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 #include <linux/types.h>
16 #include <linux/netfilter.h>
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/skbuff.h>
20 #include <linux/proc_fs.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stddef.h>
23 #include <linux/slab.h>
24 #include <linux/random.h>
25 #include <linux/jhash.h>
26 #include <linux/err.h>
27 #include <linux/percpu.h>
28 #include <linux/moduleparam.h>
29 #include <linux/notifier.h>
30 #include <linux/kernel.h>
31 #include <linux/netdevice.h>
32 #include <linux/socket.h>
34 #include <linux/nsproxy.h>
35 #include <linux/rculist_nulls.h>
37 #include <net/netfilter/nf_conntrack.h>
38 #include <net/netfilter/nf_conntrack_l3proto.h>
39 #include <net/netfilter/nf_conntrack_l4proto.h>
40 #include <net/netfilter/nf_conntrack_expect.h>
41 #include <net/netfilter/nf_conntrack_helper.h>
42 #include <net/netfilter/nf_conntrack_seqadj.h>
43 #include <net/netfilter/nf_conntrack_core.h>
44 #include <net/netfilter/nf_conntrack_extend.h>
45 #include <net/netfilter/nf_conntrack_acct.h>
46 #include <net/netfilter/nf_conntrack_ecache.h>
47 #include <net/netfilter/nf_conntrack_zones.h>
48 #include <net/netfilter/nf_conntrack_timestamp.h>
49 #include <net/netfilter/nf_conntrack_timeout.h>
50 #include <net/netfilter/nf_conntrack_labels.h>
51 #include <net/netfilter/nf_conntrack_synproxy.h>
52 #include <net/netfilter/nf_nat.h>
53 #include <net/netfilter/nf_nat_core.h>
54 #include <net/netfilter/nf_nat_helper.h>
56 #define NF_CONNTRACK_VERSION "0.5.0"
58 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
59 enum nf_nat_manip_type manip,
60 const struct nlattr *attr) __read_mostly;
61 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
63 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
64 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
66 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
67 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
69 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
71 h1 %= CONNTRACK_LOCKS;
72 h2 %= CONNTRACK_LOCKS;
73 spin_unlock(&nf_conntrack_locks[h1]);
75 spin_unlock(&nf_conntrack_locks[h2]);
78 /* return true if we need to recompute hashes (in case hash table was resized) */
79 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
80 unsigned int h2, unsigned int sequence)
82 h1 %= CONNTRACK_LOCKS;
83 h2 %= CONNTRACK_LOCKS;
85 spin_lock(&nf_conntrack_locks[h1]);
87 spin_lock_nested(&nf_conntrack_locks[h2],
88 SINGLE_DEPTH_NESTING);
90 spin_lock(&nf_conntrack_locks[h2]);
91 spin_lock_nested(&nf_conntrack_locks[h1],
92 SINGLE_DEPTH_NESTING);
94 if (read_seqcount_retry(&net->ct.generation, sequence)) {
95 nf_conntrack_double_unlock(h1, h2);
101 static void nf_conntrack_all_lock(void)
105 for (i = 0; i < CONNTRACK_LOCKS; i++)
106 spin_lock_nested(&nf_conntrack_locks[i], i);
109 static void nf_conntrack_all_unlock(void)
113 for (i = 0; i < CONNTRACK_LOCKS; i++)
114 spin_unlock(&nf_conntrack_locks[i]);
117 unsigned int nf_conntrack_htable_size __read_mostly;
118 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
120 unsigned int nf_conntrack_max __read_mostly;
121 EXPORT_SYMBOL_GPL(nf_conntrack_max);
123 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
124 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
126 unsigned int nf_conntrack_hash_rnd __read_mostly;
127 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
129 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
133 /* The direction must be ignored, so we hash everything up to the
134 * destination ports (which is a multiple of 4) and treat the last
135 * three bytes manually.
