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 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
65 enum ip_conntrack_info ctinfo,
66 unsigned int protoff);
67 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
69 DEFINE_SPINLOCK(nf_conntrack_lock);
70 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
72 unsigned int nf_conntrack_htable_size __read_mostly;
73 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
75 unsigned int nf_conntrack_max __read_mostly;
76 EXPORT_SYMBOL_GPL(nf_conntrack_max);
78 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
79 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
81 unsigned int nf_conntrack_hash_rnd __read_mostly;
82 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
84 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
88 /* The direction must be ignored, so we hash everything up to the
89 * destination ports (which is a multiple of 4) and treat the last
90 * three bytes manually.
92 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
93 return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
94 (((__force __u16)tuple->dst.u.all << 16) |
95 tuple->dst.protonum));
98 static u32 __hash_bucket(u32 hash, unsigned int size)
100 return ((u64)hash * size) >> 32;
103 static u32 hash_bucket(u32 hash, const struct net *net)
105 return __hash_bucket(hash, net->ct.htable_size);
108 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
109 u16 zone, unsigned int size)
111 return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
114 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
115 const struct nf_conntrack_tuple *tuple)
117 return __hash_conntrack(tuple, zone, net->ct.htable_size);
121 nf_ct_get_tuple(const struct sk_buff *skb,
123 unsigned int dataoff,
126 struct nf_conntrack_tuple *tuple,
127 const struct nf_conntrack_l3proto *l3proto,
128 const struct nf_conntrack_l4proto *l4proto)
130 memset(tuple, 0, sizeof(*tuple));
132 tuple->src.l3num = l3num;
133 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
136 tuple->dst.protonum = protonum;
137 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
139 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
141 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
143 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
144 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
146 struct nf_conntrack_l3proto *l3proto;
147 struct nf_conntrack_l4proto *l4proto;
148 unsigned int protoff;
154 l3proto = __nf_ct_l3proto_find(l3num);
155 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
156 if (ret != NF_ACCEPT) {
161 l4proto = __nf_ct_l4proto_find(l3num, protonum);
163 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
169 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
172 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
173 const struct nf_conntrack_tuple *orig,
174 const struct nf_conntrack_l3proto *l3proto,
175 const struct nf_conntrack_l4proto *l4proto)
177 memset(inverse, 0, sizeof(*inverse));
179 inverse->src.l3num = orig->src.l3num;
180 if (l3proto->invert_tuple(inverse, orig) == 0)
183 inverse->dst.dir = !orig->dst.dir;
185 inverse->dst.protonum = orig->dst.protonum;
186 return l4proto->invert_tuple(inverse, orig);
188 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
191 clean_from_lists(struct nf_conn *ct)
193 pr_debug("clean_from_lists(%p)\n", ct);
194 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
195 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
197 /* Destroy all pending expectations */
198 nf_ct_remove_expectations(ct);
202 destroy_conntrack(struct nf_conntrack *nfct)
204 struct nf_conn *ct = (struct nf_conn *)nfct;
205 struct net *net = nf_ct_net(ct);
206 struct nf_conntrack_l4proto *l4proto;
208 pr_debug("destroy_conntrack(%p)\n", ct);
209 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
210 NF_CT_ASSERT(!timer_pending(&ct->timeout));
212 /* To make sure we don't get any weird locking issues here:
213 * destroy_conntrack() MUST NOT be called with a write lock
214 * to nf_conntrack_lock!!! -HW */
216 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
217 if (l4proto && l4proto->destroy)
218 l4proto->destroy(ct);
222 spin_lock_bh(&nf_conntrack_lock);
223 /* Expectations will have been removed in clean_from_lists,
224 * except TFTP can create an expectation on the first packet,
225 * before connection is in the list, so we need to clean here,
227 nf_ct_remove_expectations(ct);
229 /* We overload first tuple to link into unconfirmed or dying list.*/
230 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
231 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
233 NF_CT_STAT_INC(net, delete);
234 spin_unlock_bh(&nf_conntrack_lock);
237 nf_ct_put(ct->master);
239 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
240 nf_conntrack_free(ct);
243 static void nf_ct_delete_from_lists(struct nf_conn *ct)
245 struct net *net = nf_ct_net(ct);
247 nf_ct_helper_destroy(ct);
248 spin_lock_bh(&nf_conntrack_lock);
249 /* Inside lock so preempt is disabled on module removal path.
