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 #include "nf_internals.h"
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;
83 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
84 static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
85 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
86 static __read_mostly bool nf_conntrack_locks_all;
88 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
89 #define GC_MAX_BUCKETS_DIV 128u
90 /* upper bound of full table scan */
91 #define GC_MAX_SCAN_JIFFIES (16u * HZ)
92 /* desired ratio of entries found to be expired */
93 #define GC_EVICT_RATIO 50u
95 static struct conntrack_gc_work conntrack_gc_work;
97 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
99 /* 1) Acquire the lock */
102 /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
103 * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
105 if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
108 /* fast path failed, unlock */
111 /* Slow path 1) get global lock */
112 spin_lock(&nf_conntrack_locks_all_lock);
114 /* Slow path 2) get the lock we want */
117 /* Slow path 3) release the global lock */
118 spin_unlock(&nf_conntrack_locks_all_lock);
120 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
122 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
124 h1 %= CONNTRACK_LOCKS;
125 h2 %= CONNTRACK_LOCKS;
126 spin_unlock(&nf_conntrack_locks[h1]);
128 spin_unlock(&nf_conntrack_locks[h2]);
131 /* return true if we need to recompute hashes (in case hash table was resized) */
132 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
133 unsigned int h2, unsigned int sequence)
135 h1 %= CONNTRACK_LOCKS;
136 h2 %= CONNTRACK_LOCKS;
138 nf_conntrack_lock(&nf_conntrack_locks[h1]);
140 spin_lock_nested(&nf_conntrack_locks[h2],
141 SINGLE_DEPTH_NESTING);
143 nf_conntrack_lock(&nf_conntrack_locks[h2]);
144 spin_lock_nested(&nf_conntrack_locks[h1],
145 SINGLE_DEPTH_NESTING);
147 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
148 nf_conntrack_double_unlock(h1, h2);
154 static void nf_conntrack_all_lock(void)
158 spin_lock(&nf_conntrack_locks_all_lock);
160 nf_conntrack_locks_all = true;
162 for (i = 0; i < CONNTRACK_LOCKS; i++) {
163 spin_lock(&nf_conntrack_locks[i]);
165 /* This spin_unlock provides the "release" to ensure that
166 * nf_conntrack_locks_all==true is visible to everyone that
167 * acquired spin_lock(&nf_conntrack_locks[]).
169 spin_unlock(&nf_conntrack_locks[i]);
173 static void nf_conntrack_all_unlock(void)
175 /* All prior stores must be complete before we clear
176 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
177 * might observe the false value but not the entire
179 * It pairs with the smp_load_acquire() in nf_conntrack_lock()
181 smp_store_release(&nf_conntrack_locks_all, false);
182 spin_unlock(&nf_conntrack_locks_all_lock);
185 unsigned int nf_conntrack_htable_size __read_mostly;
186 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
188 unsigned int nf_conntrack_max __read_mostly;
189 seqcount_t nf_conntrack_generation __read_mostly;
190 static unsigned int nf_conntrack_hash_rnd __read_mostly;
192 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
193 const struct net *net)
198 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
200 /* The direction must be ignored, so we hash everything up to the
201 * destination ports (which is a multiple of 4) and treat the last
202 * three bytes manually.
204 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
205 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
206 return jhash2((u32 *)tuple, n, seed ^
207 (((__force __u16)tuple->dst.u.all << 16) |
208 tuple->dst.protonum));
211 static u32 scale_hash(u32 hash)
213 return reciprocal_scale(hash, nf_conntrack_htable_size);
216 static u32 __hash_conntrack(const struct net *net,
217 const struct nf_conntrack_tuple *tuple,
220 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
223 static u32 hash_conntrack(const struct net *net,
224 const struct nf_conntrack_tuple *tuple)
226 return scale_hash(hash_conntrack_raw(tuple, net));
230 nf_ct_get_tuple(const struct sk_buff *skb,
232 unsigned int dataoff,
236 struct nf_conntrack_tuple *tuple,
237 const struct nf_conntrack_l3proto *l3proto,
238 const struct nf_conntrack_l4proto *l4proto)
240 memset(tuple, 0, sizeof(*tuple));
242 tuple->src.l3num = l3num;
243 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
246 tuple->dst.protonum = protonum;
247 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
249 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
251 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
253 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
255 struct net *net, struct nf_conntrack_tuple *tuple)
257 const struct nf_conntrack_l3proto *l3proto;
258 const struct nf_conntrack_l4proto *l4proto;
259 unsigned int protoff;
265 l3proto = __nf_ct_l3proto_find(l3num);
266 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
267 if (ret != NF_ACCEPT) {
272 l4proto = __nf_ct_l4proto_find(l3num, protonum);
274 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
280 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
283 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
284 const struct nf_conntrack_tuple *orig,
285 const struct nf_conntrack_l3proto *l3proto,
286 const struct nf_conntrack_l4proto *l4proto)