137 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
138 return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
139 (((__force __u16)tuple->dst.u.all << 16) |
140 tuple->dst.protonum));
143 static u32 __hash_bucket(u32 hash, unsigned int size)
145 return reciprocal_scale(hash, size);
148 static u32 hash_bucket(u32 hash, const struct net *net)
150 return __hash_bucket(hash, net->ct.htable_size);
153 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
154 u16 zone, unsigned int size)
156 return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
159 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
160 const struct nf_conntrack_tuple *tuple)
162 return __hash_conntrack(tuple, zone, net->ct.htable_size);
166 nf_ct_get_tuple(const struct sk_buff *skb,
168 unsigned int dataoff,
171 struct nf_conntrack_tuple *tuple,
172 const struct nf_conntrack_l3proto *l3proto,
173 const struct nf_conntrack_l4proto *l4proto)
175 memset(tuple, 0, sizeof(*tuple));
177 tuple->src.l3num = l3num;
178 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
181 tuple->dst.protonum = protonum;
182 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
184 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
186 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
188 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
189 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
191 struct nf_conntrack_l3proto *l3proto;
192 struct nf_conntrack_l4proto *l4proto;
193 unsigned int protoff;
199 l3proto = __nf_ct_l3proto_find(l3num);
200 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
201 if (ret != NF_ACCEPT) {
206 l4proto = __nf_ct_l4proto_find(l3num, protonum);
208 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
214 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
217 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
218 const struct nf_conntrack_tuple *orig,
219 const struct nf_conntrack_l3proto *l3proto,
220 const struct nf_conntrack_l4proto *l4proto)
222 memset(inverse, 0, sizeof(*inverse));
224 inverse->src.l3num = orig->src.l3num;
225 if (l3proto->invert_tuple(inverse, orig) == 0)
228 inverse->dst.dir = !orig->dst.dir;
230 inverse->dst.protonum = orig->dst.protonum;
231 return l4proto->invert_tuple(inverse, orig);
233 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
236 clean_from_lists(struct nf_conn *ct)
238 pr_debug("clean_from_lists(%p)\n", ct);
239 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
240 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
242 /* Destroy all pending expectations */
243 nf_ct_remove_expectations(ct);
246 /* must be called with local_bh_disable */
247 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
249 struct ct_pcpu *pcpu;
251 /* add this conntrack to the (per cpu) dying list */
252 ct->cpu = smp_processor_id();
253 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
255 spin_lock(&pcpu->lock);
256 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
258 spin_unlock(&pcpu->lock);
261 /* must be called with local_bh_disable */
262 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
264 struct ct_pcpu *pcpu;
266 /* add this conntrack to the (per cpu) unconfirmed list */
267 ct->cpu = smp_processor_id();
268 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
270 spin_lock(&pcpu->lock);
271 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
273 spin_unlock(&pcpu->lock);
276 /* must be called with local_bh_disable */
277 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
279 struct ct_pcpu *pcpu;
281 /* We overload first tuple to link into unconfirmed or dying list.*/
282 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
284 spin_lock(&pcpu->lock);
285 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
286 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
287 spin_unlock(&pcpu->lock);
291 destroy_conntrack(struct nf_conntrack *nfct)
293 struct nf_conn *ct = (struct nf_conn *)nfct;
294 struct net *net = nf_ct_net(ct);
295 struct nf_conntrack_l4proto *l4proto;
297 pr_debug("destroy_conntrack(%p)\n", ct);
298 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
299 NF_CT_ASSERT(!timer_pending(&ct->timeout));
302 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
303 if (l4proto && l4proto->destroy)
304 l4proto->destroy(ct);
309 /* Expectations will have been removed in clean_from_lists,
310 * except TFTP can create an expectation on the first packet,
311 * before connection is in the list, so we need to clean here,
314 nf_ct_remove_expectations(ct);
316 nf_ct_del_from_dying_or_unconfirmed_list(ct);
318 NF_CT_STAT_INC(net, delete);
322 nf_ct_put(ct->master);
324 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
325 nf_conntrack_free(ct);
328 static void nf_ct_delete_from_lists(struct nf_conn *ct)
330 struct net *net = nf_ct_net(ct);
331 unsigned int hash, reply_hash;
332 u16 zone = nf_ct_zone(ct);
333 unsigned int sequence;
335 nf_ct_helper_destroy(ct);
339 sequence = read_seqcount_begin(&net->ct.generation);
340 hash = hash_conntrack(net, zone,
341 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
342 reply_hash = hash_conntrack(net, zone,
343 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
344 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
346 clean_from_lists(ct);
347 nf_conntrack_double_unlock(hash, reply_hash);
349 nf_ct_add_to_dying_list(ct);
351 NF_CT_STAT_INC(net, delete_list);
355 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
357 struct nf_conn_tstamp *tstamp;
359 tstamp = nf_conn_tstamp_find(ct);
360 if (tstamp && tstamp->stop == 0)
361 tstamp->stop = ktime_get_real_ns();
363 if (nf_ct_is_dying(ct))
366 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
367 portid, report) < 0) {
368 /* destroy event was not delivered */
369 nf_ct_delete_from_lists(ct);
370 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
374 nf_conntrack_ecache_work(nf_ct_net(ct));
375 set_bit(IPS_DYING_BIT, &ct->status);
377 nf_ct_delete_from_lists(ct);
381 EXPORT_SYMBOL_GPL(nf_ct_delete);
383 static void death_by_timeout(unsigned long ul_conntrack)
385 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
389 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
390 const struct nf_conntrack_tuple *tuple,
393 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
395 /* A conntrack can be recreated with the equal tuple,
396 * so we need to check that the conntrack is confirmed
398 return nf_ct_tuple_equal(tuple, &h->tuple) &&
399 nf_ct_zone(ct) == zone &&
400 nf_ct_is_confirmed(ct);
405 * - Caller must take a reference on returned object
406 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
408 static struct nf_conntrack_tuple_hash *
409 ____nf_conntrack_find(struct net *net, u16 zone,
410 const struct nf_conntrack_tuple *tuple, u32 hash)
412 struct nf_conntrack_tuple_hash *h;
413 struct hlist_nulls_node *n;
414 unsigned int bucket = hash_bucket(hash, net);
416 /* Disable BHs the entire time since we normally need to disable them
417 * at least once for the stats anyway.
421 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
422 if (nf_ct_key_equal(h, tuple, zone)) {
423 NF_CT_STAT_INC(net, found);
427 NF_CT_STAT_INC(net, searched);
430 * if the nulls value we got at the end of this lookup is
431 * not the expected one, we must restart lookup.
432 * We probably met an item that was moved to another chain.
