl)) {
1282 NF_CT_STAT_INC_ATOMIC(net, ignore);
1288 /* rcu_read_lock()ed by nf_hook_slow */
1289 l3proto = __nf_ct_l3proto_find(pf);
1290 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1291 &dataoff, &protonum);
1293 pr_debug("not prepared to track yet or error occurred\n");
1294 NF_CT_STAT_INC_ATOMIC(net, error);
1295 NF_CT_STAT_INC_ATOMIC(net, invalid);
1300 l4proto = __nf_ct_l4proto_find(pf, protonum);
1302 /* It may be an special packet, error, unclean...
1303 * inverse of the return code tells to the netfilter
1304 * core what to do with the packet. */
1305 if (l4proto->error != NULL) {
1306 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1309 NF_CT_STAT_INC_ATOMIC(net, error);
1310 NF_CT_STAT_INC_ATOMIC(net, invalid);
1314 /* ICMP[v6] protocol trackers may assign one conntrack. */
1319 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1320 l3proto, l4proto, &set_reply, &ctinfo);
1322 /* Not valid part of a connection */
1323 NF_CT_STAT_INC_ATOMIC(net, invalid);
1329 /* Too stressed to deal. */
1330 NF_CT_STAT_INC_ATOMIC(net, drop);
1335 NF_CT_ASSERT(skb->nfct);
1337 /* Decide what timeout policy we want to apply to this flow. */
1338 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1340 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1342 /* Invalid: inverse of the return code tells
1343 * the netfilter core what to do */
1344 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1345 nf_conntrack_put(skb->nfct);
1347 NF_CT_STAT_INC_ATOMIC(net, invalid);
1348 if (ret == -NF_DROP)
1349 NF_CT_STAT_INC_ATOMIC(net, drop);
1354 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1355 nf_conntrack_event_cache(IPCT_REPLY, ct);
1358 /* Special case: we have to repeat this hook, assign the
1359 * template again to this packet. We assume that this packet
1360 * has no conntrack assigned. This is used by nf_ct_tcp. */
1361 if (ret == NF_REPEAT)
1362 skb->nfct = (struct nf_conntrack *)tmpl;
1369 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1371 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1372 const struct nf_conntrack_tuple *orig)
1377 ret = nf_ct_invert_tuple(inverse, orig,
1378 __nf_ct_l3proto_find(orig->src.l3num),
1379 __nf_ct_l4proto_find(orig->src.l3num,
1380 orig->dst.protonum));
1384 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1386 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1387 implicitly racy: see __nf_conntrack_confirm */
1388 void nf_conntrack_alter_reply(struct nf_conn *ct,
1389 const struct nf_conntrack_tuple *newreply)
1391 struct nf_conn_help *help = nfct_help(ct);
1393 /* Should be unconfirmed, so not in hash table yet */
1394 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1396 pr_debug("Altering reply tuple of %p to ", ct);
1397 nf_ct_dump_tuple(newreply);
1399 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1400 if (ct->master || (help && !hlist_empty(&help->expectations)))
1404 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1407 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1409 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1410 void __nf_ct_refresh_acct(struct nf_conn *ct,
1411 enum ip_conntrack_info ctinfo,
1412 const struct sk_buff *skb,
1413 unsigned long extra_jiffies,
1418 /* Only update if this is not a fixed timeout */
1419 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1422 /* If not in hash table, timer will not be active yet */
1423 if (nf_ct_is_confirmed(ct))
1424 extra_jiffies += nfct_time_stamp;
1426 ct->timeout = extra_jiffies;
1429 nf_ct_acct_update(ct, ctinfo, skb->len);
1431 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1433 bool nf_ct_kill_acct(struct nf_conn *ct,
1434 enum ip_conntrack_info ctinfo,
1435 const struct sk_buff *skb)
1437 nf_ct_acct_update(ct, ctinfo, skb->len);
1439 return nf_ct_delete(ct, 0, 0);
1441 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1443 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1445 #include <linux/netfilter/nfnetlink.h>
1446 #include <linux/netfilter/nfnetlink_conntrack.h>
1447 #include <linux/mutex.h>
1449 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1450 * in ip_conntrack_core, since we don't want the protocols to autoload
1451 * or depend on ctnetlink */
1452 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1453 const struct nf_conntrack_tuple *tuple)
1455 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1456 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1457 goto nla_put_failure;
1463 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1465 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1466 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1467 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1469 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1471 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1472 struct nf_conntrack_tuple *t)
1474 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1477 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1478 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1482 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1484 int nf_ct_port_nlattr_tuple_size(void)
1486 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1488 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1491 /* Used by ipt_REJECT and ip6t_REJECT. */
