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)
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, 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,
156 return __hash_bucket(hash_conntrack_raw(tuple), size);
159 static inline u_int32_t hash_conntrack(const struct net *net,
160 const struct nf_conntrack_tuple *tuple)
162 return __hash_conntrack(tuple, 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);
290 /* Released via destroy_conntrack() */
291 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
292 const struct nf_conntrack_zone *zone,
295 struct nf_conn *tmpl;
297 tmpl = kzalloc(sizeof(*tmpl), flags);
301 tmpl->status = IPS_TEMPLATE;
302 write_pnet(&tmpl->ct_net, net);
304 if (nf_ct_zone_add(tmpl, flags, zone) < 0)
307 atomic_set(&tmpl->ct_general.use, 0);
314 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
316 void nf_ct_tmpl_free(struct nf_conn *tmpl)
318 nf_ct_ext_destroy(tmpl);
319 nf_ct_ext_free(tmpl);
322 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
325 destroy_conntrack(struct nf_conntrack *nfct)
327 struct nf_conn *ct = (struct nf_conn *)nfct;
328 struct net *net = nf_ct_net(ct);
329 struct nf_conntrack_l4proto *l4proto;
331 pr_debug("destroy_conntrack(%p)\n", ct);
332 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
333 NF_CT_ASSERT(!timer_pending(&ct->timeout));
335 if (unlikely(nf_ct_is_template(ct))) {
340 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
341 if (l4proto && l4proto->destroy)
342 l4proto->destroy(ct);
347 /* Expectations will have been removed in clean_from_lists,
348 * except TFTP can create an expectation on the first packet,
349 * before connection is in the list, so we need to clean here,
352 nf_ct_remove_expectations(ct);
354 nf_ct_del_from_dying_or_unconfirmed_list(ct);
356 NF_CT_STAT_INC(net, delete);
360 nf_ct_put(ct->master);
362 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
363 nf_conntrack_free(ct);
366 static void nf_ct_delete_from_lists(struct nf_conn *ct)
368 struct net *net = nf_ct_net(ct);
369 unsigned int hash, reply_hash;
370 unsigned int sequence;
372 nf_ct_helper_destroy(ct);
376 sequence = read_seqcount_begin(&net->ct.generation);
377 hash = hash_conntrack(net,
378 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
379 reply_hash = hash_conntrack(net,
380 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
381 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
383 clean_from_lists(ct);
384 nf_conntrack_double_unlock(hash, reply_hash);
386 nf_ct_add_to_dying_list(ct);
388 NF_CT_STAT_INC(net, delete_list);
392 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
394 struct nf_conn_tstamp *tstamp;
396 tstamp = nf_conn_tstamp_find(ct);
397 if (tstamp && tstamp->stop == 0)
398 tstamp->stop = ktime_get_real_ns();
400 if (nf_ct_is_dying(ct))
403 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
404 portid, report) < 0) {
405 /* destroy event was not delivered */
406 nf_ct_delete_from_lists(ct);
407 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
411 nf_conntrack_ecache_work(nf_ct_net(ct));
412 set_bit(IPS_DYING_BIT, &ct->status);
414 nf_ct_delete_from_lists(ct);
418 EXPORT_SYMBOL_GPL(nf_ct_delete);
420 static void death_by_timeout(unsigned long ul_conntrack)
422 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
426 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
427 const struct nf_conntrack_tuple *tuple,
428 const struct nf_conntrack_zone *zone)
430 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
432 /* A conntrack can be recreated with the equal tuple,
433 * so we need to check that the conntrack is confirmed
435 return nf_ct_tuple_equal(tuple, &h->tuple) &&
436 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
437 nf_ct_is_confirmed(ct);
442 * - Caller must take a reference on returned object
443 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
445 static struct nf_conntrack_tuple_hash *
446 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
447 const struct nf_conntrack_tuple *tuple, u32 hash)
449 struct nf_conntrack_tuple_hash *h;
450 struct hlist_nulls_node *n;
451 unsigned int bucket = hash_bucket(hash, net);
453 /* Disable BHs the entire time since we normally need to disable them
454 * at least once for the stats anyway.
458 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
459 if (nf_ct_key_equal(h, tuple, zone)) {
460 NF_CT_STAT_INC(net, found);
464 NF_CT_STAT_INC(net, searched);
467 * if the nulls value we got at the end of this lookup is
468 * not the expected one, we must restart lookup.
469 * We probably met an item that was moved to another chain.