434 if (get_nulls_value(n) != bucket) {
435 NF_CT_STAT_INC(net, search_restart);
443 /* Find a connection corresponding to a tuple. */
444 static struct nf_conntrack_tuple_hash *
445 __nf_conntrack_find_get(struct net *net, u16 zone,
446 const struct nf_conntrack_tuple *tuple, u32 hash)
448 struct nf_conntrack_tuple_hash *h;
453 h = ____nf_conntrack_find(net, zone, tuple, hash);
455 ct = nf_ct_tuplehash_to_ctrack(h);
456 if (unlikely(nf_ct_is_dying(ct) ||
457 !atomic_inc_not_zero(&ct->ct_general.use)))
460 if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
471 struct nf_conntrack_tuple_hash *
472 nf_conntrack_find_get(struct net *net, u16 zone,
473 const struct nf_conntrack_tuple *tuple)
475 return __nf_conntrack_find_get(net, zone, tuple,
476 hash_conntrack_raw(tuple, zone));
478 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
480 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
482 unsigned int reply_hash)
484 struct net *net = nf_ct_net(ct);
486 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
487 &net->ct.hash[hash]);
488 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
489 &net->ct.hash[reply_hash]);
493 nf_conntrack_hash_check_insert(struct nf_conn *ct)
495 struct net *net = nf_ct_net(ct);
496 unsigned int hash, reply_hash;
497 struct nf_conntrack_tuple_hash *h;
498 struct hlist_nulls_node *n;
500 unsigned int sequence;
502 zone = nf_ct_zone(ct);
506 sequence = read_seqcount_begin(&net->ct.generation);
507 hash = hash_conntrack(net, zone,
508 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
509 reply_hash = hash_conntrack(net, zone,
510 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
511 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
513 /* See if there's one in the list already, including reverse */
514 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
515 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
517 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
519 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
520 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
522 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
525 add_timer(&ct->timeout);
527 /* The caller holds a reference to this object */
528 atomic_set(&ct->ct_general.use, 2);
529 __nf_conntrack_hash_insert(ct, hash, reply_hash);
530 nf_conntrack_double_unlock(hash, reply_hash);
531 NF_CT_STAT_INC(net, insert);
536 nf_conntrack_double_unlock(hash, reply_hash);
537 NF_CT_STAT_INC(net, insert_failed);
541 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
543 /* deletion from this larval template list happens via nf_ct_put() */
544 void nf_conntrack_tmpl_insert(struct net *net, struct nf_conn *tmpl)
546 struct ct_pcpu *pcpu;
548 __set_bit(IPS_TEMPLATE_BIT, &tmpl->status);
549 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
550 nf_conntrack_get(&tmpl->ct_general);
552 /* add this conntrack to the (per cpu) tmpl list */
554 tmpl->cpu = smp_processor_id();
555 pcpu = per_cpu_ptr(nf_ct_net(tmpl)->ct.pcpu_lists, tmpl->cpu);
557 spin_lock(&pcpu->lock);
558 /* Overload tuple linked list to put us in template list. */
559 hlist_nulls_add_head_rcu(&tmpl->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
561 spin_unlock_bh(&pcpu->lock);
563 EXPORT_SYMBOL_GPL(nf_conntrack_tmpl_insert);
565 /* Confirm a connection given skb; places it in hash table */
567 __nf_conntrack_confirm(struct sk_buff *skb)
569 unsigned int hash, reply_hash;
570 struct nf_conntrack_tuple_hash *h;
572 struct nf_conn_help *help;
573 struct nf_conn_tstamp *tstamp;
574 struct hlist_nulls_node *n;
575 enum ip_conntrack_info ctinfo;
578 unsigned int sequence;
580 ct = nf_ct_get(skb, &ctinfo);
583 /* ipt_REJECT uses nf_conntrack_attach to attach related
584 ICMP/TCP RST packets in other direction. Actual packet
585 which created connection will be IP_CT_NEW or for an
586 expected connection, IP_CT_RELATED. */
587 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
590 zone = nf_ct_zone(ct);
594 sequence = read_seqcount_begin(&net->ct.generation);
595 /* reuse the hash saved before */
596 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
597 hash = hash_bucket(hash, net);
598 reply_hash = hash_conntrack(net, zone,
599 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
601 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
603 /* We're not in hash table, and we refuse to set up related
604 * connections for unconfirmed conns. But packet copies and
605 * REJECT will give spurious warnings here.
607 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
609 /* No external references means no one else could have
612 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
613 pr_debug("Confirming conntrack %p\n", ct);
615 /* We have to check the DYING flag after unlink to prevent
616 * a race against nf_ct_get_next_corpse() possibly called from
617 * user context, else we insert an already 'dead' hash, blocking
618 * further use of that particular connection -JM.
620 nf_ct_del_from_dying_or_unconfirmed_list(ct);
622 if (unlikely(nf_ct_is_dying(ct))) {
623 nf_ct_add_to_dying_list(ct);
624 nf_conntrack_double_unlock(hash, reply_hash);
629 /* See if there's one in the list already, including reverse:
630 NAT could have grabbed it without realizing, since we're
631 not in the hash. If there is, we lost race. */
632 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
633 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
635 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
637 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
638 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
640 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
643 /* Timer relative to confirmation time, not original
644 setting time, otherwise we'd get timer wrap in
645 weird delay cases. */
646 ct->timeout.expires += jiffies;
647 add_timer(&ct->timeout);
648 atomic_inc(&ct->ct_general.use);
649 ct->status |= IPS_CONFIRMED;
651 /* set conntrack timestamp, if enabled. */
652 tstamp = nf_conn_tstamp_find(ct);
654 if (skb->tstamp.tv64 == 0)
655 __net_timestamp(skb);
657 tstamp->start = ktime_to_ns(skb->tstamp);
659 /* Since the lookup is lockless, hash insertion must be done after
660 * starting the timer and setting the CONFIRMED bit. The RCU barriers
661 * guarantee that no other CPU can find the conntrack before the above
662 * stores are visible.