250 * Otherwise we can get spurious warnings. */
251 NF_CT_STAT_INC(net, delete_list);
252 clean_from_lists(ct);
253 /* add this conntrack to the dying list */
254 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
256 spin_unlock_bh(&nf_conntrack_lock);
259 static void death_by_event(unsigned long ul_conntrack)
261 struct nf_conn *ct = (void *)ul_conntrack;
262 struct net *net = nf_ct_net(ct);
263 struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
265 BUG_ON(ecache == NULL);
267 if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
268 /* bad luck, let's retry again */
269 ecache->timeout.expires = jiffies +
270 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
271 add_timer(&ecache->timeout);
274 /* we've got the event delivered, now it's dying */
275 set_bit(IPS_DYING_BIT, &ct->status);
279 static void nf_ct_dying_timeout(struct nf_conn *ct)
281 struct net *net = nf_ct_net(ct);
282 struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
284 BUG_ON(ecache == NULL);
286 /* set a new timer to retry event delivery */
287 setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
288 ecache->timeout.expires = jiffies +
289 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
290 add_timer(&ecache->timeout);
293 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
295 struct nf_conn_tstamp *tstamp;
297 tstamp = nf_conn_tstamp_find(ct);
298 if (tstamp && tstamp->stop == 0)
299 tstamp->stop = ktime_to_ns(ktime_get_real());
301 if (!nf_ct_is_dying(ct) &&
302 unlikely(nf_conntrack_event_report(IPCT_DESTROY, ct,
303 portid, report) < 0)) {
304 /* destroy event was not delivered */
305 nf_ct_delete_from_lists(ct);
306 nf_ct_dying_timeout(ct);
309 set_bit(IPS_DYING_BIT, &ct->status);
310 nf_ct_delete_from_lists(ct);
314 EXPORT_SYMBOL_GPL(nf_ct_delete);
316 static void death_by_timeout(unsigned long ul_conntrack)
318 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
323 * - Caller must take a reference on returned object
324 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
326 * - Caller must lock nf_conntrack_lock before calling this function
328 static struct nf_conntrack_tuple_hash *
329 ____nf_conntrack_find(struct net *net, u16 zone,
330 const struct nf_conntrack_tuple *tuple, u32 hash)
332 struct nf_conntrack_tuple_hash *h;
333 struct hlist_nulls_node *n;
334 unsigned int bucket = hash_bucket(hash, net);
336 /* Disable BHs the entire time since we normally need to disable them
337 * at least once for the stats anyway.
341 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
342 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
343 nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
344 NF_CT_STAT_INC(net, found);
348 NF_CT_STAT_INC(net, searched);
351 * if the nulls value we got at the end of this lookup is
352 * not the expected one, we must restart lookup.
353 * We probably met an item that was moved to another chain.
355 if (get_nulls_value(n) != bucket) {
356 NF_CT_STAT_INC(net, search_restart);
364 struct nf_conntrack_tuple_hash *
365 __nf_conntrack_find(struct net *net, u16 zone,
366 const struct nf_conntrack_tuple *tuple)
368 return ____nf_conntrack_find(net, zone, tuple,
369 hash_conntrack_raw(tuple, zone));
371 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
373 /* Find a connection corresponding to a tuple. */
374 static struct nf_conntrack_tuple_hash *
375 __nf_conntrack_find_get(struct net *net, u16 zone,
376 const struct nf_conntrack_tuple *tuple, u32 hash)
378 struct nf_conntrack_tuple_hash *h;
383 h = ____nf_conntrack_find(net, zone, tuple, hash);
385 ct = nf_ct_tuplehash_to_ctrack(h);
386 if (unlikely(nf_ct_is_dying(ct) ||
387 !atomic_inc_not_zero(&ct->ct_general.use)))
390 if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
391 nf_ct_zone(ct) != zone)) {
402 struct nf_conntrack_tuple_hash *
403 nf_conntrack_find_get(struct net *net, u16 zone,
404 const struct nf_conntrack_tuple *tuple)
406 return __nf_conntrack_find_get(net, zone, tuple,
407 hash_conntrack_raw(tuple, zone));
409 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
411 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
413 unsigned int repl_hash)
415 struct net *net = nf_ct_net(ct);
417 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
418 &net->ct.hash[hash]);
419 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
420 &net->ct.hash[repl_hash]);
424 nf_conntrack_hash_check_insert(struct nf_conn *ct)
426 struct net *net = nf_ct_net(ct);
427 unsigned int hash, repl_hash;
428 struct nf_conntrack_tuple_hash *h;
429 struct hlist_nulls_node *n;
432 zone = nf_ct_zone(ct);
433 hash = hash_conntrack(net, zone,
434 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
435 repl_hash = hash_conntrack(net, zone,