471 if (get_nulls_value(n) != bucket) {
472 NF_CT_STAT_INC(net, search_restart);
480 /* Find a connection corresponding to a tuple. */
481 static struct nf_conntrack_tuple_hash *
482 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
483 const struct nf_conntrack_tuple *tuple, u32 hash)
485 struct nf_conntrack_tuple_hash *h;
490 h = ____nf_conntrack_find(net, zone, tuple, hash);
492 ct = nf_ct_tuplehash_to_ctrack(h);
493 if (unlikely(nf_ct_is_dying(ct) ||
494 !atomic_inc_not_zero(&ct->ct_general.use)))
497 if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
508 struct nf_conntrack_tuple_hash *
509 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
510 const struct nf_conntrack_tuple *tuple)
512 return __nf_conntrack_find_get(net, zone, tuple,
513 hash_conntrack_raw(tuple));
515 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
517 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
519 unsigned int reply_hash)
521 struct net *net = nf_ct_net(ct);
523 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
524 &net->ct.hash[hash]);
525 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
526 &net->ct.hash[reply_hash]);
530 nf_conntrack_hash_check_insert(struct nf_conn *ct)
532 const struct nf_conntrack_zone *zone;
533 struct net *net = nf_ct_net(ct);
534 unsigned int hash, reply_hash;
535 struct nf_conntrack_tuple_hash *h;
536 struct hlist_nulls_node *n;
537 unsigned int sequence;
539 zone = nf_ct_zone(ct);
543 sequence = read_seqcount_begin(&net->ct.generation);
544 hash = hash_conntrack(net,
545 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
546 reply_hash = hash_conntrack(net,
547 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
548 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
550 /* See if there's one in the list already, including reverse */
551 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
552 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
554 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
557 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
558 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
560 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
564 add_timer(&ct->timeout);
566 /* The caller holds a reference to this object */
567 atomic_set(&ct->ct_general.use, 2);
568 __nf_conntrack_hash_insert(ct, hash, reply_hash);
569 nf_conntrack_double_unlock(hash, reply_hash);
570 NF_CT_STAT_INC(net, insert);
575 nf_conntrack_double_unlock(hash, reply_hash);
576 NF_CT_STAT_INC(net, insert_failed);
580 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
582 /* Confirm a connection given skb; places it in hash table */
584 __nf_conntrack_confirm(struct sk_buff *skb)
586 const struct nf_conntrack_zone *zone;
587 unsigned int hash, reply_hash;
588 struct nf_conntrack_tuple_hash *h;
590 struct nf_conn_help *help;
591 struct nf_conn_tstamp *tstamp;
592 struct hlist_nulls_node *n;
593 enum ip_conntrack_info ctinfo;
595 unsigned int sequence;
597 ct = nf_ct_get(skb, &ctinfo);
600 /* ipt_REJECT uses nf_conntrack_attach to attach related
601 ICMP/TCP RST packets in other direction. Actual packet
602 which created connection will be IP_CT_NEW or for an
603 expected connection, IP_CT_RELATED. */
604 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
607 zone = nf_ct_zone(ct);
611 sequence = read_seqcount_begin(&net->ct.generation);
612 /* reuse the hash saved before */
613 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
614 hash = hash_bucket(hash, net);
615 reply_hash = hash_conntrack(net,
616 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
618 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
620 /* We're not in hash table, and we refuse to set up related
621 * connections for unconfirmed conns. But packet copies and
622 * REJECT will give spurious warnings here.
624 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
626 /* No external references means no one else could have
629 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
630 pr_debug("Confirming conntrack %p\n", ct);
631 /* We have to check the DYING flag after unlink to prevent
632 * a race against nf_ct_get_next_corpse() possibly called from
633 * user context, else we insert an already 'dead' hash, blocking
634 * further use of that particular connection -JM.
636 nf_ct_del_from_dying_or_unconfirmed_list(ct);
638 if (unlikely(nf_ct_is_dying(ct)))
641 /* See if there's one in the list already, including reverse:
642 NAT could have grabbed it without realizing, since we're
643 not in the hash. If there is, we lost race. */
644 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
645 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
647 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
650 hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
651 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
653 nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
657 /* Timer relative to confirmation time, not original
658 setting time, otherwise we'd get timer wrap in
659 weird delay cases. */
660 ct->timeout.expires += jiffies;
661 add_timer(&ct->timeout);
662 atomic_inc(&ct->ct_general.use);
663 ct->status |= IPS_CONFIRMED;
665 /* set conntrack timestamp, if enabled. */
666 tstamp = nf_conn_tstamp_find(ct);
668 if (skb->tstamp.tv64 == 0)
669 __net_timestamp(skb);
671 tstamp->start = ktime_to_ns(skb->tstamp);
673 /* Since the lookup is lockless, hash insertion must be done after
674 * starting the timer and setting the CONFIRMED bit. The RCU barriers
675 * guarantee that no other CPU can find the conntrack before the above
676 * stores are visible.