664 __nf_conntrack_hash_insert(ct, hash, reply_hash);
665 nf_conntrack_double_unlock(hash, reply_hash);
666 NF_CT_STAT_INC(net, insert);
669 help = nfct_help(ct);
670 if (help && help->helper)
671 nf_conntrack_event_cache(IPCT_HELPER, ct);
673 nf_conntrack_event_cache(master_ct(ct) ?
674 IPCT_RELATED : IPCT_NEW, ct);
678 nf_conntrack_double_unlock(hash, reply_hash);
679 NF_CT_STAT_INC(net, insert_failed);
683 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
685 /* Returns true if a connection correspondings to the tuple (required
688 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
689 const struct nf_conn *ignored_conntrack)
691 struct net *net = nf_ct_net(ignored_conntrack);
692 struct nf_conntrack_tuple_hash *h;
693 struct hlist_nulls_node *n;
695 u16 zone = nf_ct_zone(ignored_conntrack);
696 unsigned int hash = hash_conntrack(net, zone, tuple);
698 /* Disable BHs the entire time since we need to disable them at
699 * least once for the stats anyway.
702 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
703 ct = nf_ct_tuplehash_to_ctrack(h);
704 if (ct != ignored_conntrack &&
705 nf_ct_tuple_equal(tuple, &h->tuple) &&
706 nf_ct_zone(ct) == zone) {
707 NF_CT_STAT_INC(net, found);
708 rcu_read_unlock_bh();
711 NF_CT_STAT_INC(net, searched);
713 rcu_read_unlock_bh();
717 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
719 #define NF_CT_EVICTION_RANGE 8
721 /* There's a small race here where we may free a just-assured
722 connection. Too bad: we're in trouble anyway. */
723 static noinline int early_drop(struct net *net, unsigned int _hash)
725 /* Use oldest entry, which is roughly LRU */
726 struct nf_conntrack_tuple_hash *h;
727 struct nf_conn *ct = NULL, *tmp;
728 struct hlist_nulls_node *n;
729 unsigned int i = 0, cnt = 0;
731 unsigned int hash, sequence;
736 sequence = read_seqcount_begin(&net->ct.generation);
737 hash = hash_bucket(_hash, net);
738 for (; i < net->ct.htable_size; i++) {
739 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
741 if (read_seqcount_retry(&net->ct.generation, sequence)) {
745 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
747 tmp = nf_ct_tuplehash_to_ctrack(h);
748 if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
749 !nf_ct_is_dying(tmp) &&
750 atomic_inc_not_zero(&tmp->ct_general.use)) {
757 hash = (hash + 1) % net->ct.htable_size;
760 if (ct || cnt >= NF_CT_EVICTION_RANGE)
769 if (del_timer(&ct->timeout)) {
770 if (nf_ct_delete(ct, 0, 0)) {
772 NF_CT_STAT_INC_ATOMIC(net, early_drop);
779 void init_nf_conntrack_hash_rnd(void)
784 * Why not initialize nf_conntrack_rnd in a "init()" function ?
785 * Because there isn't enough entropy when system initializing,
786 * and we initialize it as late as possible.
789 get_random_bytes(&rand, sizeof(rand));
791 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
794 static struct nf_conn *
795 __nf_conntrack_alloc(struct net *net, u16 zone,
796 const struct nf_conntrack_tuple *orig,
797 const struct nf_conntrack_tuple *repl,
802 if (unlikely(!nf_conntrack_hash_rnd)) {
803 init_nf_conntrack_hash_rnd();
804 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
805 hash = hash_conntrack_raw(orig, zone);
808 /* We don't want any race condition at early drop stage */
809 atomic_inc(&net->ct.count);
811 if (nf_conntrack_max &&
812 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
813 if (!early_drop(net, hash)) {
814 atomic_dec(&net->ct.count);
815 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
816 return ERR_PTR(-ENOMEM);
821 * Do not use kmem_cache_zalloc(), as this cache uses
822 * SLAB_DESTROY_BY_RCU.
824 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
826 atomic_dec(&net->ct.count);
827 return ERR_PTR(-ENOMEM);
830 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
831 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
833 memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
834 offsetof(struct nf_conn, proto) -
835 offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
836 spin_lock_init(&ct->lock);
837 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
838 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
839 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
840 /* save hash for reusing when confirming */
841 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
842 /* Don't set timer yet: wait for confirmation */
843 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
844 write_pnet(&ct->ct_net, net);
845 #ifdef CONFIG_NF_CONNTRACK_ZONES
847 struct nf_conntrack_zone *nf_ct_zone;
849 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
852 nf_ct_zone->id = zone;
855 /* Because we use RCU lookups, we set ct_general.use to zero before
856 * this is inserted in any list.
