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 DEFINE_SPINLOCK(nf_conntrack_lock);
64 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
66 unsigned int nf_conntrack_htable_size __read_mostly;
67 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
69 unsigned int nf_conntrack_max __read_mostly;
70 EXPORT_SYMBOL_GPL(nf_conntrack_max);
72 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
73 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
75 unsigned int nf_conntrack_hash_rnd __read_mostly;
76 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
78 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
82 /* The direction must be ignored, so we hash everything up to the
83 * destination ports (which is a multiple of 4) and treat the last
84 * three bytes manually.
86 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
87 return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
88 (((__force __u16)tuple->dst.u.all << 16) |
89 tuple->dst.protonum));
92 static u32 __hash_bucket(u32 hash, unsigned int size)
94 return ((u64)hash * size) >> 32;
97 static u32 hash_bucket(u32 hash, const struct net *net)
99 return __hash_bucket(hash, net->ct.htable_size);
102 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
103 u16 zone, unsigned int size)
105 return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
108 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
109 const struct nf_conntrack_tuple *tuple)
111 return __hash_conntrack(tuple, zone, net->ct.htable_size);
115 nf_ct_get_tuple(const struct sk_buff *skb,
117 unsigned int dataoff,
120 struct nf_conntrack_tuple *tuple,
121 const struct nf_conntrack_l3proto *l3proto,
122 const struct nf_conntrack_l4proto *l4proto)
124 memset(tuple, 0, sizeof(*tuple));
126 tuple->src.l3num = l3num;
127 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
130 tuple->dst.protonum = protonum;
131 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
133 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
135 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
137 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
138 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
140 struct nf_conntrack_l3proto *l3proto;
141 struct nf_conntrack_l4proto *l4proto;
142 unsigned int protoff;
148 l3proto = __nf_ct_l3proto_find(l3num);
149 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
150 if (ret != NF_ACCEPT) {
155 l4proto = __nf_ct_l4proto_find(l3num, protonum);
157 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
163 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
166 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
167 const struct nf_conntrack_tuple *orig,
168 const struct nf_conntrack_l3proto *l3proto,
169 const struct nf_conntrack_l4proto *l4proto)
171 memset(inverse, 0, sizeof(*inverse));
173 inverse->src.l3num = orig->src.l3num;
174 if (l3proto->invert_tuple(inverse, orig) == 0)
177 inverse->dst.dir = !orig->dst.dir;
179 inverse->dst.protonum = orig->dst.protonum;
180 return l4proto->invert_tuple(inverse, orig);
182 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
185 clean_from_lists(struct nf_conn *ct)
187 pr_debug("clean_from_lists(%p)\n", ct);
188 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
189 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
191 /* Destroy all pending expectations */
192 nf_ct_remove_expectations(ct);
196 destroy_conntrack(struct nf_conntrack *nfct)
198 struct nf_conn *ct = (struct nf_conn *)nfct;
199 struct net *net = nf_ct_net(ct);
200 struct nf_conntrack_l4proto *l4proto;
202 pr_debug("destroy_conntrack(%p)\n", ct);
203 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
204 NF_CT_ASSERT(!timer_pending(&ct->timeout));
206 /* To make sure we don't get any weird locking issues here:
207 * destroy_conntrack() MUST NOT be called with a write lock
208 * to nf_conntrack_lock!!! -HW */
210 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
211 if (l4proto && l4proto->destroy)
212 l4proto->destroy(ct);
216 spin_lock_bh(&nf_conntrack_lock);
217 /* Expectations will have been removed in clean_from_lists,
218 * except TFTP can create an expectation on the first packet,
219 * before connection is in the list, so we need to clean here,
221 nf_ct_remove_expectations(ct);
223 /* We overload first tuple to link into unconfirmed or dying list.*/
224 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
225 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
227 NF_CT_STAT_INC(net, delete);
228 spin_unlock_bh(&nf_conntrack_lock);
231 nf_ct_put(ct->master);
233 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
234 nf_conntrack_free(ct);
237 static void nf_ct_delete_from_lists(struct nf_conn *ct)
239 struct net *net = nf_ct_net(ct);
241 nf_ct_helper_destroy(ct);
242 spin_lock_bh(&nf_conntrack_lock);
243 /* Inside lock so preempt is disabled on module removal path.