678 __nf_conntrack_hash_insert(ct, hash, reply_hash);
679 nf_conntrack_double_unlock(hash, reply_hash);
680 NF_CT_STAT_INC(net, insert);
683 help = nfct_help(ct);
684 if (help && help->helper)
685 nf_conntrack_event_cache(IPCT_HELPER, ct);
687 nf_conntrack_event_cache(master_ct(ct) ?
688 IPCT_RELATED : IPCT_NEW, ct);
692 nf_ct_add_to_dying_list(ct);
693 nf_conntrack_double_unlock(hash, reply_hash);
694 NF_CT_STAT_INC(net, insert_failed);
698 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
700 /* Returns true if a connection correspondings to the tuple (required
703 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
704 const struct nf_conn *ignored_conntrack)
706 struct net *net = nf_ct_net(ignored_conntrack);
707 const struct nf_conntrack_zone *zone;
708 struct nf_conntrack_tuple_hash *h;
709 struct hlist_nulls_node *n;
713 zone = nf_ct_zone(ignored_conntrack);
714 hash = hash_conntrack(net, tuple);
716 /* Disable BHs the entire time since we need to disable them at
717 * least once for the stats anyway.
720 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
721 ct = nf_ct_tuplehash_to_ctrack(h);
722 if (ct != ignored_conntrack &&
723 nf_ct_tuple_equal(tuple, &h->tuple) &&
724 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h))) {
725 NF_CT_STAT_INC(net, found);
726 rcu_read_unlock_bh();
729 NF_CT_STAT_INC(net, searched);
731 rcu_read_unlock_bh();
735 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
737 #define NF_CT_EVICTION_RANGE 8
739 /* There's a small race here where we may free a just-assured
740 connection. Too bad: we're in trouble anyway. */
741 static noinline int early_drop(struct net *net, unsigned int _hash)
743 /* Use oldest entry, which is roughly LRU */
744 struct nf_conntrack_tuple_hash *h;
745 struct nf_conn *ct = NULL, *tmp;
746 struct hlist_nulls_node *n;
747 unsigned int i = 0, cnt = 0;
749 unsigned int hash, sequence;
754 sequence = read_seqcount_begin(&net->ct.generation);
755 hash = hash_bucket(_hash, net);
756 for (; i < net->ct.htable_size; i++) {
757 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
759 if (read_seqcount_retry(&net->ct.generation, sequence)) {
763 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
765 tmp = nf_ct_tuplehash_to_ctrack(h);
766 if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
767 !nf_ct_is_dying(tmp) &&
768 atomic_inc_not_zero(&tmp->ct_general.use)) {
775 hash = (hash + 1) % net->ct.htable_size;
778 if (ct || cnt >= NF_CT_EVICTION_RANGE)
787 if (del_timer(&ct->timeout)) {
788 if (nf_ct_delete(ct, 0, 0)) {
790 NF_CT_STAT_INC_ATOMIC(net, early_drop);
797 void init_nf_conntrack_hash_rnd(void)
802 * Why not initialize nf_conntrack_rnd in a "init()" function ?
803 * Because there isn't enough entropy when system initializing,
804 * and we initialize it as late as possible.
807 get_random_bytes(&rand, sizeof(rand));
809 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
812 static struct nf_conn *
813 __nf_conntrack_alloc(struct net *net,
814 const struct nf_conntrack_zone *zone,
815 const struct nf_conntrack_tuple *orig,
816 const struct nf_conntrack_tuple *repl,
821 if (unlikely(!nf_conntrack_hash_rnd)) {
822 init_nf_conntrack_hash_rnd();
823 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
824 hash = hash_conntrack_raw(orig);
827 /* We don't want any race condition at early drop stage */
828 atomic_inc(&net->ct.count);
830 if (nf_conntrack_max &&
831 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
832 if (!early_drop(net, hash)) {
833 atomic_dec(&net->ct.count);
834 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
835 return ERR_PTR(-ENOMEM);
840 * Do not use kmem_cache_zalloc(), as this cache uses
841 * SLAB_DESTROY_BY_RCU.