858 atomic_set(&ct->ct_general.use, 0);
861 #ifdef CONFIG_NF_CONNTRACK_ZONES
863 atomic_dec(&net->ct.count);
864 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
865 return ERR_PTR(-ENOMEM);
869 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
870 const struct nf_conntrack_tuple *orig,
871 const struct nf_conntrack_tuple *repl,
874 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
876 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
878 void nf_conntrack_free(struct nf_conn *ct)
880 struct net *net = nf_ct_net(ct);
882 /* A freed object has refcnt == 0, that's
883 * the golden rule for SLAB_DESTROY_BY_RCU
885 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
887 nf_ct_ext_destroy(ct);
889 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
890 smp_mb__before_atomic();
891 atomic_dec(&net->ct.count);
893 EXPORT_SYMBOL_GPL(nf_conntrack_free);
896 /* Allocate a new conntrack: we return -ENOMEM if classification
897 failed due to stress. Otherwise it really is unclassifiable. */
898 static struct nf_conntrack_tuple_hash *
899 init_conntrack(struct net *net, struct nf_conn *tmpl,
900 const struct nf_conntrack_tuple *tuple,
901 struct nf_conntrack_l3proto *l3proto,
902 struct nf_conntrack_l4proto *l4proto,
904 unsigned int dataoff, u32 hash)
907 struct nf_conn_help *help;
908 struct nf_conntrack_tuple repl_tuple;
909 struct nf_conntrack_ecache *ecache;
910 struct nf_conntrack_expect *exp = NULL;
911 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
912 struct nf_conn_timeout *timeout_ext;
913 unsigned int *timeouts;
915 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
916 pr_debug("Can't invert tuple.\n");
920 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
923 return (struct nf_conntrack_tuple_hash *)ct;
925 if (tmpl && nfct_synproxy(tmpl)) {
926 nfct_seqadj_ext_add(ct);
927 nfct_synproxy_ext_add(ct);
930 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
932 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
934 timeouts = l4proto->get_timeouts(net);
936 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
937 nf_conntrack_free(ct);
938 pr_debug("init conntrack: can't track with proto module\n");
943 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
945 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
946 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
947 nf_ct_labels_ext_add(ct);
949 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
950 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
951 ecache ? ecache->expmask : 0,
955 if (net->ct.expect_count) {
956 spin_lock(&nf_conntrack_expect_lock);
957 exp = nf_ct_find_expectation(net, zone, tuple);
959 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
961 /* Welcome, Mr. Bond. We've been expecting you... */
962 __set_bit(IPS_EXPECTED_BIT, &ct->status);
963 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
964 ct->master = exp->master;
966 help = nf_ct_helper_ext_add(ct, exp->helper,
969 rcu_assign_pointer(help->helper, exp->helper);
972 #ifdef CONFIG_NF_CONNTRACK_MARK
973 ct->mark = exp->master->mark;
975 #ifdef CONFIG_NF_CONNTRACK_SECMARK
976 ct->secmark = exp->master->secmark;
978 NF_CT_STAT_INC(net, expect_new);
980 spin_unlock(&nf_conntrack_expect_lock);
983 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
984 NF_CT_STAT_INC(net, new);
987 /* Now it is inserted into the unconfirmed list, bump refcount */
988 nf_conntrack_get(&ct->ct_general);
989 nf_ct_add_to_unconfirmed_list(ct);
995 exp->expectfn(ct, exp);
996 nf_ct_expect_put(exp);
999 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1002 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1003 static inline struct nf_conn *
1004 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1005 struct sk_buff *skb,
1006 unsigned int dataoff,
1009 struct nf_conntrack_l3proto *l3proto,
1010 struct nf_conntrack_l4proto *l4proto,
1012 enum ip_conntrack_info *ctinfo)
1014 struct nf_conntrack_tuple tuple;
1015 struct nf_conntrack_tuple_hash *h;
1017 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
1020 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1021 dataoff, l3num, protonum, &tuple, l3proto,
1023 pr_debug("resolve_normal_ct: Can't get tuple\n");
1027 /* look for tuple match */
1028 hash = hash_conntrack_raw(&tuple, zone);
1029 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1031 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1032 skb, dataoff, hash);
1038 ct = nf_ct_tuplehash_to_ctrack(h);
1040 /* It exists; we have (non-exclusive) reference. */
1041 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1042 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1043 /* Please set reply bit if this packet OK */
1046 /* Once we've had two way comms, always ESTABLISHED. */
1047 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1048 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
1049 *ctinfo = IP_CT_ESTABLISHED;
1050 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1051 pr_debug("nf_conntrack_in: related packet for %p\n",
1053 *ctinfo = IP_CT_RELATED;
1055 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
1056 *ctinfo = IP_CT_NEW;
1060 skb->nfct = &ct->ct_general;
1061 skb->nfctinfo = *ctinfo;
1066 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1067 struct sk_buff *skb)
1069 struct nf_conn *ct, *tmpl = NULL;
1070 enum ip_conntrack_info ctinfo;
1071 struct nf_conntrack_l3proto *l3proto;
1072 struct nf_conntrack_l4proto *l4proto;
1073 unsigned int *timeouts;
1074 unsigned int dataoff;
1080 /* Previously seen (loopback or untracked)? Ignore. */
1081 tmpl = (struct nf_conn *)skb->nfct;
1082 if (!nf_ct_is_template(tmpl)) {
1083 NF_CT_STAT_INC_ATOMIC(net, ignore);
1089 /* rcu_read_lock()ed by nf_hook_slow */
1090 l3proto = __nf_ct_l3proto_find(pf);
1091 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1092 &dataoff, &protonum);
1094 pr_debug("not prepared to track yet or error occurred\n");
1095 NF_CT_STAT_INC_ATOMIC(net, error);
1096 NF_CT_STAT_INC_ATOMIC(net, invalid);
1101 l4proto = __nf_ct_l4proto_find(pf, protonum);
1103 /* It may be an special packet, error, unclean...