288 memset(inverse, 0, sizeof(*inverse));
290 inverse->src.l3num = orig->src.l3num;
291 if (l3proto->invert_tuple(inverse, orig) == 0)
294 inverse->dst.dir = !orig->dst.dir;
296 inverse->dst.protonum = orig->dst.protonum;
297 return l4proto->invert_tuple(inverse, orig);
299 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
302 clean_from_lists(struct nf_conn *ct)
304 pr_debug("clean_from_lists(%p)\n", ct);
305 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
306 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
308 /* Destroy all pending expectations */
309 nf_ct_remove_expectations(ct);
312 /* must be called with local_bh_disable */
313 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
315 struct ct_pcpu *pcpu;
317 /* add this conntrack to the (per cpu) dying list */
318 ct->cpu = smp_processor_id();
319 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
321 spin_lock(&pcpu->lock);
322 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
324 spin_unlock(&pcpu->lock);
327 /* must be called with local_bh_disable */
328 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
330 struct ct_pcpu *pcpu;
332 /* add this conntrack to the (per cpu) unconfirmed list */
333 ct->cpu = smp_processor_id();
334 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
336 spin_lock(&pcpu->lock);
337 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
339 spin_unlock(&pcpu->lock);
342 /* must be called with local_bh_disable */
343 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
345 struct ct_pcpu *pcpu;
347 /* We overload first tuple to link into unconfirmed or dying list.*/
348 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
350 spin_lock(&pcpu->lock);
351 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
352 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
353 spin_unlock(&pcpu->lock);
356 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
358 /* Released via destroy_conntrack() */
359 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
360 const struct nf_conntrack_zone *zone,
363 struct nf_conn *tmpl, *p;
365 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
366 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
371 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
373 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
374 tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
377 tmpl = kzalloc(sizeof(*tmpl), flags);
382 tmpl->status = IPS_TEMPLATE;
383 write_pnet(&tmpl->ct_net, net);
384 nf_ct_zone_add(tmpl, zone);
385 atomic_set(&tmpl->ct_general.use, 0);
389 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
391 void nf_ct_tmpl_free(struct nf_conn *tmpl)
393 nf_ct_ext_destroy(tmpl);
394 nf_ct_ext_free(tmpl);
396 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
397 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
401 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
404 destroy_conntrack(struct nf_conntrack *nfct)
406 struct nf_conn *ct = (struct nf_conn *)nfct;
407 const struct nf_conntrack_l4proto *l4proto;
409 pr_debug("destroy_conntrack(%p)\n", ct);
410 WARN_ON(atomic_read(&nfct->use) != 0);
412 if (unlikely(nf_ct_is_template(ct))) {
416 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
417 if (l4proto->destroy)
418 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 const struct nf_conntrack_l4proto *l4proto;
703 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
704 if (l4proto->allow_clash &&
705 ((ct->status & IPS_NAT_DONE_MASK) == 0) &&
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.
764 /* No external references means no one else could have
767 WARN_ON(nf_ct_is_confirmed(ct));
768 pr_debug("Confirming conntrack %p\n", ct);
769 /* We have to check the DYING flag after unlink to prevent
770 * a race against nf_ct_get_next_corpse() possibly called from
771 * user context, else we insert an already 'dead' hash, blocking
772 * further use of that particular connection -JM.
774 nf_ct_del_from_dying_or_unconfirmed_list(ct);
776 if (unlikely(nf_ct_is_dying(ct))) {
777 nf_ct_add_to_dying_list(ct);
781 /* See if there's one in the list already, including reverse:
782 NAT could have grabbed it without realizing, since we're
783 not in the hash. If there is, we lost race. */
784 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
785 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
789 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
790 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
794 /* Timer relative to confirmation time, not original
795 setting time, otherwise we'd get timer wrap in
796 weird delay cases. */
797 ct->timeout += nfct_time_stamp;
798 atomic_inc(&ct->ct_general.use);
799 ct->status |= IPS_CONFIRMED;
801 /* set conntrack timestamp, if enabled. */
802 tstamp = nf_conn_tstamp_find(ct);
804 if (skb->tstamp == 0)
805 __net_timestamp(skb);
807 tstamp->start = ktime_to_ns(skb->tstamp);
809 /* Since the lookup is lockless, hash insertion must be done after
810 * starting the timer and setting the CONFIRMED bit. The RCU barriers
811 * guarantee that no other CPU can find the conntrack before the above
812 * stores are visible.
814 __nf_conntrack_hash_insert(ct, hash, reply_hash);
815 nf_conntrack_double_unlock(hash, reply_hash);
818 help = nfct_help(ct);
819 if (help && help->helper)
820 nf_conntrack_event_cache(IPCT_HELPER, ct);
822 nf_conntrack_event_cache(master_ct(ct) ?