843 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
847 spin_lock_init(&ct->lock);
848 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
849 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
850 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
851 /* save hash for reusing when confirming */
852 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
854 /* Don't set timer yet: wait for confirmation */
855 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
856 write_pnet(&ct->ct_net, net);
857 memset(&ct->__nfct_init_offset[0], 0,
858 offsetof(struct nf_conn, proto) -
859 offsetof(struct nf_conn, __nfct_init_offset[0]));
861 if (zone && nf_ct_zone_add(ct, GFP_ATOMIC, zone) < 0)
864 /* Because we use RCU lookups, we set ct_general.use to zero before
865 * this is inserted in any list.
867 atomic_set(&ct->ct_general.use, 0);
870 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
872 atomic_dec(&net->ct.count);
873 return ERR_PTR(-ENOMEM);
876 struct nf_conn *nf_conntrack_alloc(struct net *net,
877 const struct nf_conntrack_zone *zone,
878 const struct nf_conntrack_tuple *orig,
879 const struct nf_conntrack_tuple *repl,
882 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
884 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
886 void nf_conntrack_free(struct nf_conn *ct)
888 struct net *net = nf_ct_net(ct);
890 /* A freed object has refcnt == 0, that's
891 * the golden rule for SLAB_DESTROY_BY_RCU
893 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
895 nf_ct_ext_destroy(ct);
897 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
898 smp_mb__before_atomic();
899 atomic_dec(&net->ct.count);
901 EXPORT_SYMBOL_GPL(nf_conntrack_free);
904 /* Allocate a new conntrack: we return -ENOMEM if classification
905 failed due to stress. Otherwise it really is unclassifiable. */
906 static struct nf_conntrack_tuple_hash *
907 init_conntrack(struct net *net, struct nf_conn *tmpl,
908 const struct nf_conntrack_tuple *tuple,
909 struct nf_conntrack_l3proto *l3proto,
910 struct nf_conntrack_l4proto *l4proto,
912 unsigned int dataoff, u32 hash)
915 struct nf_conn_help *help;
916 struct nf_conntrack_tuple repl_tuple;
917 struct nf_conntrack_ecache *ecache;
918 struct nf_conntrack_expect *exp = NULL;
919 const struct nf_conntrack_zone *zone;
920 struct nf_conn_timeout *timeout_ext;
921 struct nf_conntrack_zone tmp;
922 unsigned int *timeouts;
924 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
925 pr_debug("Can't invert tuple.\n");
929 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
930 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
933 return (struct nf_conntrack_tuple_hash *)ct;
935 if (tmpl && nfct_synproxy(tmpl)) {
936 nfct_seqadj_ext_add(ct);
937 nfct_synproxy_ext_add(ct);
940 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
942 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
944 timeouts = l4proto->get_timeouts(net);
946 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
947 nf_conntrack_free(ct);
948 pr_debug("init conntrack: can't track with proto module\n");
953 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
955 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
956 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
957 nf_ct_labels_ext_add(ct);
959 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
960 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
961 ecache ? ecache->expmask : 0,
965 if (net->ct.expect_count) {
966 spin_lock(&nf_conntrack_expect_lock);
967 exp = nf_ct_find_expectation(net, zone, tuple);
969 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
971 /* Welcome, Mr. Bond. We've been expecting you... */
972 __set_bit(IPS_EXPECTED_BIT, &ct->status);
973 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
974 ct->master = exp->master;
976 help = nf_ct_helper_ext_add(ct, exp->helper,
979 rcu_assign_pointer(help->helper, exp->helper);
982 #ifdef CONFIG_NF_CONNTRACK_MARK
983 ct->mark = exp->master->mark;
985 #ifdef CONFIG_NF_CONNTRACK_SECMARK
986 ct->secmark = exp->master->secmark;
988 NF_CT_STAT_INC(net, expect_new);
990 spin_unlock(&nf_conntrack_expect_lock);
993 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
994 NF_CT_STAT_INC(net, new);
997 /* Now it is inserted into the unconfirmed list, bump refcount */
998 nf_conntrack_get(&ct->ct_general);
999 nf_ct_add_to_unconfirmed_list(ct);
1005 exp->expectfn(ct, exp);
1006 nf_ct_expect_put(exp);
1009 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1012 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1013 static inline struct nf_conn *
1014 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1015 struct sk_buff *skb,
1016 unsigned int dataoff,
1019 struct nf_conntrack_l3proto *l3proto,
1020 struct nf_conntrack_l4proto *l4proto,
1022 enum ip_conntrack_info *ctinfo)
1024 const struct nf_conntrack_zone *zone;
1025 struct nf_conntrack_tuple tuple;