436 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
438 spin_lock_bh(&nf_conntrack_lock);
440 /* See if there's one in the list already, including reverse */
441 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
442 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
444 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
446 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
447 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
449 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
452 add_timer(&ct->timeout);
453 nf_conntrack_get(&ct->ct_general);
454 __nf_conntrack_hash_insert(ct, hash, repl_hash);
455 NF_CT_STAT_INC(net, insert);
456 spin_unlock_bh(&nf_conntrack_lock);
461 NF_CT_STAT_INC(net, insert_failed);
462 spin_unlock_bh(&nf_conntrack_lock);
465 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
467 /* Confirm a connection given skb; places it in hash table */
469 __nf_conntrack_confirm(struct sk_buff *skb)
471 unsigned int hash, repl_hash;
472 struct nf_conntrack_tuple_hash *h;
474 struct nf_conn_help *help;
475 struct nf_conn_tstamp *tstamp;
476 struct hlist_nulls_node *n;
477 enum ip_conntrack_info ctinfo;
481 ct = nf_ct_get(skb, &ctinfo);
484 /* ipt_REJECT uses nf_conntrack_attach to attach related
485 ICMP/TCP RST packets in other direction. Actual packet
486 which created connection will be IP_CT_NEW or for an
487 expected connection, IP_CT_RELATED. */
488 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
491 zone = nf_ct_zone(ct);
492 /* reuse the hash saved before */
493 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
494 hash = hash_bucket(hash, net);
495 repl_hash = hash_conntrack(net, zone,
496 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
498 /* We're not in hash table, and we refuse to set up related
499 connections for unconfirmed conns. But packet copies and
500 REJECT will give spurious warnings here. */
501 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
503 /* No external references means no one else could have
505 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
506 pr_debug("Confirming conntrack %p\n", ct);
508 spin_lock_bh(&nf_conntrack_lock);
510 /* We have to check the DYING flag inside the lock to prevent
511 a race against nf_ct_get_next_corpse() possibly called from
512 user context, else we insert an already 'dead' hash, blocking
513 further use of that particular connection -JM */
515 if (unlikely(nf_ct_is_dying(ct))) {
516 spin_unlock_bh(&nf_conntrack_lock);
520 /* See if there's one in the list already, including reverse:
521 NAT could have grabbed it without realizing, since we're
522 not in the hash. If there is, we lost race. */
523 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
524 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
526 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
528 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
529 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
531 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
534 /* Remove from unconfirmed list */
535 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
537 /* Timer relative to confirmation time, not original
538 setting time, otherwise we'd get timer wrap in
539 weird delay cases. */
540 ct->timeout.expires += jiffies;
541 add_timer(&ct->timeout);
542 atomic_inc(&ct->ct_general.use);
543 ct->status |= IPS_CONFIRMED;
545 /* set conntrack timestamp, if enabled. */
546 tstamp = nf_conn_tstamp_find(ct);
548 if (skb->tstamp.tv64 == 0)
549 __net_timestamp(skb);
551 tstamp->start = ktime_to_ns(skb->tstamp);
553 /* Since the lookup is lockless, hash insertion must be done after
554 * starting the timer and setting the CONFIRMED bit. The RCU barriers
555 * guarantee that no other CPU can find the conntrack before the above
556 * stores are visible.
558 __nf_conntrack_hash_insert(ct, hash, repl_hash);
559 NF_CT_STAT_INC(net, insert);
560 spin_unlock_bh(&nf_conntrack_lock);
562 help = nfct_help(ct);
563 if (help && help->helper)
564 nf_conntrack_event_cache(IPCT_HELPER, ct);
566 nf_conntrack_event_cache(master_ct(ct) ?
567 IPCT_RELATED : IPCT_NEW, ct);
571 NF_CT_STAT_INC(net, insert_failed);
572 spin_unlock_bh(&nf_conntrack_lock);
575 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
577 /* Returns true if a connection correspondings to the tuple (required
580 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
581 const struct nf_conn *ignored_conntrack)
583 struct net *net = nf_ct_net(ignored_conntrack);
584 struct nf_conntrack_tuple_hash *h;
585 struct hlist_nulls_node *n;
587 u16 zone = nf_ct_zone(ignored_conntrack);
588 unsigned int hash = hash_conntrack(net, zone, tuple);
590 /* Disable BHs the entire time since we need to disable them at
591 * least once for the stats anyway.