823 IPCT_RELATED : IPCT_NEW, ct);
827 nf_ct_add_to_dying_list(ct);
828 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
830 nf_conntrack_double_unlock(hash, reply_hash);
831 NF_CT_STAT_INC(net, insert_failed);
835 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
837 /* Returns true if a connection correspondings to the tuple (required
840 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
841 const struct nf_conn *ignored_conntrack)
843 struct net *net = nf_ct_net(ignored_conntrack);
844 const struct nf_conntrack_zone *zone;
845 struct nf_conntrack_tuple_hash *h;
846 struct hlist_nulls_head *ct_hash;
847 unsigned int hash, hsize;
848 struct hlist_nulls_node *n;
851 zone = nf_ct_zone(ignored_conntrack);
855 nf_conntrack_get_ht(&ct_hash, &hsize);
856 hash = __hash_conntrack(net, tuple, hsize);
858 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
859 ct = nf_ct_tuplehash_to_ctrack(h);
861 if (ct == ignored_conntrack)
864 if (nf_ct_is_expired(ct)) {
865 nf_ct_gc_expired(ct);
869 if (nf_ct_key_equal(h, tuple, zone, net)) {
870 NF_CT_STAT_INC_ATOMIC(net, found);
876 if (get_nulls_value(n) != hash) {
877 NF_CT_STAT_INC_ATOMIC(net, search_restart);
885 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
887 #define NF_CT_EVICTION_RANGE 8
889 /* There's a small race here where we may free a just-assured
890 connection. Too bad: we're in trouble anyway. */
891 static unsigned int early_drop_list(struct net *net,
892 struct hlist_nulls_head *head)
894 struct nf_conntrack_tuple_hash *h;
895 struct hlist_nulls_node *n;
896 unsigned int drops = 0;
899 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
900 tmp = nf_ct_tuplehash_to_ctrack(h);
902 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status))
905 if (nf_ct_is_expired(tmp)) {
906 nf_ct_gc_expired(tmp);
910 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
911 !net_eq(nf_ct_net(tmp), net) ||
915 if (!atomic_inc_not_zero(&tmp->ct_general.use))
918 /* kill only if still in same netns -- might have moved due to
919 * SLAB_TYPESAFE_BY_RCU rules.
921 * We steal the timer reference. If that fails timer has
922 * already fired or someone else deleted it. Just drop ref
923 * and move to next entry.
925 if (net_eq(nf_ct_net(tmp), net) &&
926 nf_ct_is_confirmed(tmp) &&
927 nf_ct_delete(tmp, 0, 0))
936 static noinline int early_drop(struct net *net, unsigned int _hash)
940 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
941 struct hlist_nulls_head *ct_hash;
942 unsigned int hash, hsize, drops;
945 nf_conntrack_get_ht(&ct_hash, &hsize);
946 hash = reciprocal_scale(_hash++, hsize);
948 drops = early_drop_list(net, &ct_hash[hash]);
952 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
960 static bool gc_worker_skip_ct(const struct nf_conn *ct)
962 return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct);
965 static bool gc_worker_can_early_drop(const struct nf_conn *ct)
967 const struct nf_conntrack_l4proto *l4proto;
969 if (!test_bit(IPS_ASSURED_BIT, &ct->status))
972 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
973 if (l4proto->can_early_drop && l4proto->can_early_drop(ct))
979 #define DAY (86400 * HZ)
981 /* Set an arbitrary timeout large enough not to ever expire, this save
982 * us a check for the IPS_OFFLOAD_BIT from the packet path via
983 * nf_ct_is_expired().
985 static void nf_ct_offload_timeout(struct nf_conn *ct)
987 if (nf_ct_expires(ct) < DAY / 2)
988 ct->timeout = nfct_time_stamp + DAY;
991 static void gc_worker(struct work_struct *work)
993 unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
994 unsigned int i, goal, buckets = 0, expired_count = 0;
995 unsigned int nf_conntrack_max95 = 0;
996 struct conntrack_gc_work *gc_work;
997 unsigned int ratio, scanned = 0;
998 unsigned long next_run;
1000 gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
1002 goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
1003 i = gc_work->last_bucket;
1004 if (gc_work->early_drop)
1005 nf_conntrack_max95 = nf_conntrack_max / 100u * 95u;
1008 struct nf_conntrack_tuple_hash *h;
1009 struct hlist_nulls_head *ct_hash;
1010 struct hlist_nulls_node *n;
1011 unsigned int hashsz;
1012 struct nf_conn *tmp;
1017 nf_conntrack_get_ht(&ct_hash, &hashsz);
1021 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
1024 tmp = nf_ct_tuplehash_to_ctrack(h);
1027 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
1028 nf_ct_offload_timeout(tmp);
1032 if (nf_ct_is_expired(tmp)) {
1033 nf_ct_gc_expired(tmp);
1038 if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp))
1041 net = nf_ct_net(tmp);
1042 if (atomic_read(&net->ct.count) < nf_conntrack_max95)
1045 /* need to take reference to avoid possible races */
1046 if (!atomic_inc_not_zero(&tmp->ct_general.use))
1049 if (gc_worker_skip_ct(tmp)) {
1054 if (gc_worker_can_early_drop(tmp))
1060 /* could check get_nulls_value() here and restart if ct
1061 * was moved to another chain. But given gc is best-effort
1062 * we will just continue with next hash slot.
1066 } while (++buckets < goal);
1068 if (gc_work->exiting)
1072 * Eviction will normally happen from the packet path, and not
1073 * from this gc worker.
1075 * This worker is only here to reap expired entries when system went
1076 * idle after a busy period.
1078 * The heuristics below are supposed to balance conflicting goals:
1080 * 1. Minimize time until we notice a stale entry
1081 * 2. Maximize scan intervals to not waste cycles
1083 * Normally, expire ratio will be close to 0.
1085 * As soon as a sizeable fraction of the entries have expired
1086 * increase scan frequency.