1492 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1495 enum ip_conntrack_info ctinfo;
1497 /* This ICMP is in reverse direction to the packet which caused it */
1498 ct = nf_ct_get(skb, &ctinfo);
1499 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1500 ctinfo = IP_CT_RELATED_REPLY;
1502 ctinfo = IP_CT_RELATED;
1504 /* Attach to new skbuff, and increment count */
1505 nskb->nfct = &ct->ct_general;
1506 nskb->nfctinfo = ctinfo;
1507 nf_conntrack_get(nskb->nfct);
1510 /* Bring out ya dead! */
1511 static struct nf_conn *
1512 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1513 void *data, unsigned int *bucket)
1515 struct nf_conntrack_tuple_hash *h;
1517 struct hlist_nulls_node *n;
1521 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1522 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1524 nf_conntrack_lock(lockp);
1525 if (*bucket < nf_conntrack_htable_size) {
1526 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1527 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1529 ct = nf_ct_tuplehash_to_ctrack(h);
1530 if (net_eq(nf_ct_net(ct), net) &&
1540 for_each_possible_cpu(cpu) {
1541 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1543 spin_lock_bh(&pcpu->lock);
1544 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1545 ct = nf_ct_tuplehash_to_ctrack(h);
1547 set_bit(IPS_DYING_BIT, &ct->status);
1549 spin_unlock_bh(&pcpu->lock);
1554 atomic_inc(&ct->ct_general.use);
1560 void nf_ct_iterate_cleanup(struct net *net,
1561 int (*iter)(struct nf_conn *i, void *data),
1562 void *data, u32 portid, int report)
1565 unsigned int bucket = 0;
1569 if (atomic_read(&net->ct.count) == 0)
1572 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1573 /* Time to push up daises... */
1575 nf_ct_delete(ct, portid, report);
1580 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1582 static int kill_all(struct nf_conn *i, void *data)
1587 void nf_ct_free_hashtable(void *hash, unsigned int size)
1589 if (is_vmalloc_addr(hash))
1592 free_pages((unsigned long)hash,
1593 get_order(sizeof(struct hlist_head) * size));
1595 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1597 static int untrack_refs(void)
1601 for_each_possible_cpu(cpu) {
1602 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1604 cnt += atomic_read(&ct->ct_general.use) - 1;
1609 void nf_conntrack_cleanup_start(void)
1611 conntrack_gc_work.exiting = true;
1612 RCU_INIT_POINTER(ip_ct_attach, NULL);
1615 void nf_conntrack_cleanup_end(void)
1617 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1618 while (untrack_refs() > 0)
1621 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
1622 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1624 nf_conntrack_proto_fini();
1625 nf_conntrack_seqadj_fini();
1626 nf_conntrack_labels_fini();
1627 nf_conntrack_helper_fini();
1628 nf_conntrack_timeout_fini();
1629 nf_conntrack_ecache_fini();
1630 nf_conntrack_tstamp_fini();
1631 nf_conntrack_acct_fini();
1632 nf_conntrack_expect_fini();
1634 kmem_cache_destroy(nf_conntrack_cachep);
1638 * Mishearing the voices in his head, our hero wonders how he's
1639 * supposed to kill the mall.
1641 void nf_conntrack_cleanup_net(struct net *net)
1645 list_add(&net->exit_list, &single);
1646 nf_conntrack_cleanup_net_list(&single);
1649 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1655 * This makes sure all current packets have passed through
1656 * netfilter framework. Roll on, two-stage module
1662 list_for_each_entry(net, net_exit_list, exit_list) {
1663 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1664 if (atomic_read(&net->ct.count) != 0)
1669 goto i_see_dead_people;
1672 list_for_each_entry(net, net_exit_list, exit_list) {
1673 nf_conntrack_proto_pernet_fini(net);
1674 nf_conntrack_helper_pernet_fini(net);
1675 nf_conntrack_ecache_pernet_fini(net);
1676 nf_conntrack_tstamp_pernet_fini(net);
1677 nf_conntrack_acct_pernet_fini(net);
1678 nf_conntrack_expect_pernet_fini(net);
1679 free_percpu(net->ct.stat);
1680 free_percpu(net->ct.pcpu_lists);
1684 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1686 struct hlist_nulls_head *hash;
1687 unsigned int nr_slots, i;
1690 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1693 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1694 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1696 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1699 sz = nr_slots * sizeof(struct hlist_nulls_head);
1700 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1706 for (i = 0; i < nr_slots; i++)
1707 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1711 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1713 int nf_conntrack_hash_resize(unsigned int hashsize)
1716 unsigned int old_size;
1717 struct hlist_nulls_head *hash, *old_hash;
1718 struct nf_conntrack_tuple_hash *h;
1724 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1728 old_size = nf_conntrack_htable_size;
1729 if (old_size == hashsize) {
1730 nf_ct_free_hashtable(hash, hashsize);
1735 nf_conntrack_all_lock();
1736 write_seqcount_begin(&nf_conntrack_generation);
1738 /* Lookups in the old hash might happen in parallel, which means we
1739 * might get false negatives during connection lookup. New connections
1740 * created because of a false negative won't make it into the hash
1741 * though since that required taking the locks.