594 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
595 ct = nf_ct_tuplehash_to_ctrack(h);
596 if (ct != ignored_conntrack &&
597 nf_ct_tuple_equal(tuple, &h->tuple) &&
598 nf_ct_zone(ct) == zone) {
599 NF_CT_STAT_INC(net, found);
600 rcu_read_unlock_bh();
603 NF_CT_STAT_INC(net, searched);
605 rcu_read_unlock_bh();
609 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
611 #define NF_CT_EVICTION_RANGE 8
613 /* There's a small race here where we may free a just-assured
614 connection. Too bad: we're in trouble anyway. */
615 static noinline int early_drop(struct net *net, unsigned int hash)
617 /* Use oldest entry, which is roughly LRU */
618 struct nf_conntrack_tuple_hash *h;
619 struct nf_conn *ct = NULL, *tmp;
620 struct hlist_nulls_node *n;
621 unsigned int i, cnt = 0;
625 for (i = 0; i < net->ct.htable_size; i++) {
626 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
628 tmp = nf_ct_tuplehash_to_ctrack(h);
629 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
635 if (likely(!nf_ct_is_dying(ct) &&
636 atomic_inc_not_zero(&ct->ct_general.use)))
642 if (cnt >= NF_CT_EVICTION_RANGE)
645 hash = (hash + 1) % net->ct.htable_size;
652 if (del_timer(&ct->timeout)) {
653 if (nf_ct_delete(ct, 0, 0)) {
655 NF_CT_STAT_INC_ATOMIC(net, early_drop);
662 void init_nf_conntrack_hash_rnd(void)
667 * Why not initialize nf_conntrack_rnd in a "init()" function ?
668 * Because there isn't enough entropy when system initializing,
669 * and we initialize it as late as possible.
672 get_random_bytes(&rand, sizeof(rand));
674 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
677 static struct nf_conn *
678 __nf_conntrack_alloc(struct net *net, u16 zone,
679 const struct nf_conntrack_tuple *orig,
680 const struct nf_conntrack_tuple *repl,
685 if (unlikely(!nf_conntrack_hash_rnd)) {
686 init_nf_conntrack_hash_rnd();
687 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
688 hash = hash_conntrack_raw(orig, zone);
691 /* We don't want any race condition at early drop stage */
692 atomic_inc(&net->ct.count);
694 if (nf_conntrack_max &&
695 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
696 if (!early_drop(net, hash_bucket(hash, net))) {
697 atomic_dec(&net->ct.count);
698 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
699 return ERR_PTR(-ENOMEM);
704 * Do not use kmem_cache_zalloc(), as this cache uses
705 * SLAB_DESTROY_BY_RCU.
707 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
709 atomic_dec(&net->ct.count);
710 return ERR_PTR(-ENOMEM);
713 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
714 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
716 memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
717 offsetof(struct nf_conn, proto) -
718 offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
719 spin_lock_init(&ct->lock);
720 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
721 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
722 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
723 /* save hash for reusing when confirming */
724 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
725 /* Don't set timer yet: wait for confirmation */
726 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
727 write_pnet(&ct->ct_net, net);
728 #ifdef CONFIG_NF_CONNTRACK_ZONES
730 struct nf_conntrack_zone *nf_ct_zone;
732 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
735 nf_ct_zone->id = zone;
739 * changes to lookup keys must be done before setting refcnt to 1
742 atomic_set(&ct->ct_general.use, 1);
745 #ifdef CONFIG_NF_CONNTRACK_ZONES
747 atomic_dec(&net->ct.count);
748 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
749 return ERR_PTR(-ENOMEM);
753 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
754 const struct nf_conntrack_tuple *orig,
755 const struct nf_conntrack_tuple *repl,
758 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
760 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
762 void nf_conntrack_free(struct nf_conn *ct)
764 struct net *net = nf_ct_net(ct);
766 nf_ct_ext_destroy(ct);
767 atomic_dec(&net->ct.count);
769 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
771 EXPORT_SYMBOL_GPL(nf_conntrack_free);
774 /* Allocate a new conntrack: we return -ENOMEM if classification
775 failed due to stress. Otherwise it really is unclassifiable. */
776 static struct nf_conntrack_tuple_hash *
777 init_conntrack(struct net *net, struct nf_conn *tmpl,
778 const struct nf_conntrack_tuple *tuple,
779 struct nf_conntrack_l3proto *l3proto,
780 struct nf_conntrack_l4proto *l4proto,
782 unsigned int dataoff, u32 hash)
785 struct nf_conn_help *help;