1088 ratio = scanned ? expired_count * 100 / scanned : 0;
1089 if (ratio > GC_EVICT_RATIO) {
1090 gc_work->next_gc_run = min_interval;
1092 unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV;
1094 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0);
1096 gc_work->next_gc_run += min_interval;
1097 if (gc_work->next_gc_run > max)
1098 gc_work->next_gc_run = max;
1101 next_run = gc_work->next_gc_run;
1102 gc_work->last_bucket = i;
1103 gc_work->early_drop = false;
1104 queue_delayed_work(system_power_efficient_wq, &gc_work->dwork, next_run);
1107 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1109 INIT_DEFERRABLE_WORK(&gc_work->dwork, gc_worker);
1110 gc_work->next_gc_run = HZ;
1111 gc_work->exiting = false;
1114 static struct nf_conn *
1115 __nf_conntrack_alloc(struct net *net,
1116 const struct nf_conntrack_zone *zone,
1117 const struct nf_conntrack_tuple *orig,
1118 const struct nf_conntrack_tuple *repl,
1119 gfp_t gfp, u32 hash)
1123 /* We don't want any race condition at early drop stage */
1124 atomic_inc(&net->ct.count);
1126 if (nf_conntrack_max &&
1127 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1128 if (!early_drop(net, hash)) {
1129 if (!conntrack_gc_work.early_drop)
1130 conntrack_gc_work.early_drop = true;
1131 atomic_dec(&net->ct.count);
1132 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1133 return ERR_PTR(-ENOMEM);
1138 * Do not use kmem_cache_zalloc(), as this cache uses
1139 * SLAB_TYPESAFE_BY_RCU.
1141 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1145 spin_lock_init(&ct->lock);
1146 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1147 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1148 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1149 /* save hash for reusing when confirming */
1150 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1152 write_pnet(&ct->ct_net, net);
1153 memset(&ct->__nfct_init_offset[0], 0,
1154 offsetof(struct nf_conn, proto) -
1155 offsetof(struct nf_conn, __nfct_init_offset[0]));
1157 nf_ct_zone_add(ct, zone);
1159 /* Because we use RCU lookups, we set ct_general.use to zero before
1160 * this is inserted in any list.
1162 atomic_set(&ct->ct_general.use, 0);
1165 atomic_dec(&net->ct.count);
1166 return ERR_PTR(-ENOMEM);
1169 struct nf_conn *nf_conntrack_alloc(struct net *net,
1170 const struct nf_conntrack_zone *zone,
1171 const struct nf_conntrack_tuple *orig,
1172 const struct nf_conntrack_tuple *repl,
1175 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1177 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1179 void nf_conntrack_free(struct nf_conn *ct)
1181 struct net *net = nf_ct_net(ct);
1183 /* A freed object has refcnt == 0, that's
1184 * the golden rule for SLAB_TYPESAFE_BY_RCU
1186 WARN_ON(atomic_read(&ct->ct_general.use) != 0);
1188 nf_ct_ext_destroy(ct);
1190 kmem_cache_free(nf_conntrack_cachep, ct);
1191 smp_mb__before_atomic();
1192 atomic_dec(&net->ct.count);
1194 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1197 /* Allocate a new conntrack: we return -ENOMEM if classification
1198 failed due to stress. Otherwise it really is unclassifiable. */
1199 static noinline struct nf_conntrack_tuple_hash *
1200 init_conntrack(struct net *net, struct nf_conn *tmpl,
1201 const struct nf_conntrack_tuple *tuple,
1202 const struct nf_conntrack_l3proto *l3proto,
1203 const struct nf_conntrack_l4proto *l4proto,
1204 struct sk_buff *skb,
1205 unsigned int dataoff, u32 hash)
1208 struct nf_conn_help *help;
1209 struct nf_conntrack_tuple repl_tuple;
1210 struct nf_conntrack_ecache *ecache;
1211 struct nf_conntrack_expect *exp = NULL;
1212 const struct nf_conntrack_zone *zone;
1213 struct nf_conn_timeout *timeout_ext;
1214 struct nf_conntrack_zone tmp;
1215 unsigned int *timeouts;
1217 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1218 pr_debug("Can't invert tuple.\n");
1222 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1223 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1226 return (struct nf_conntrack_tuple_hash *)ct;
1228 if (!nf_ct_add_synproxy(ct, tmpl)) {
1229 nf_conntrack_free(ct);
1230 return ERR_PTR(-ENOMEM);
1233 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1235 timeouts = nf_ct_timeout_data(timeout_ext);
1236 if (unlikely(!timeouts))
1237 timeouts = l4proto->get_timeouts(net);
1239 timeouts = l4proto->get_timeouts(net);
1242 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1243 nf_conntrack_free(ct);
1244 pr_debug("can't track with proto module\n");
1249 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1252 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1253 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1254 nf_ct_labels_ext_add(ct);
1256 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1257 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1258 ecache ? ecache->expmask : 0,
1262 if (net->ct.expect_count) {
1263 spin_lock(&nf_conntrack_expect_lock);
1264 exp = nf_ct_find_expectation(net, zone, tuple);
1266 pr_debug("expectation arrives ct=%p exp=%p\n",
1268 /* Welcome, Mr. Bond. We've been expecting you... */
1269 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1270 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1271 ct->master = exp->master;
1273 help = nf_ct_helper_ext_add(ct, exp->helper,
1276 rcu_assign_pointer(help->helper, exp->helper);
1279 #ifdef CONFIG_NF_CONNTRACK_MARK
1280 ct->mark = exp->master->mark;