1026 struct nf_conntrack_tuple_hash *h;
1027 struct nf_conntrack_zone tmp;
1031 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1032 dataoff, l3num, protonum, &tuple, l3proto,
1034 pr_debug("resolve_normal_ct: Can't get tuple\n");
1038 /* look for tuple match */
1039 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1040 hash = hash_conntrack_raw(&tuple);
1041 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1043 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1044 skb, dataoff, hash);
1050 ct = nf_ct_tuplehash_to_ctrack(h);
1052 /* It exists; we have (non-exclusive) reference. */
1053 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1054 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1055 /* Please set reply bit if this packet OK */
1058 /* Once we've had two way comms, always ESTABLISHED. */
1059 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1060 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
1061 *ctinfo = IP_CT_ESTABLISHED;
1062 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1063 pr_debug("nf_conntrack_in: related packet for %p\n",
1065 *ctinfo = IP_CT_RELATED;
1067 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
1068 *ctinfo = IP_CT_NEW;
1072 skb->nfct = &ct->ct_general;
1073 skb->nfctinfo = *ctinfo;
1078 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1079 struct sk_buff *skb)
1081 struct nf_conn *ct, *tmpl = NULL;
1082 enum ip_conntrack_info ctinfo;
1083 struct nf_conntrack_l3proto *l3proto;
1084 struct nf_conntrack_l4proto *l4proto;
1085 unsigned int *timeouts;
1086 unsigned int dataoff;
1092 /* Previously seen (loopback or untracked)? Ignore. */
1093 tmpl = (struct nf_conn *)skb->nfct;
1094 if (!nf_ct_is_template(tmpl)) {
1095 NF_CT_STAT_INC_ATOMIC(net, ignore);
1101 /* rcu_read_lock()ed by nf_hook_slow */
1102 l3proto = __nf_ct_l3proto_find(pf);
1103 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1104 &dataoff, &protonum);
1106 pr_debug("not prepared to track yet or error occurred\n");
1107 NF_CT_STAT_INC_ATOMIC(net, error);
1108 NF_CT_STAT_INC_ATOMIC(net, invalid);
1113 l4proto = __nf_ct_l4proto_find(pf, protonum);
1115 /* It may be an special packet, error, unclean...
1116 * inverse of the return code tells to the netfilter
1117 * core what to do with the packet. */
1118 if (l4proto->error != NULL) {
1119 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1122 NF_CT_STAT_INC_ATOMIC(net, error);
1123 NF_CT_STAT_INC_ATOMIC(net, invalid);
1127 /* ICMP[v6] protocol trackers may assign one conntrack. */
1132 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1133 l3proto, l4proto, &set_reply, &ctinfo);
1135 /* Not valid part of a connection */
1136 NF_CT_STAT_INC_ATOMIC(net, invalid);
1142 /* Too stressed to deal. */
1143 NF_CT_STAT_INC_ATOMIC(net, drop);
1148 NF_CT_ASSERT(skb->nfct);
1150 /* Decide what timeout policy we want to apply to this flow. */
1151 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1153 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1155 /* Invalid: inverse of the return code tells
1156 * the netfilter core what to do */
1157 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1158 nf_conntrack_put(skb->nfct);
1160 NF_CT_STAT_INC_ATOMIC(net, invalid);
1161 if (ret == -NF_DROP)
1162 NF_CT_STAT_INC_ATOMIC(net, drop);
1167 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1168 nf_conntrack_event_cache(IPCT_REPLY, ct);
1171 /* Special case: we have to repeat this hook, assign the
1172 * template again to this packet. We assume that this packet
1173 * has no conntrack assigned. This is used by nf_ct_tcp. */
1174 if (ret == NF_REPEAT)
1175 skb->nfct = (struct nf_conntrack *)tmpl;
1182 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1184 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1185 const struct nf_conntrack_tuple *orig)
1190 ret = nf_ct_invert_tuple(inverse, orig,
1191 __nf_ct_l3proto_find(orig->src.l3num),
1192 __nf_ct_l4proto_find(orig->src.l3num,
1193 orig->dst.protonum));
1197 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1199 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1200 implicitly racy: see __nf_conntrack_confirm */
1201 void nf_conntrack_alter_reply(struct nf_conn *ct,
1202 const struct nf_conntrack_tuple *newreply)
1204 struct nf_conn_help *help = nfct_help(ct);
1206 /* Should be unconfirmed, so not in hash table yet */
1207 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1209 pr_debug("Altering reply tuple of %p to ", ct);
1210 nf_ct_dump_tuple(newreply);
1212 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1213 if (ct->master || (help && !hlist_empty(&help->expectations)))
1217 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1220 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1222 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1223 void __nf_ct_refresh_acct(struct nf_conn *ct,
1224 enum ip_conntrack_info ctinfo,