1104 * inverse of the return code tells to the netfilter
1105 * core what to do with the packet. */
1106 if (l4proto->error != NULL) {
1107 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1110 NF_CT_STAT_INC_ATOMIC(net, error);
1111 NF_CT_STAT_INC_ATOMIC(net, invalid);
1115 /* ICMP[v6] protocol trackers may assign one conntrack. */
1120 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1121 l3proto, l4proto, &set_reply, &ctinfo);
1123 /* Not valid part of a connection */
1124 NF_CT_STAT_INC_ATOMIC(net, invalid);
1130 /* Too stressed to deal. */
1131 NF_CT_STAT_INC_ATOMIC(net, drop);
1136 NF_CT_ASSERT(skb->nfct);
1138 /* Decide what timeout policy we want to apply to this flow. */
1139 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1141 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1143 /* Invalid: inverse of the return code tells
1144 * the netfilter core what to do */
1145 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1146 nf_conntrack_put(skb->nfct);
1148 NF_CT_STAT_INC_ATOMIC(net, invalid);
1149 if (ret == -NF_DROP)
1150 NF_CT_STAT_INC_ATOMIC(net, drop);
1155 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1156 nf_conntrack_event_cache(IPCT_REPLY, ct);
1159 /* Special case: we have to repeat this hook, assign the
1160 * template again to this packet. We assume that this packet
1161 * has no conntrack assigned. This is used by nf_ct_tcp. */
1162 if (ret == NF_REPEAT)
1163 skb->nfct = (struct nf_conntrack *)tmpl;
1170 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1172 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1173 const struct nf_conntrack_tuple *orig)
1178 ret = nf_ct_invert_tuple(inverse, orig,
1179 __nf_ct_l3proto_find(orig->src.l3num),
1180 __nf_ct_l4proto_find(orig->src.l3num,
1181 orig->dst.protonum));
1185 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1187 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1188 implicitly racy: see __nf_conntrack_confirm */
1189 void nf_conntrack_alter_reply(struct nf_conn *ct,
1190 const struct nf_conntrack_tuple *newreply)
1192 struct nf_conn_help *help = nfct_help(ct);
1194 /* Should be unconfirmed, so not in hash table yet */
1195 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1197 pr_debug("Altering reply tuple of %p to ", ct);
1198 nf_ct_dump_tuple(newreply);
1200 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1201 if (ct->master || (help && !hlist_empty(&help->expectations)))
1205 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1208 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1210 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1211 void __nf_ct_refresh_acct(struct nf_conn *ct,
1212 enum ip_conntrack_info ctinfo,
1213 const struct sk_buff *skb,
1214 unsigned long extra_jiffies,
1217 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1220 /* Only update if this is not a fixed timeout */
1221 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1224 /* If not in hash table, timer will not be active yet */
1225 if (!nf_ct_is_confirmed(ct)) {
1226 ct->timeout.expires = extra_jiffies;
1228 unsigned long newtime = jiffies + extra_jiffies;
1230 /* Only update the timeout if the new timeout is at least
1231 HZ jiffies from the old timeout. Need del_timer for race
1232 avoidance (may already be dying). */
1233 if (newtime - ct->timeout.expires >= HZ)
1234 mod_timer_pending(&ct->timeout, newtime);
1239 struct nf_conn_acct *acct;
1241 acct = nf_conn_acct_find(ct);
1243 struct nf_conn_counter *counter = acct->counter;
1245 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1246 atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1250 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1252 bool __nf_ct_kill_acct(struct nf_conn *ct,
1253 enum ip_conntrack_info ctinfo,
1254 const struct sk_buff *skb,
1258 struct nf_conn_acct *acct;
1260 acct = nf_conn_acct_find(ct);
1262 struct nf_conn_counter *counter = acct->counter;
1264 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1265 atomic64_add(skb->len - skb_network_offset(skb),
1266 &counter[CTINFO2DIR(ctinfo)].bytes);
1270 if (del_timer(&ct->timeout)) {
1271 ct->timeout.function((unsigned long)ct);
1276 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1278 #ifdef CONFIG_NF_CONNTRACK_ZONES
1279 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1280 .len = sizeof(struct nf_conntrack_zone),
1281 .align = __alignof__(struct nf_conntrack_zone),
1282 .id = NF_CT_EXT_ZONE,
1286 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1288 #include <linux/netfilter/nfnetlink.h>
1289 #include <linux/netfilter/nfnetlink_conntrack.h>
1290 #include <linux/mutex.h>
1292 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1293 * in ip_conntrack_core, since we don't want the protocols to autoload
1294 * or depend on ctnetlink */
1295 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1296 const struct nf_conntrack_tuple *tuple)
1298 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1299 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1300 goto nla_put_failure;
1306 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1308 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1309 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1310 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1312 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1314 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1315 struct nf_conntrack_tuple *t)
1317 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1320 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1321 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1325 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1327 int nf_ct_port_nlattr_tuple_size(void)