1282 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1283 ct->secmark = exp->master->secmark;
1285 NF_CT_STAT_INC(net, expect_new);
1287 spin_unlock(&nf_conntrack_expect_lock);
1290 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1292 /* Now it is inserted into the unconfirmed list, bump refcount */
1293 nf_conntrack_get(&ct->ct_general);
1294 nf_ct_add_to_unconfirmed_list(ct);
1300 exp->expectfn(ct, exp);
1301 nf_ct_expect_put(exp);
1304 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1307 /* On success, returns 0, sets skb->_nfct | ctinfo */
1309 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1310 struct sk_buff *skb,
1311 unsigned int dataoff,
1314 const struct nf_conntrack_l3proto *l3proto,
1315 const struct nf_conntrack_l4proto *l4proto)
1317 const struct nf_conntrack_zone *zone;
1318 struct nf_conntrack_tuple tuple;
1319 struct nf_conntrack_tuple_hash *h;
1320 enum ip_conntrack_info ctinfo;
1321 struct nf_conntrack_zone tmp;
1325 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1326 dataoff, l3num, protonum, net, &tuple, l3proto,
1328 pr_debug("Can't get tuple\n");
1332 /* look for tuple match */
1333 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1334 hash = hash_conntrack_raw(&tuple, net);
1335 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1337 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1338 skb, dataoff, hash);
1344 ct = nf_ct_tuplehash_to_ctrack(h);
1346 /* It exists; we have (non-exclusive) reference. */
1347 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1348 ctinfo = IP_CT_ESTABLISHED_REPLY;
1350 /* Once we've had two way comms, always ESTABLISHED. */
1351 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1352 pr_debug("normal packet for %p\n", ct);
1353 ctinfo = IP_CT_ESTABLISHED;
1354 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1355 pr_debug("related packet for %p\n", ct);
1356 ctinfo = IP_CT_RELATED;
1358 pr_debug("new packet for %p\n", ct);
1362 nf_ct_set(skb, ct, ctinfo);
1367 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1368 struct sk_buff *skb)
1370 const struct nf_conntrack_l3proto *l3proto;
1371 const struct nf_conntrack_l4proto *l4proto;
1372 struct nf_conn *ct, *tmpl;
1373 enum ip_conntrack_info ctinfo;
1374 unsigned int *timeouts;
1375 unsigned int dataoff;
1379 tmpl = nf_ct_get(skb, &ctinfo);
1380 if (tmpl || ctinfo == IP_CT_UNTRACKED) {
1381 /* Previously seen (loopback or untracked)? Ignore. */
1382 if ((tmpl && !nf_ct_is_template(tmpl)) ||
1383 ctinfo == IP_CT_UNTRACKED) {
1384 NF_CT_STAT_INC_ATOMIC(net, ignore);
1390 /* rcu_read_lock()ed by nf_hook_thresh */
1391 l3proto = __nf_ct_l3proto_find(pf);
1392 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1393 &dataoff, &protonum);
1395 pr_debug("not prepared to track yet or error occurred\n");
1396 NF_CT_STAT_INC_ATOMIC(net, error);
1397 NF_CT_STAT_INC_ATOMIC(net, invalid);
1402 l4proto = __nf_ct_l4proto_find(pf, protonum);
1404 /* It may be an special packet, error, unclean...
1405 * inverse of the return code tells to the netfilter
1406 * core what to do with the packet. */
1407 if (l4proto->error != NULL) {
1408 ret = l4proto->error(net, tmpl, skb, dataoff, pf, hooknum);
1410 NF_CT_STAT_INC_ATOMIC(net, error);
1411 NF_CT_STAT_INC_ATOMIC(net, invalid);
1415 /* ICMP[v6] protocol trackers may assign one conntrack. */
1420 ret = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1423 /* Too stressed to deal. */
1424 NF_CT_STAT_INC_ATOMIC(net, drop);
1429 ct = nf_ct_get(skb, &ctinfo);
1431 /* Not valid part of a connection */
1432 NF_CT_STAT_INC_ATOMIC(net, invalid);
1437 /* Decide what timeout policy we want to apply to this flow. */
1438 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1440 ret = l4proto->packet(ct, skb, dataoff, ctinfo, timeouts);
1442 /* Invalid: inverse of the return code tells
1443 * the netfilter core what to do */
1444 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1445 nf_conntrack_put(&ct->ct_general);
1447 NF_CT_STAT_INC_ATOMIC(net, invalid);
1448 if (ret == -NF_DROP)
1449 NF_CT_STAT_INC_ATOMIC(net, drop);
1450 /* Special case: TCP tracker reports an attempt to reopen a
1451 * closed/aborted connection. We have to go back and create a
1454 if (ret == -NF_REPEAT)
1460 if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
1461 !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1462 nf_conntrack_event_cache(IPCT_REPLY, ct);
1469 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1471 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1472 const struct nf_conntrack_tuple *orig)
1477 ret = nf_ct_invert_tuple(inverse, orig,
1478 __nf_ct_l3proto_find(orig->src.l3num),
1479 __nf_ct_l4proto_find(orig->src.l3num,
1480 orig->dst.protonum));
1484 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1486 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1487 implicitly racy: see __nf_conntrack_confirm */
1488 void nf_conntrack_alter_reply(struct nf_conn *ct,
1489 const struct nf_conntrack_tuple *newreply)
1491 struct nf_conn_help *help = nfct_help(ct);
1493 /* Should be unconfirmed, so not in hash table yet */
1494 WARN_ON(nf_ct_is_confirmed(ct));
1496 pr_debug("Altering reply tuple of %p to ", ct);
1497 nf_ct_dump_tuple(newreply);
1499 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1500 if (ct->master || (help && !hlist_empty(&help->expectations)))
1504 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1507 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1509 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1510 void __nf_ct_refresh_acct(struct nf_conn *ct,