1225 const struct sk_buff *skb,
1226 unsigned long extra_jiffies,
1229 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1232 /* Only update if this is not a fixed timeout */
1233 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1236 /* If not in hash table, timer will not be active yet */
1237 if (!nf_ct_is_confirmed(ct)) {
1238 ct->timeout.expires = extra_jiffies;
1240 unsigned long newtime = jiffies + extra_jiffies;
1242 /* Only update the timeout if the new timeout is at least
1243 HZ jiffies from the old timeout. Need del_timer for race
1244 avoidance (may already be dying). */
1245 if (newtime - ct->timeout.expires >= HZ)
1246 mod_timer_pending(&ct->timeout, newtime);
1251 struct nf_conn_acct *acct;
1253 acct = nf_conn_acct_find(ct);
1255 struct nf_conn_counter *counter = acct->counter;
1257 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1258 atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1262 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1264 bool __nf_ct_kill_acct(struct nf_conn *ct,
1265 enum ip_conntrack_info ctinfo,
1266 const struct sk_buff *skb,
1270 struct nf_conn_acct *acct;
1272 acct = nf_conn_acct_find(ct);
1274 struct nf_conn_counter *counter = acct->counter;
1276 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1277 atomic64_add(skb->len - skb_network_offset(skb),
1278 &counter[CTINFO2DIR(ctinfo)].bytes);
1282 if (del_timer(&ct->timeout)) {
1283 ct->timeout.function((unsigned long)ct);
1288 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1290 #ifdef CONFIG_NF_CONNTRACK_ZONES
1291 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1292 .len = sizeof(struct nf_conntrack_zone),
1293 .align = __alignof__(struct nf_conntrack_zone),
1294 .id = NF_CT_EXT_ZONE,
1298 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1300 #include <linux/netfilter/nfnetlink.h>
1301 #include <linux/netfilter/nfnetlink_conntrack.h>
1302 #include <linux/mutex.h>
1304 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1305 * in ip_conntrack_core, since we don't want the protocols to autoload
1306 * or depend on ctnetlink */
1307 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1308 const struct nf_conntrack_tuple *tuple)
1310 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1311 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1312 goto nla_put_failure;
1318 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1320 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1321 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1322 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1324 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1326 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1327 struct nf_conntrack_tuple *t)
1329 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1332 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1333 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1337 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1339 int nf_ct_port_nlattr_tuple_size(void)
1341 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1343 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1346 /* Used by ipt_REJECT and ip6t_REJECT. */
1347 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1350 enum ip_conntrack_info ctinfo;
1352 /* This ICMP is in reverse direction to the packet which caused it */
1353 ct = nf_ct_get(skb, &ctinfo);
1354 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1355 ctinfo = IP_CT_RELATED_REPLY;
1357 ctinfo = IP_CT_RELATED;
1359 /* Attach to new skbuff, and increment count */
1360 nskb->nfct = &ct->ct_general;
1361 nskb->nfctinfo = ctinfo;
1362 nf_conntrack_get(nskb->nfct);
1365 /* Bring out ya dead! */
1366 static struct nf_conn *
1367 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1368 void *data, unsigned int *bucket)
1370 struct nf_conntrack_tuple_hash *h;
1372 struct hlist_nulls_node *n;
1376 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1377 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1380 if (*bucket < net->ct.htable_size) {
1381 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1382 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1384 ct = nf_ct_tuplehash_to_ctrack(h);
1393 for_each_possible_cpu(cpu) {
1394 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1396 spin_lock_bh(&pcpu->lock);
1397 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1398 ct = nf_ct_tuplehash_to_ctrack(h);
1400 set_bit(IPS_DYING_BIT, &ct->status);
1402 spin_unlock_bh(&pcpu->lock);
1406 atomic_inc(&ct->ct_general.use);
1412 void nf_ct_iterate_cleanup(struct net *net,
1413 int (*iter)(struct nf_conn *i, void *data),
1414 void *data, u32 portid, int report)
1417 unsigned int bucket = 0;
1419 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1420 /* Time to push up daises... */
1421 if (del_timer(&ct->timeout))
1422 nf_ct_delete(ct, portid, report);