1744 for (i = 0; i < nf_conntrack_htable_size; i++) {
1745 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1746 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1747 struct nf_conntrack_tuple_hash, hnnode);
1748 ct = nf_ct_tuplehash_to_ctrack(h);
1749 hlist_nulls_del_rcu(&h->hnnode);
1750 bucket = __hash_conntrack(nf_ct_net(ct),
1751 &h->tuple, hashsize);
1752 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1755 old_size = nf_conntrack_htable_size;
1756 old_hash = nf_conntrack_hash;
1758 nf_conntrack_hash = hash;
1759 nf_conntrack_htable_size = hashsize;
1761 write_seqcount_end(&nf_conntrack_generation);
1762 nf_conntrack_all_unlock();
1766 nf_ct_free_hashtable(old_hash, old_size);
1770 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1772 unsigned int hashsize;
1775 if (current->nsproxy->net_ns != &init_net)
1778 /* On boot, we can set this without any fancy locking. */
1779 if (!nf_conntrack_htable_size)
1780 return param_set_uint(val, kp);
1782 rc = kstrtouint(val, 0, &hashsize);
1786 return nf_conntrack_hash_resize(hashsize);
1788 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1790 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1791 &nf_conntrack_htable_size, 0600);
1793 void nf_ct_untracked_status_or(unsigned long bits)
1797 for_each_possible_cpu(cpu)
1798 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1800 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1802 int nf_conntrack_init_start(void)
1808 seqcount_init(&nf_conntrack_generation);
1810 for (i = 0; i < CONNTRACK_LOCKS; i++)
1811 spin_lock_init(&nf_conntrack_locks[i]);
1813 if (!nf_conntrack_htable_size) {
1814 /* Idea from tcp.c: use 1/16384 of memory.
1815 * On i386: 32MB machine has 512 buckets.
1816 * >= 1GB machines have 16384 buckets.
1817 * >= 4GB machines have 65536 buckets.
1819 nf_conntrack_htable_size
1820 = (((totalram_pages << PAGE_SHIFT) / 16384)
1821 / sizeof(struct hlist_head));
1822 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1823 nf_conntrack_htable_size = 65536;
1824 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1825 nf_conntrack_htable_size = 16384;
1826 if (nf_conntrack_htable_size < 32)
1827 nf_conntrack_htable_size = 32;
1829 /* Use a max. factor of four by default to get the same max as
1830 * with the old struct list_heads. When a table size is given
1831 * we use the old value of 8 to avoid reducing the max.
1836 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1837 if (!nf_conntrack_hash)
1840 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1842 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1843 sizeof(struct nf_conn), 0,
1844 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1845 if (!nf_conntrack_cachep)
1848 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1849 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1852 ret = nf_conntrack_expect_init();
1856 ret = nf_conntrack_acct_init();
1860 ret = nf_conntrack_tstamp_init();
1864 ret = nf_conntrack_ecache_init();
1868 ret = nf_conntrack_timeout_init();
1872 ret = nf_conntrack_helper_init();
1876 ret = nf_conntrack_labels_init();
1880 ret = nf_conntrack_seqadj_init();
1884 ret = nf_conntrack_proto_init();
1888 /* Set up fake conntrack: to never be deleted, not in any hashes */
1889 for_each_possible_cpu(cpu) {
1890 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1891 write_pnet(&ct->ct_net, &init_net);
1892 atomic_set(&ct->ct_general.use, 1);
1894 /* - and look it like as a confirmed connection */
1895 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1897 conntrack_gc_work_init(&conntrack_gc_work);
1898 schedule_delayed_work(&conntrack_gc_work.dwork, GC_INTERVAL);
1903 nf_conntrack_seqadj_fini();
1905 nf_conntrack_labels_fini();
1907 nf_conntrack_helper_fini();
1909 nf_conntrack_timeout_fini();
1911 nf_conntrack_ecache_fini();
1913 nf_conntrack_tstamp_fini();
1915 nf_conntrack_acct_fini();
1917 nf_conntrack_expect_fini();
1919 kmem_cache_destroy(nf_conntrack_cachep);
1921 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1925 void nf_conntrack_init_end(void)
1927 /* For use by REJECT target */
1928 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1929 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1933 * We need to use special "null" values, not used in hash table
1935 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1936 #define DYING_NULLS_VAL ((1<<30)+1)
1937 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1939 int nf_conntrack_init_net(struct net *net)
1944 atomic_set(&net->ct.count, 0);
1946 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1947 if (!net->ct.pcpu_lists)
1950 for_each_possible_cpu(cpu) {
1951 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1953 spin_lock_init(&pcpu->lock);
1954 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1955 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1958 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1960 goto err_pcpu_lists;
1962 ret = nf_conntrack_expect_pernet_init(net);
1965 ret = nf_conntrack_acct_pernet_init(net);
1968 ret = nf_conntrack_tstamp_pernet_init(net);
1971 ret = nf_conntrack_ecache_pernet_init(net);
1974 ret = nf_conntrack_helper_pernet_init(net);
1977 ret = nf_conntrack_proto_pernet_init(net);
1983 nf_conntrack_helper_pernet_fini(net);
1985 nf_conntrack_ecache_pernet_fini(net);
1987 nf_conntrack_tstamp_pernet_fini(net);
1989 nf_conntrack_acct_pernet_fini(net);
1991 nf_conntrack_expect_pernet_fini(net);
1993 free_percpu(net->ct.stat);
1995 free_percpu(net->ct.pcpu_lists);