786 struct nf_conntrack_tuple repl_tuple;
787 struct nf_conntrack_ecache *ecache;
788 struct nf_conntrack_expect *exp;
789 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
790 struct nf_conn_timeout *timeout_ext;
791 unsigned int *timeouts;
793 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
794 pr_debug("Can't invert tuple.\n");
798 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
801 return (struct nf_conntrack_tuple_hash *)ct;
803 if (tmpl && nfct_synproxy(tmpl)) {
804 nfct_seqadj_ext_add(ct);
805 nfct_synproxy_ext_add(ct);
808 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
810 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
812 timeouts = l4proto->get_timeouts(net);
814 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
815 nf_conntrack_free(ct);
816 pr_debug("init conntrack: can't track with proto module\n");
821 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
823 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
824 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
825 nf_ct_labels_ext_add(ct);
827 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
828 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
829 ecache ? ecache->expmask : 0,
832 spin_lock_bh(&nf_conntrack_lock);
833 exp = nf_ct_find_expectation(net, zone, tuple);
835 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
837 /* Welcome, Mr. Bond. We've been expecting you... */
838 __set_bit(IPS_EXPECTED_BIT, &ct->status);
839 ct->master = exp->master;
841 help = nf_ct_helper_ext_add(ct, exp->helper,
844 rcu_assign_pointer(help->helper, exp->helper);
847 #ifdef CONFIG_NF_CONNTRACK_MARK
848 ct->mark = exp->master->mark;
850 #ifdef CONFIG_NF_CONNTRACK_SECMARK
851 ct->secmark = exp->master->secmark;
853 nf_conntrack_get(&ct->master->ct_general);
854 NF_CT_STAT_INC(net, expect_new);
856 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
857 NF_CT_STAT_INC(net, new);
860 /* Overload tuple linked list to put us in unconfirmed list. */
861 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
862 &net->ct.unconfirmed);
864 spin_unlock_bh(&nf_conntrack_lock);
868 exp->expectfn(ct, exp);
869 nf_ct_expect_put(exp);
872 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
875 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
876 static inline struct nf_conn *
877 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
879 unsigned int dataoff,
882 struct nf_conntrack_l3proto *l3proto,
883 struct nf_conntrack_l4proto *l4proto,
885 enum ip_conntrack_info *ctinfo)
887 struct nf_conntrack_tuple tuple;
888 struct nf_conntrack_tuple_hash *h;
890 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
893 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
894 dataoff, l3num, protonum, &tuple, l3proto,
896 pr_debug("resolve_normal_ct: Can't get tuple\n");
900 /* look for tuple match */
901 hash = hash_conntrack_raw(&tuple, zone);
902 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
904 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
911 ct = nf_ct_tuplehash_to_ctrack(h);
913 /* It exists; we have (non-exclusive) reference. */
914 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
915 *ctinfo = IP_CT_ESTABLISHED_REPLY;
916 /* Please set reply bit if this packet OK */
919 /* Once we've had two way comms, always ESTABLISHED. */
920 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
921 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
922 *ctinfo = IP_CT_ESTABLISHED;
923 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
924 pr_debug("nf_conntrack_in: related packet for %p\n",
926 *ctinfo = IP_CT_RELATED;
928 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
933 skb->nfct = &ct->ct_general;
934 skb->nfctinfo = *ctinfo;
939 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
942 struct nf_conn *ct, *tmpl = NULL;
943 enum ip_conntrack_info ctinfo;
944 struct nf_conntrack_l3proto *l3proto;
945 struct nf_conntrack_l4proto *l4proto;
946 unsigned int *timeouts;
947 unsigned int dataoff;
953 /* Previously seen (loopback or untracked)? Ignore. */
954 tmpl = (struct nf_conn *)skb->nfct;
955 if (!nf_ct_is_template(tmpl)) {
956 NF_CT_STAT_INC_ATOMIC(net, ignore);
962 /* rcu_read_lock()ed by nf_hook_slow */
963 l3proto = __nf_ct_l3proto_find(pf);
964 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
965 &dataoff, &protonum);
967 pr_debug("not prepared to track yet or error occurred\n");
968 NF_CT_STAT_INC_ATOMIC(net, error);
969 NF_CT_STAT_INC_ATOMIC(net, invalid);
974 l4proto = __nf_ct_l4proto_find(pf, protonum);
976 /* It may be an special packet, error, unclean...