1424 /* ... else the timer will get him soon. */
1429 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1431 static int kill_all(struct nf_conn *i, void *data)
1436 void nf_ct_free_hashtable(void *hash, unsigned int size)
1438 if (is_vmalloc_addr(hash))
1441 free_pages((unsigned long)hash,
1442 get_order(sizeof(struct hlist_head) * size));
1444 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1446 static int untrack_refs(void)
1450 for_each_possible_cpu(cpu) {
1451 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1453 cnt += atomic_read(&ct->ct_general.use) - 1;
1458 void nf_conntrack_cleanup_start(void)
1460 RCU_INIT_POINTER(ip_ct_attach, NULL);
1463 void nf_conntrack_cleanup_end(void)
1465 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1466 while (untrack_refs() > 0)
1469 #ifdef CONFIG_NF_CONNTRACK_ZONES
1470 nf_ct_extend_unregister(&nf_ct_zone_extend);
1472 nf_conntrack_proto_fini();
1473 nf_conntrack_seqadj_fini();
1474 nf_conntrack_labels_fini();
1475 nf_conntrack_helper_fini();
1476 nf_conntrack_timeout_fini();
1477 nf_conntrack_ecache_fini();
1478 nf_conntrack_tstamp_fini();
1479 nf_conntrack_acct_fini();
1480 nf_conntrack_expect_fini();
1484 * Mishearing the voices in his head, our hero wonders how he's
1485 * supposed to kill the mall.
1487 void nf_conntrack_cleanup_net(struct net *net)
1491 list_add(&net->exit_list, &single);
1492 nf_conntrack_cleanup_net_list(&single);
1495 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1501 * This makes sure all current packets have passed through
1502 * netfilter framework. Roll on, two-stage module
1508 list_for_each_entry(net, net_exit_list, exit_list) {
1509 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1510 if (atomic_read(&net->ct.count) != 0)
1515 goto i_see_dead_people;
1518 list_for_each_entry(net, net_exit_list, exit_list) {
1519 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1520 nf_conntrack_proto_pernet_fini(net);
1521 nf_conntrack_helper_pernet_fini(net);
1522 nf_conntrack_ecache_pernet_fini(net);
1523 nf_conntrack_tstamp_pernet_fini(net);
1524 nf_conntrack_acct_pernet_fini(net);
1525 nf_conntrack_expect_pernet_fini(net);
1526 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1527 kfree(net->ct.slabname);
1528 free_percpu(net->ct.stat);
1529 free_percpu(net->ct.pcpu_lists);
1533 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1535 struct hlist_nulls_head *hash;
1536 unsigned int nr_slots, i;
1539 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1540 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1541 sz = nr_slots * sizeof(struct hlist_nulls_head);
1542 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1548 for (i = 0; i < nr_slots; i++)
1549 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1553 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1555 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1558 unsigned int hashsize, old_size;
1559 struct hlist_nulls_head *hash, *old_hash;
1560 struct nf_conntrack_tuple_hash *h;
1563 if (current->nsproxy->net_ns != &init_net)
1566 /* On boot, we can set this without any fancy locking. */
1567 if (!nf_conntrack_htable_size)
1568 return param_set_uint(val, kp);
1570 rc = kstrtouint(val, 0, &hashsize);
1576 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1581 nf_conntrack_all_lock();
1582 write_seqcount_begin(&init_net.ct.generation);
1584 /* Lookups in the old hash might happen in parallel, which means we
1585 * might get false negatives during connection lookup. New connections
1586 * created because of a false negative won't make it into the hash
1587 * though since that required taking the locks.
1590 for (i = 0; i < init_net.ct.htable_size; i++) {
1591 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1592 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1593 struct nf_conntrack_tuple_hash, hnnode);
1594 ct = nf_ct_tuplehash_to_ctrack(h);
1595 hlist_nulls_del_rcu(&h->hnnode);
1596 bucket = __hash_conntrack(&h->tuple, hashsize);
1597 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1600 old_size = init_net.ct.htable_size;
1601 old_hash = init_net.ct.hash;
1603 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1604 init_net.ct.hash = hash;
1606 write_seqcount_end(&init_net.ct.generation);
1607 nf_conntrack_all_unlock();
1610 nf_ct_free_hashtable(old_hash, old_size);
1613 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1615 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1616 &nf_conntrack_htable_size, 0600);
1618 void nf_ct_untracked_status_or(unsigned long bits)
1622 for_each_possible_cpu(cpu)
1623 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1625 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1627 int nf_conntrack_init_start(void)
1632 for (i = 0; i < CONNTRACK_LOCKS; i++)
1633 spin_lock_init(&nf_conntrack_locks[i]);
1635 if (!nf_conntrack_htable_size) {
1636 /* Idea from tcp.c: use 1/16384 of memory.