1511 enum ip_conntrack_info ctinfo,
1512 const struct sk_buff *skb,
1513 unsigned long extra_jiffies,
1518 /* Only update if this is not a fixed timeout */
1519 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1522 /* If not in hash table, timer will not be active yet */
1523 if (nf_ct_is_confirmed(ct))
1524 extra_jiffies += nfct_time_stamp;
1526 ct->timeout = extra_jiffies;
1529 nf_ct_acct_update(ct, ctinfo, skb->len);
1531 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1533 bool nf_ct_kill_acct(struct nf_conn *ct,
1534 enum ip_conntrack_info ctinfo,
1535 const struct sk_buff *skb)
1537 nf_ct_acct_update(ct, ctinfo, skb->len);
1539 return nf_ct_delete(ct, 0, 0);
1541 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1543 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1545 #include <linux/netfilter/nfnetlink.h>
1546 #include <linux/netfilter/nfnetlink_conntrack.h>
1547 #include <linux/mutex.h>
1549 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1550 * in ip_conntrack_core, since we don't want the protocols to autoload
1551 * or depend on ctnetlink */
1552 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1553 const struct nf_conntrack_tuple *tuple)
1555 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1556 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1557 goto nla_put_failure;
1563 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1565 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1566 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1567 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1569 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1571 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1572 struct nf_conntrack_tuple *t)
1574 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1577 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1578 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1582 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1584 unsigned int nf_ct_port_nlattr_tuple_size(void)
1586 static unsigned int size __read_mostly;
1589 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1593 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1596 /* Used by ipt_REJECT and ip6t_REJECT. */
1597 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1600 enum ip_conntrack_info ctinfo;
1602 /* This ICMP is in reverse direction to the packet which caused it */
1603 ct = nf_ct_get(skb, &ctinfo);
1604 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1605 ctinfo = IP_CT_RELATED_REPLY;
1607 ctinfo = IP_CT_RELATED;
1609 /* Attach to new skbuff, and increment count */
1610 nf_ct_set(nskb, ct, ctinfo);
1611 nf_conntrack_get(skb_nfct(nskb));
1614 /* Bring out ya dead! */
1615 static struct nf_conn *
1616 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
1617 void *data, unsigned int *bucket)
1619 struct nf_conntrack_tuple_hash *h;
1621 struct hlist_nulls_node *n;
1624 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1625 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1627 nf_conntrack_lock(lockp);
1628 if (*bucket < nf_conntrack_htable_size) {
1629 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1630 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1632 ct = nf_ct_tuplehash_to_ctrack(h);
1644 atomic_inc(&ct->ct_general.use);
1650 static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data),
1651 void *data, u32 portid, int report)
1653 unsigned int bucket = 0, sequence;
1659 sequence = read_seqcount_begin(&nf_conntrack_generation);
1661 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
1662 /* Time to push up daises... */
1664 nf_ct_delete(ct, portid, report);
1669 if (!read_seqcount_retry(&nf_conntrack_generation, sequence))
1676 int (*iter)(struct nf_conn *i, void *data);
1681 static int iter_net_only(struct nf_conn *i, void *data)
1683 struct iter_data *d = data;
1685 if (!net_eq(d->net, nf_ct_net(i)))
1688 return d->iter(i, d->data);
1692 __nf_ct_unconfirmed_destroy(struct net *net)
1696 for_each_possible_cpu(cpu) {
1697 struct nf_conntrack_tuple_hash *h;
1698 struct hlist_nulls_node *n;
1699 struct ct_pcpu *pcpu;
1701 pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1703 spin_lock_bh(&pcpu->lock);
1704 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1707 ct = nf_ct_tuplehash_to_ctrack(h);
1709 /* we cannot call iter() on unconfirmed list, the
1710 * owning cpu can reallocate ct->ext at any time.
1712 set_bit(IPS_DYING_BIT, &ct->status);
1714 spin_unlock_bh(&pcpu->lock);
1719 void nf_ct_unconfirmed_destroy(struct net *net)
1723 if (atomic_read(&net->ct.count) > 0) {
1724 __nf_ct_unconfirmed_destroy(net);
1725 nf_queue_nf_hook_drop(net);
1729 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy);
1731 void nf_ct_iterate_cleanup_net(struct net *net,
1732 int (*iter)(struct nf_conn *i, void *data),
1733 void *data, u32 portid, int report)
1739 if (atomic_read(&net->ct.count) == 0)
1746 nf_ct_iterate_cleanup(iter_net_only, &d, portid, report);
1748 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net);
1751 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
1752 * @iter: callback to invoke for each conntrack
1753 * @data: data to pass to @iter
1755 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
1756 * unconfirmed list as dying (so they will not be inserted into
1759 * Can only be called in module exit path.