977 * inverse of the return code tells to the netfilter
978 * core what to do with the packet. */
979 if (l4proto->error != NULL) {
980 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
983 NF_CT_STAT_INC_ATOMIC(net, error);
984 NF_CT_STAT_INC_ATOMIC(net, invalid);
988 /* ICMP[v6] protocol trackers may assign one conntrack. */
993 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
994 l3proto, l4proto, &set_reply, &ctinfo);
996 /* Not valid part of a connection */
997 NF_CT_STAT_INC_ATOMIC(net, invalid);
1003 /* Too stressed to deal. */
1004 NF_CT_STAT_INC_ATOMIC(net, drop);
1009 NF_CT_ASSERT(skb->nfct);
1011 /* Decide what timeout policy we want to apply to this flow. */
1012 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1014 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1016 /* Invalid: inverse of the return code tells
1017 * the netfilter core what to do */
1018 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1019 nf_conntrack_put(skb->nfct);
1021 NF_CT_STAT_INC_ATOMIC(net, invalid);
1022 if (ret == -NF_DROP)
1023 NF_CT_STAT_INC_ATOMIC(net, drop);
1028 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1029 nf_conntrack_event_cache(IPCT_REPLY, ct);
1032 /* Special case: we have to repeat this hook, assign the
1033 * template again to this packet. We assume that this packet
1034 * has no conntrack assigned. This is used by nf_ct_tcp. */
1035 if (ret == NF_REPEAT)
1036 skb->nfct = (struct nf_conntrack *)tmpl;
1043 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1045 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1046 const struct nf_conntrack_tuple *orig)
1051 ret = nf_ct_invert_tuple(inverse, orig,
1052 __nf_ct_l3proto_find(orig->src.l3num),
1053 __nf_ct_l4proto_find(orig->src.l3num,
1054 orig->dst.protonum));
1058 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1060 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1061 implicitly racy: see __nf_conntrack_confirm */
1062 void nf_conntrack_alter_reply(struct nf_conn *ct,
1063 const struct nf_conntrack_tuple *newreply)
1065 struct nf_conn_help *help = nfct_help(ct);
1067 /* Should be unconfirmed, so not in hash table yet */
1068 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1070 pr_debug("Altering reply tuple of %p to ", ct);
1071 nf_ct_dump_tuple(newreply);
1073 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1074 if (ct->master || (help && !hlist_empty(&help->expectations)))
1078 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1081 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1083 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1084 void __nf_ct_refresh_acct(struct nf_conn *ct,
1085 enum ip_conntrack_info ctinfo,
1086 const struct sk_buff *skb,
1087 unsigned long extra_jiffies,
1090 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1093 /* Only update if this is not a fixed timeout */
1094 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1097 /* If not in hash table, timer will not be active yet */
1098 if (!nf_ct_is_confirmed(ct)) {
1099 ct->timeout.expires = extra_jiffies;
1101 unsigned long newtime = jiffies + extra_jiffies;
1103 /* Only update the timeout if the new timeout is at least
1104 HZ jiffies from the old timeout. Need del_timer for race
1105 avoidance (may already be dying). */
1106 if (newtime - ct->timeout.expires >= HZ)
1107 mod_timer_pending(&ct->timeout, newtime);
1112 struct nf_conn_counter *acct;
1114 acct = nf_conn_acct_find(ct);
1116 atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1117 atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1121 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1123 bool __nf_ct_kill_acct(struct nf_conn *ct,
1124 enum ip_conntrack_info ctinfo,
1125 const struct sk_buff *skb,
1129 struct nf_conn_counter *acct;
1131 acct = nf_conn_acct_find(ct);
1133 atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1134 atomic64_add(skb->len - skb_network_offset(skb),
1135 &acct[CTINFO2DIR(ctinfo)].bytes);
1139 if (del_timer(&ct->timeout)) {
1140 ct->timeout.function((unsigned long)ct);
1145 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1147 #ifdef CONFIG_NF_CONNTRACK_ZONES
1148 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1149 .len = sizeof(struct nf_conntrack_zone),
1150 .align = __alignof__(struct nf_conntrack_zone),
1151 .id = NF_CT_EXT_ZONE,
1155 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1157 #include <linux/netfilter/nfnetlink.h>
1158 #include <linux/netfilter/nfnetlink_conntrack.h>
1159 #include <linux/mutex.h>
1161 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1162 * in ip_conntrack_core, since we don't want the protocols to autoload
1163 * or depend on ctnetlink */
1164 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1165 const struct nf_conntrack_tuple *tuple)
1167 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1168 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1169 goto nla_put_failure;
1175 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1177 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1178 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1179 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1181 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1183 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1184 struct nf_conntrack_tuple *t)
1186 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1189 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1190 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1194 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1196 int nf_ct_port_nlattr_tuple_size(void)
1198 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1200 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1203 /* Used by ipt_REJECT and ip6t_REJECT. */