1637 * On i386: 32MB machine has 512 buckets.
1638 * >= 1GB machines have 16384 buckets.
1639 * >= 4GB machines have 65536 buckets.
1641 nf_conntrack_htable_size
1642 = (((totalram_pages << PAGE_SHIFT) / 16384)
1643 / sizeof(struct hlist_head));
1644 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1645 nf_conntrack_htable_size = 65536;
1646 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1647 nf_conntrack_htable_size = 16384;
1648 if (nf_conntrack_htable_size < 32)
1649 nf_conntrack_htable_size = 32;
1651 /* Use a max. factor of four by default to get the same max as
1652 * with the old struct list_heads. When a table size is given
1653 * we use the old value of 8 to avoid reducing the max.
1657 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1659 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1660 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1663 ret = nf_conntrack_expect_init();
1667 ret = nf_conntrack_acct_init();
1671 ret = nf_conntrack_tstamp_init();
1675 ret = nf_conntrack_ecache_init();
1679 ret = nf_conntrack_timeout_init();
1683 ret = nf_conntrack_helper_init();
1687 ret = nf_conntrack_labels_init();
1691 ret = nf_conntrack_seqadj_init();
1695 #ifdef CONFIG_NF_CONNTRACK_ZONES
1696 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1700 ret = nf_conntrack_proto_init();
1704 /* Set up fake conntrack: to never be deleted, not in any hashes */
1705 for_each_possible_cpu(cpu) {
1706 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1707 write_pnet(&ct->ct_net, &init_net);
1708 atomic_set(&ct->ct_general.use, 1);
1710 /* - and look it like as a confirmed connection */
1711 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1715 #ifdef CONFIG_NF_CONNTRACK_ZONES
1716 nf_ct_extend_unregister(&nf_ct_zone_extend);
1719 nf_conntrack_seqadj_fini();
1721 nf_conntrack_labels_fini();
1723 nf_conntrack_helper_fini();
1725 nf_conntrack_timeout_fini();
1727 nf_conntrack_ecache_fini();
1729 nf_conntrack_tstamp_fini();
1731 nf_conntrack_acct_fini();
1733 nf_conntrack_expect_fini();
1738 void nf_conntrack_init_end(void)
1740 /* For use by REJECT target */
1741 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1742 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1746 * We need to use special "null" values, not used in hash table
1748 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1749 #define DYING_NULLS_VAL ((1<<30)+1)
1750 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1752 int nf_conntrack_init_net(struct net *net)
1757 atomic_set(&net->ct.count, 0);
1758 seqcount_init(&net->ct.generation);
1760 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1761 if (!net->ct.pcpu_lists)
1764 for_each_possible_cpu(cpu) {
1765 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1767 spin_lock_init(&pcpu->lock);
1768 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1769 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1772 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1774 goto err_pcpu_lists;
1776 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1777 if (!net->ct.slabname)
1780 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1781 sizeof(struct nf_conn), 0,
1782 SLAB_DESTROY_BY_RCU, NULL);
1783 if (!net->ct.nf_conntrack_cachep) {
1784 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1788 net->ct.htable_size = nf_conntrack_htable_size;
1789 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1790 if (!net->ct.hash) {
1791 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1794 ret = nf_conntrack_expect_pernet_init(net);
1797 ret = nf_conntrack_acct_pernet_init(net);
1800 ret = nf_conntrack_tstamp_pernet_init(net);
1803 ret = nf_conntrack_ecache_pernet_init(net);
1806 ret = nf_conntrack_helper_pernet_init(net);
1809 ret = nf_conntrack_proto_pernet_init(net);
1815 nf_conntrack_helper_pernet_fini(net);
1817 nf_conntrack_ecache_pernet_fini(net);
1819 nf_conntrack_tstamp_pernet_fini(net);
1821 nf_conntrack_acct_pernet_fini(net);
1823 nf_conntrack_expect_pernet_fini(net);
1825 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1827 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1829 kfree(net->ct.slabname);
1831 free_percpu(net->ct.stat);
1833 free_percpu(net->ct.pcpu_lists);