1762 nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
1766 down_read(&net_rwsem);
1768 if (atomic_read(&net->ct.count) == 0)
1770 __nf_ct_unconfirmed_destroy(net);
1771 nf_queue_nf_hook_drop(net);
1773 up_read(&net_rwsem);
1775 /* Need to wait for netns cleanup worker to finish, if its
1776 * running -- it might have deleted a net namespace from
1777 * the global list, so our __nf_ct_unconfirmed_destroy() might
1778 * not have affected all namespaces.
1782 /* a conntrack could have been unlinked from unconfirmed list
1783 * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
1784 * This makes sure its inserted into conntrack table.
1788 nf_ct_iterate_cleanup(iter, data, 0, 0);
1790 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
1792 static int kill_all(struct nf_conn *i, void *data)
1794 return net_eq(nf_ct_net(i), data);
1797 void nf_ct_free_hashtable(void *hash, unsigned int size)
1799 if (is_vmalloc_addr(hash))
1802 free_pages((unsigned long)hash,
1803 get_order(sizeof(struct hlist_head) * size));
1805 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1807 void nf_conntrack_cleanup_start(void)
1809 conntrack_gc_work.exiting = true;
1810 RCU_INIT_POINTER(ip_ct_attach, NULL);
1813 void nf_conntrack_cleanup_end(void)
1815 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1817 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
1818 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1820 nf_conntrack_proto_fini();
1821 nf_conntrack_seqadj_fini();
1822 nf_conntrack_labels_fini();
1823 nf_conntrack_helper_fini();
1824 nf_conntrack_timeout_fini();
1825 nf_conntrack_ecache_fini();
1826 nf_conntrack_tstamp_fini();
1827 nf_conntrack_acct_fini();
1828 nf_conntrack_expect_fini();
1830 kmem_cache_destroy(nf_conntrack_cachep);
1834 * Mishearing the voices in his head, our hero wonders how he's
1835 * supposed to kill the mall.
1837 void nf_conntrack_cleanup_net(struct net *net)
1841 list_add(&net->exit_list, &single);
1842 nf_conntrack_cleanup_net_list(&single);
1845 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1851 * This makes sure all current packets have passed through
1852 * netfilter framework. Roll on, two-stage module
1858 list_for_each_entry(net, net_exit_list, exit_list) {
1859 nf_ct_iterate_cleanup(kill_all, net, 0, 0);
1860 if (atomic_read(&net->ct.count) != 0)
1865 goto i_see_dead_people;
1868 list_for_each_entry(net, net_exit_list, exit_list) {
1869 nf_conntrack_proto_pernet_fini(net);
1870 nf_conntrack_helper_pernet_fini(net);
1871 nf_conntrack_ecache_pernet_fini(net);
1872 nf_conntrack_tstamp_pernet_fini(net);
1873 nf_conntrack_acct_pernet_fini(net);
1874 nf_conntrack_expect_pernet_fini(net);
1875 free_percpu(net->ct.stat);
1876 free_percpu(net->ct.pcpu_lists);
1880 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1882 struct hlist_nulls_head *hash;
1883 unsigned int nr_slots, i;
1886 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1889 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1890 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1892 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1895 sz = nr_slots * sizeof(struct hlist_nulls_head);
1896 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1902 for (i = 0; i < nr_slots; i++)
1903 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1907 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1909 int nf_conntrack_hash_resize(unsigned int hashsize)
1912 unsigned int old_size;
1913 struct hlist_nulls_head *hash, *old_hash;
1914 struct nf_conntrack_tuple_hash *h;
1920 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1924 old_size = nf_conntrack_htable_size;
1925 if (old_size == hashsize) {
1926 nf_ct_free_hashtable(hash, hashsize);
1931 nf_conntrack_all_lock();
1932 write_seqcount_begin(&nf_conntrack_generation);
1934 /* Lookups in the old hash might happen in parallel, which means we
1935 * might get false negatives during connection lookup. New connections
1936 * created because of a false negative won't make it into the hash
1937 * though since that required taking the locks.