1204 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1207 enum ip_conntrack_info ctinfo;
1209 /* This ICMP is in reverse direction to the packet which caused it */
1210 ct = nf_ct_get(skb, &ctinfo);
1211 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1212 ctinfo = IP_CT_RELATED_REPLY;
1214 ctinfo = IP_CT_RELATED;
1216 /* Attach to new skbuff, and increment count */
1217 nskb->nfct = &ct->ct_general;
1218 nskb->nfctinfo = ctinfo;
1219 nf_conntrack_get(nskb->nfct);
1222 /* Bring out ya dead! */
1223 static struct nf_conn *
1224 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1225 void *data, unsigned int *bucket)
1227 struct nf_conntrack_tuple_hash *h;
1229 struct hlist_nulls_node *n;
1231 spin_lock_bh(&nf_conntrack_lock);
1232 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1233 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1234 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1236 ct = nf_ct_tuplehash_to_ctrack(h);
1241 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1242 ct = nf_ct_tuplehash_to_ctrack(h);
1244 set_bit(IPS_DYING_BIT, &ct->status);
1246 spin_unlock_bh(&nf_conntrack_lock);
1249 atomic_inc(&ct->ct_general.use);
1250 spin_unlock_bh(&nf_conntrack_lock);
1254 void nf_ct_iterate_cleanup(struct net *net,
1255 int (*iter)(struct nf_conn *i, void *data),
1256 void *data, u32 portid, int report)
1259 unsigned int bucket = 0;
1261 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1262 /* Time to push up daises... */
1263 if (del_timer(&ct->timeout))
1264 nf_ct_delete(ct, portid, report);
1266 /* ... else the timer will get him soon. */
1271 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1273 static int kill_all(struct nf_conn *i, void *data)
1278 void nf_ct_free_hashtable(void *hash, unsigned int size)
1280 if (is_vmalloc_addr(hash))
1283 free_pages((unsigned long)hash,
1284 get_order(sizeof(struct hlist_head) * size));
1286 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1288 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1290 nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1292 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1294 static void nf_ct_release_dying_list(struct net *net)
1296 struct nf_conntrack_tuple_hash *h;
1298 struct hlist_nulls_node *n;
1300 spin_lock_bh(&nf_conntrack_lock);
1301 hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1302 ct = nf_ct_tuplehash_to_ctrack(h);
1303 /* never fails to remove them, no listeners at this point */
1306 spin_unlock_bh(&nf_conntrack_lock);
1309 static int untrack_refs(void)
1313 for_each_possible_cpu(cpu) {
1314 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1316 cnt += atomic_read(&ct->ct_general.use) - 1;
1321 void nf_conntrack_cleanup_start(void)
1323 RCU_INIT_POINTER(ip_ct_attach, NULL);
1326 void nf_conntrack_cleanup_end(void)
1328 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1329 while (untrack_refs() > 0)
1332 #ifdef CONFIG_NF_CONNTRACK_ZONES
1333 nf_ct_extend_unregister(&nf_ct_zone_extend);
1335 nf_conntrack_proto_fini();
1336 nf_conntrack_seqadj_fini();
1337 nf_conntrack_labels_fini();
1338 nf_conntrack_helper_fini();
1339 nf_conntrack_timeout_fini();
1340 nf_conntrack_ecache_fini();
1341 nf_conntrack_tstamp_fini();
1342 nf_conntrack_acct_fini();
1343 nf_conntrack_expect_fini();
1347 * Mishearing the voices in his head, our hero wonders how he's
1348 * supposed to kill the mall.
1350 void nf_conntrack_cleanup_net(struct net *net)
1354 list_add(&net->exit_list, &single);
1355 nf_conntrack_cleanup_net_list(&single);
1358 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1364 * This makes sure all current packets have passed through
1365 * netfilter framework. Roll on, two-stage module
1371 list_for_each_entry(net, net_exit_list, exit_list) {
1372 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1373 nf_ct_release_dying_list(net);
1374 if (atomic_read(&net->ct.count) != 0)
1379 goto i_see_dead_people;
1382 list_for_each_entry(net, net_exit_list, exit_list) {
1383 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1384 nf_conntrack_proto_pernet_fini(net);
1385 nf_conntrack_helper_pernet_fini(net);
1386 nf_conntrack_ecache_pernet_fini(net);
1387 nf_conntrack_tstamp_pernet_fini(net);
1388 nf_conntrack_acct_pernet_fini(net);
1389 nf_conntrack_expect_pernet_fini(net);
1390 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1391 kfree(net->ct.slabname);
1392 free_percpu(net->ct.stat);
1396 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1398 struct hlist_nulls_head *hash;
1399 unsigned int nr_slots, i;
1402 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1403 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1404 sz = nr_slots * sizeof(struct hlist_nulls_head);
1405 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1408 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1413 for (i = 0; i < nr_slots; i++)
1414 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1418 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1420 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1423 unsigned int hashsize, old_size;
1424 struct hlist_nulls_head *hash, *old_hash;
1425 struct nf_conntrack_tuple_hash *h;
1428 if (current->nsproxy->net_ns != &init_net)
1431 /* On boot, we can set this without any fancy locking. */
1432 if (!nf_conntrack_htable_size)
1433 return param_set_uint(val, kp);
1435 rc = kstrtouint(val, 0, &hashsize);
1441 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1445 /* Lookups in the old hash might happen in parallel, which means we
1446 * might get false negatives during connection lookup. New connections
1447 * created because of a false negative won't make it into the hash
1448 * though since that required taking the lock.