1329 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1331 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1334 /* Used by ipt_REJECT and ip6t_REJECT. */
1335 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1338 enum ip_conntrack_info ctinfo;
1340 /* This ICMP is in reverse direction to the packet which caused it */
1341 ct = nf_ct_get(skb, &ctinfo);
1342 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1343 ctinfo = IP_CT_RELATED_REPLY;
1345 ctinfo = IP_CT_RELATED;
1347 /* Attach to new skbuff, and increment count */
1348 nskb->nfct = &ct->ct_general;
1349 nskb->nfctinfo = ctinfo;
1350 nf_conntrack_get(nskb->nfct);
1353 /* Bring out ya dead! */
1354 static struct nf_conn *
1355 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1356 void *data, unsigned int *bucket)
1358 struct nf_conntrack_tuple_hash *h;
1360 struct hlist_nulls_node *n;
1364 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1365 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1368 if (*bucket < net->ct.htable_size) {
1369 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1370 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1372 ct = nf_ct_tuplehash_to_ctrack(h);
1381 for_each_possible_cpu(cpu) {
1382 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1384 spin_lock_bh(&pcpu->lock);
1385 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1386 ct = nf_ct_tuplehash_to_ctrack(h);
1388 set_bit(IPS_DYING_BIT, &ct->status);
1390 spin_unlock_bh(&pcpu->lock);
1394 atomic_inc(&ct->ct_general.use);
1400 void nf_ct_iterate_cleanup(struct net *net,
1401 int (*iter)(struct nf_conn *i, void *data),
1402 void *data, u32 portid, int report)
1405 unsigned int bucket = 0;
1407 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1408 /* Time to push up daises... */
1409 if (del_timer(&ct->timeout))
1410 nf_ct_delete(ct, portid, report);
1412 /* ... else the timer will get him soon. */
1417 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1419 static int kill_all(struct nf_conn *i, void *data)
1424 void nf_ct_free_hashtable(void *hash, unsigned int size)
1426 if (is_vmalloc_addr(hash))
1429 free_pages((unsigned long)hash,
1430 get_order(sizeof(struct hlist_head) * size));
1432 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1434 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1436 nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1438 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1440 static int untrack_refs(void)
1444 for_each_possible_cpu(cpu) {
1445 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1447 cnt += atomic_read(&ct->ct_general.use) - 1;
1452 void nf_conntrack_cleanup_start(void)
1454 RCU_INIT_POINTER(ip_ct_attach, NULL);
1457 void nf_conntrack_cleanup_end(void)
1459 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1460 while (untrack_refs() > 0)
1463 #ifdef CONFIG_NF_CONNTRACK_ZONES
1464 nf_ct_extend_unregister(&nf_ct_zone_extend);
1466 nf_conntrack_proto_fini();
1467 nf_conntrack_seqadj_fini();
1468 nf_conntrack_labels_fini();
1469 nf_conntrack_helper_fini();
1470 nf_conntrack_timeout_fini();
1471 nf_conntrack_ecache_fini();
1472 nf_conntrack_tstamp_fini();
1473 nf_conntrack_acct_fini();
1474 nf_conntrack_expect_fini();
1478 * Mishearing the voices in his head, our hero wonders how he's
1479 * supposed to kill the mall.
1481 void nf_conntrack_cleanup_net(struct net *net)
1485 list_add(&net->exit_list, &single);
1486 nf_conntrack_cleanup_net_list(&single);
1489 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1495 * This makes sure all current packets have passed through
1496 * netfilter framework. Roll on, two-stage module
1502 list_for_each_entry(net, net_exit_list, exit_list) {
1503 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1504 if (atomic_read(&net->ct.count) != 0)
1509 goto i_see_dead_people;
1512 list_for_each_entry(net, net_exit_list, exit_list) {
1513 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1514 nf_conntrack_proto_pernet_fini(net);
1515 nf_conntrack_helper_pernet_fini(net);
1516 nf_conntrack_ecache_pernet_fini(net);
1517 nf_conntrack_tstamp_pernet_fini(net);
1518 nf_conntrack_acct_pernet_fini(net);
1519 nf_conntrack_expect_pernet_fini(net);
1520 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1521 kfree(net->ct.slabname);
1522 free_percpu(net->ct.stat);
1523 free_percpu(net->ct.pcpu_lists);
1527 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1529 struct hlist_nulls_head *hash;
1530 unsigned int nr_slots, i;
1533 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1534 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1535 sz = nr_slots * sizeof(struct hlist_nulls_head);
1536 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1539 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1544 for (i = 0; i < nr_slots; i++)
1545 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1549 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1551 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1554 unsigned int hashsize, old_size;
1555 struct hlist_nulls_head *hash, *old_hash;
1556 struct nf_conntrack_tuple_hash *h;
1559 if (current->nsproxy->net_ns != &init_net)
1562 /* On boot, we can set this without any fancy locking. */
1563 if (!nf_conntrack_htable_size)
1564 return param_set_uint(val, kp);
1566 rc = kstrtouint(val, 0, &hashsize);
1572 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1577 nf_conntrack_all_lock();
1578 write_seqcount_begin(&init_net.ct.generation);
1580 /* Lookups in the old hash might happen in parallel, which means we
1581 * might get false negatives during connection lookup. New connections
1582 * created because of a false negative won't make it into the hash
1583 * though since that required taking the locks.