1940 for (i = 0; i < nf_conntrack_htable_size; i++) {
1941 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1942 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1943 struct nf_conntrack_tuple_hash, hnnode);
1944 ct = nf_ct_tuplehash_to_ctrack(h);
1945 hlist_nulls_del_rcu(&h->hnnode);
1946 bucket = __hash_conntrack(nf_ct_net(ct),
1947 &h->tuple, hashsize);
1948 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1951 old_size = nf_conntrack_htable_size;
1952 old_hash = nf_conntrack_hash;
1954 nf_conntrack_hash = hash;
1955 nf_conntrack_htable_size = hashsize;
1957 write_seqcount_end(&nf_conntrack_generation);
1958 nf_conntrack_all_unlock();
1962 nf_ct_free_hashtable(old_hash, old_size);
1966 int nf_conntrack_set_hashsize(const char *val, const struct kernel_param *kp)
1968 unsigned int hashsize;
1971 if (current->nsproxy->net_ns != &init_net)
1974 /* On boot, we can set this without any fancy locking. */
1975 if (!nf_conntrack_htable_size)
1976 return param_set_uint(val, kp);
1978 rc = kstrtouint(val, 0, &hashsize);
1982 return nf_conntrack_hash_resize(hashsize);
1984 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1986 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1987 &nf_conntrack_htable_size, 0600);
1989 static __always_inline unsigned int total_extension_size(void)
1991 /* remember to add new extensions below */
1992 BUILD_BUG_ON(NF_CT_EXT_NUM > 9);
1994 return sizeof(struct nf_ct_ext) +
1995 sizeof(struct nf_conn_help)
1996 #if IS_ENABLED(CONFIG_NF_NAT)
1997 + sizeof(struct nf_conn_nat)
1999 + sizeof(struct nf_conn_seqadj)
2000 + sizeof(struct nf_conn_acct)
2001 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2002 + sizeof(struct nf_conntrack_ecache)
2004 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2005 + sizeof(struct nf_conn_tstamp)
2007 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2008 + sizeof(struct nf_conn_timeout)
2010 #ifdef CONFIG_NF_CONNTRACK_LABELS
2011 + sizeof(struct nf_conn_labels)
2013 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2014 + sizeof(struct nf_conn_synproxy)
2019 int nf_conntrack_init_start(void)
2025 /* struct nf_ct_ext uses u8 to store offsets/size */
2026 BUILD_BUG_ON(total_extension_size() > 255u);
2028 seqcount_init(&nf_conntrack_generation);
2030 for (i = 0; i < CONNTRACK_LOCKS; i++)
2031 spin_lock_init(&nf_conntrack_locks[i]);
2033 if (!nf_conntrack_htable_size) {
2034 /* Idea from tcp.c: use 1/16384 of memory.
2035 * On i386: 32MB machine has 512 buckets.
2036 * >= 1GB machines have 16384 buckets.
2037 * >= 4GB machines have 65536 buckets.
2039 nf_conntrack_htable_size
2040 = (((totalram_pages << PAGE_SHIFT) / 16384)
2041 / sizeof(struct hlist_head));
2042 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
2043 nf_conntrack_htable_size = 65536;
2044 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
2045 nf_conntrack_htable_size = 16384;
2046 if (nf_conntrack_htable_size < 32)
2047 nf_conntrack_htable_size = 32;
2049 /* Use a max. factor of four by default to get the same max as
2050 * with the old struct list_heads. When a table size is given
2051 * we use the old value of 8 to avoid reducing the max.
2056 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
2057 if (!nf_conntrack_hash)
2060 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
2062 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
2063 sizeof(struct nf_conn),
2065 SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
2066 if (!nf_conntrack_cachep)
2069 ret = nf_conntrack_expect_init();
2073 ret = nf_conntrack_acct_init();
2077 ret = nf_conntrack_tstamp_init();
2081 ret = nf_conntrack_ecache_init();
2085 ret = nf_conntrack_timeout_init();
2089 ret = nf_conntrack_helper_init();
2093 ret = nf_conntrack_labels_init();
2097 ret = nf_conntrack_seqadj_init();
2101 ret = nf_conntrack_proto_init();
2105 conntrack_gc_work_init(&conntrack_gc_work);
2106 queue_delayed_work(system_power_efficient_wq, &conntrack_gc_work.dwork, HZ);
2111 nf_conntrack_seqadj_fini();
2113 nf_conntrack_labels_fini();
2115 nf_conntrack_helper_fini();
2117 nf_conntrack_timeout_fini();
2119 nf_conntrack_ecache_fini();
2121 nf_conntrack_tstamp_fini();
2123 nf_conntrack_acct_fini();
2125 nf_conntrack_expect_fini();
2127 kmem_cache_destroy(nf_conntrack_cachep);
2129 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
2133 void nf_conntrack_init_end(void)
2135 /* For use by REJECT target */
2136 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
2137 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
2141 * We need to use special "null" values, not used in hash table
2143 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2144 #define DYING_NULLS_VAL ((1<<30)+1)
2145 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
2147 int nf_conntrack_init_net(struct net *net)
2152 BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
2153 atomic_set(&net->ct.count, 0);
2155 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
2156 if (!net->ct.pcpu_lists)
2159 for_each_possible_cpu(cpu) {
2160 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2162 spin_lock_init(&pcpu->lock);
2163 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
2164 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
2167 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
2169 goto err_pcpu_lists;
2171 ret = nf_conntrack_expect_pernet_init(net);
2174 ret = nf_conntrack_acct_pernet_init(net);
2177 ret = nf_conntrack_tstamp_pernet_init(net);
2180 ret = nf_conntrack_ecache_pernet_init(net);
2183 ret = nf_conntrack_helper_pernet_init(net);
2186 ret = nf_conntrack_proto_pernet_init(net);
2192 nf_conntrack_helper_pernet_fini(net);
2194 nf_conntrack_ecache_pernet_fini(net);
2196 nf_conntrack_tstamp_pernet_fini(net);
2198 nf_conntrack_acct_pernet_fini(net);
2200 nf_conntrack_expect_pernet_fini(net);
2202 free_percpu(net->ct.stat);
2204 free_percpu(net->ct.pcpu_lists);