1450 spin_lock_bh(&nf_conntrack_lock);
1451 for (i = 0; i < init_net.ct.htable_size; i++) {
1452 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1453 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1454 struct nf_conntrack_tuple_hash, hnnode);
1455 ct = nf_ct_tuplehash_to_ctrack(h);
1456 hlist_nulls_del_rcu(&h->hnnode);
1457 bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1459 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1462 old_size = init_net.ct.htable_size;
1463 old_hash = init_net.ct.hash;
1465 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1466 init_net.ct.hash = hash;
1467 spin_unlock_bh(&nf_conntrack_lock);
1469 nf_ct_free_hashtable(old_hash, old_size);
1472 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1474 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1475 &nf_conntrack_htable_size, 0600);
1477 void nf_ct_untracked_status_or(unsigned long bits)
1481 for_each_possible_cpu(cpu)
1482 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1484 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1486 int nf_conntrack_init_start(void)
1491 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1492 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1493 if (!nf_conntrack_htable_size) {
1494 nf_conntrack_htable_size
1495 = (((totalram_pages << PAGE_SHIFT) / 16384)
1496 / sizeof(struct hlist_head));
1497 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1498 nf_conntrack_htable_size = 16384;
1499 if (nf_conntrack_htable_size < 32)
1500 nf_conntrack_htable_size = 32;
1502 /* Use a max. factor of four by default to get the same max as
1503 * with the old struct list_heads. When a table size is given
1504 * we use the old value of 8 to avoid reducing the max.
1508 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1510 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1511 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1514 ret = nf_conntrack_expect_init();
1518 ret = nf_conntrack_acct_init();
1522 ret = nf_conntrack_tstamp_init();
1526 ret = nf_conntrack_ecache_init();
1530 ret = nf_conntrack_timeout_init();
1534 ret = nf_conntrack_helper_init();
1538 ret = nf_conntrack_labels_init();
1542 ret = nf_conntrack_seqadj_init();
1546 #ifdef CONFIG_NF_CONNTRACK_ZONES
1547 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1551 ret = nf_conntrack_proto_init();
1555 /* Set up fake conntrack: to never be deleted, not in any hashes */
1556 for_each_possible_cpu(cpu) {
1557 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1558 write_pnet(&ct->ct_net, &init_net);
1559 atomic_set(&ct->ct_general.use, 1);
1561 /* - and look it like as a confirmed connection */
1562 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1566 #ifdef CONFIG_NF_CONNTRACK_ZONES
1567 nf_ct_extend_unregister(&nf_ct_zone_extend);
1570 nf_conntrack_seqadj_fini();
1572 nf_conntrack_labels_fini();
1574 nf_conntrack_helper_fini();
1576 nf_conntrack_timeout_fini();
1578 nf_conntrack_ecache_fini();
1580 nf_conntrack_tstamp_fini();
1582 nf_conntrack_acct_fini();
1584 nf_conntrack_expect_fini();
1589 void nf_conntrack_init_end(void)
1591 /* For use by REJECT target */
1592 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1593 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1597 * We need to use special "null" values, not used in hash table
1599 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1600 #define DYING_NULLS_VAL ((1<<30)+1)
1601 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1603 int nf_conntrack_init_net(struct net *net)
1607 atomic_set(&net->ct.count, 0);
1608 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1609 INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1610 INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1611 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1612 if (!net->ct.stat) {
1617 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1618 if (!net->ct.slabname) {
1623 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1624 sizeof(struct nf_conn), 0,
1625 SLAB_DESTROY_BY_RCU, NULL);
1626 if (!net->ct.nf_conntrack_cachep) {
1627 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1632 net->ct.htable_size = nf_conntrack_htable_size;
1633 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1634 if (!net->ct.hash) {
1636 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1639 ret = nf_conntrack_expect_pernet_init(net);
1642 ret = nf_conntrack_acct_pernet_init(net);
1645 ret = nf_conntrack_tstamp_pernet_init(net);
1648 ret = nf_conntrack_ecache_pernet_init(net);
1651 ret = nf_conntrack_helper_pernet_init(net);
1654 ret = nf_conntrack_proto_pernet_init(net);
1660 nf_conntrack_helper_pernet_fini(net);
1662 nf_conntrack_ecache_pernet_fini(net);
1664 nf_conntrack_tstamp_pernet_fini(net);
1666 nf_conntrack_acct_pernet_fini(net);
1668 nf_conntrack_expect_pernet_fini(net);
1670 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1672 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1674 kfree(net->ct.slabname);
1676 free_percpu(net->ct.stat);