1586 for (i = 0; i < init_net.ct.htable_size; i++) {
1587 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1588 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1589 struct nf_conntrack_tuple_hash, hnnode);
1590 ct = nf_ct_tuplehash_to_ctrack(h);
1591 hlist_nulls_del_rcu(&h->hnnode);
1592 bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1594 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1597 old_size = init_net.ct.htable_size;
1598 old_hash = init_net.ct.hash;
1600 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1601 init_net.ct.hash = hash;
1603 write_seqcount_end(&init_net.ct.generation);
1604 nf_conntrack_all_unlock();
1607 nf_ct_free_hashtable(old_hash, old_size);
1610 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1612 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1613 &nf_conntrack_htable_size, 0600);
1615 void nf_ct_untracked_status_or(unsigned long bits)
1619 for_each_possible_cpu(cpu)
1620 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1622 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1624 int nf_conntrack_init_start(void)
1629 for (i = 0; i < CONNTRACK_LOCKS; i++)
1630 spin_lock_init(&nf_conntrack_locks[i]);
1632 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1633 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1634 if (!nf_conntrack_htable_size) {
1635 nf_conntrack_htable_size
1636 = (((totalram_pages << PAGE_SHIFT) / 16384)
1637 / sizeof(struct hlist_head));
1638 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1639 nf_conntrack_htable_size = 16384;
1640 if (nf_conntrack_htable_size < 32)
1641 nf_conntrack_htable_size = 32;
1643 /* Use a max. factor of four by default to get the same max as
1644 * with the old struct list_heads. When a table size is given
1645 * we use the old value of 8 to avoid reducing the max.
1649 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1651 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1652 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1655 ret = nf_conntrack_expect_init();
1659 ret = nf_conntrack_acct_init();
1663 ret = nf_conntrack_tstamp_init();
1667 ret = nf_conntrack_ecache_init();
1671 ret = nf_conntrack_timeout_init();
1675 ret = nf_conntrack_helper_init();
1679 ret = nf_conntrack_labels_init();
1683 ret = nf_conntrack_seqadj_init();
1687 #ifdef CONFIG_NF_CONNTRACK_ZONES
1688 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1692 ret = nf_conntrack_proto_init();
1696 /* Set up fake conntrack: to never be deleted, not in any hashes */
1697 for_each_possible_cpu(cpu) {
1698 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1699 write_pnet(&ct->ct_net, &init_net);
1700 atomic_set(&ct->ct_general.use, 1);
1702 /* - and look it like as a confirmed connection */
1703 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1707 #ifdef CONFIG_NF_CONNTRACK_ZONES
1708 nf_ct_extend_unregister(&nf_ct_zone_extend);
1711 nf_conntrack_seqadj_fini();
1713 nf_conntrack_labels_fini();
1715 nf_conntrack_helper_fini();
1717 nf_conntrack_timeout_fini();
1719 nf_conntrack_ecache_fini();
1721 nf_conntrack_tstamp_fini();
1723 nf_conntrack_acct_fini();
1725 nf_conntrack_expect_fini();
1730 void nf_conntrack_init_end(void)
1732 /* For use by REJECT target */
1733 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1734 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1738 * We need to use special "null" values, not used in hash table
1740 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1741 #define DYING_NULLS_VAL ((1<<30)+1)
1742 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1744 int nf_conntrack_init_net(struct net *net)
1749 atomic_set(&net->ct.count, 0);
1750 seqcount_init(&net->ct.generation);
1752 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1753 if (!net->ct.pcpu_lists)
1756 for_each_possible_cpu(cpu) {
1757 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1759 spin_lock_init(&pcpu->lock);
1760 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1761 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1762 INIT_HLIST_NULLS_HEAD(&pcpu->tmpl, TEMPLATE_NULLS_VAL);
1765 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1767 goto err_pcpu_lists;
1769 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1770 if (!net->ct.slabname)
1773 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1774 sizeof(struct nf_conn), 0,
1775 SLAB_DESTROY_BY_RCU, NULL);
1776 if (!net->ct.nf_conntrack_cachep) {
1777 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1781 net->ct.htable_size = nf_conntrack_htable_size;
1782 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1783 if (!net->ct.hash) {
1784 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1787 ret = nf_conntrack_expect_pernet_init(net);
1790 ret = nf_conntrack_acct_pernet_init(net);
1793 ret = nf_conntrack_tstamp_pernet_init(net);
1796 ret = nf_conntrack_ecache_pernet_init(net);
1799 ret = nf_conntrack_helper_pernet_init(net);
1802 ret = nf_conntrack_proto_pernet_init(net);
1808 nf_conntrack_helper_pernet_fini(net);
1810 nf_conntrack_ecache_pernet_fini(net);
1812 nf_conntrack_tstamp_pernet_fini(net);
1814 nf_conntrack_acct_pernet_fini(net);
1816 nf_conntrack_expect_pernet_fini(net);
1818 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1820 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1822 kfree(net->ct.slabname);
1824 free_percpu(net->ct.stat);
1826 free_percpu(net->ct.pcpu_lists);