244 * Otherwise we can get spurious warnings. */
245 NF_CT_STAT_INC(net, delete_list);
246 clean_from_lists(ct);
247 /* add this conntrack to the dying list */
248 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
250 spin_unlock_bh(&nf_conntrack_lock);
253 static void death_by_event(unsigned long ul_conntrack)
255 struct nf_conn *ct = (void *)ul_conntrack;
256 struct net *net = nf_ct_net(ct);
257 struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
259 BUG_ON(ecache == NULL);
261 if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
262 /* bad luck, let's retry again */
263 ecache->timeout.expires = jiffies +
264 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
265 add_timer(&ecache->timeout);
268 /* we've got the event delivered, now it's dying */
269 set_bit(IPS_DYING_BIT, &ct->status);
273 static void nf_ct_dying_timeout(struct nf_conn *ct)
275 struct net *net = nf_ct_net(ct);
276 struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
278 BUG_ON(ecache == NULL);
280 /* set a new timer to retry event delivery */
281 setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
282 ecache->timeout.expires = jiffies +
283 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
284 add_timer(&ecache->timeout);
287 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
289 struct nf_conn_tstamp *tstamp;
291 tstamp = nf_conn_tstamp_find(ct);
292 if (tstamp && tstamp->stop == 0)
293 tstamp->stop = ktime_to_ns(ktime_get_real());
295 if (!nf_ct_is_dying(ct) &&
296 unlikely(nf_conntrack_event_report(IPCT_DESTROY, ct,
297 portid, report) < 0)) {
298 /* destroy event was not delivered */
299 nf_ct_delete_from_lists(ct);
300 nf_ct_dying_timeout(ct);
303 set_bit(IPS_DYING_BIT, &ct->status);
304 nf_ct_delete_from_lists(ct);
308 EXPORT_SYMBOL_GPL(nf_ct_delete);
310 static void death_by_timeout(unsigned long ul_conntrack)
312 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
316 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
317 const struct nf_conntrack_tuple *tuple,
320 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
322 /* A conntrack can be recreated with the equal tuple,
323 * so we need to check that the conntrack is confirmed
325 return nf_ct_tuple_equal(tuple, &h->tuple) &&
326 nf_ct_zone(ct) == zone &&
327 nf_ct_is_confirmed(ct);
332 * - Caller must take a reference on returned object
333 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
335 * - Caller must lock nf_conntrack_lock before calling this function
337 static struct nf_conntrack_tuple_hash *
338 ____nf_conntrack_find(struct net *net, u16 zone,
339 const struct nf_conntrack_tuple *tuple, u32 hash)
341 struct nf_conntrack_tuple_hash *h;
342 struct hlist_nulls_node *n;
343 unsigned int bucket = hash_bucket(hash, net);
345 /* Disable BHs the entire time since we normally need to disable them
346 * at least once for the stats anyway.
350 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
351 if (nf_ct_key_equal(h, tuple, zone)) {
352 NF_CT_STAT_INC(net, found);
356 NF_CT_STAT_INC(net, searched);
359 * if the nulls value we got at the end of this lookup is
360 * not the expected one, we must restart lookup.
361 * We probably met an item that was moved to another chain.
363 if (get_nulls_value(n) != bucket) {
364 NF_CT_STAT_INC(net, search_restart);
372 /* Find a connection corresponding to a tuple. */
373 static struct nf_conntrack_tuple_hash *
374 __nf_conntrack_find_get(struct net *net, u16 zone,
375 const struct nf_conntrack_tuple *tuple, u32 hash)
377 struct nf_conntrack_tuple_hash *h;
382 h = ____nf_conntrack_find(net, zone, tuple, hash);
384 ct = nf_ct_tuplehash_to_ctrack(h);
385 if (unlikely(nf_ct_is_dying(ct) ||
386 !atomic_inc_not_zero(&ct->ct_general.use)))
389 if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
400 struct nf_conntrack_tuple_hash *
401 nf_conntrack_find_get(struct net *net, u16 zone,
402 const struct nf_conntrack_tuple *tuple)
404 return __nf_conntrack_find_get(net, zone, tuple,
405 hash_conntrack_raw(tuple, zone));
407 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
409 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
411 unsigned int repl_hash)
413 struct net *net = nf_ct_net(ct);
415 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
416 &net->ct.hash[hash]);
417 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
418 &net->ct.hash[repl_hash]);
422 nf_conntrack_hash_check_insert(struct nf_conn *ct)
424 struct net *net = nf_ct_net(ct);
425 unsigned int hash, repl_hash;
426 struct nf_conntrack_tuple_hash *h;
427 struct hlist_nulls_node *n;
430 zone = nf_ct_zone(ct);
431 hash = hash_conntrack(net, zone,
432 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
433 repl_hash = hash_conntrack(net, zone,
434 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
436 spin_lock_bh(&nf_conntrack_lock);
438 /* See if there's one in the list already, including reverse */
439 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
440 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
442 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
444 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
445 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
447 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
450 add_timer(&ct->timeout);
451 nf_conntrack_get(&ct->ct_general);
452 __nf_conntrack_hash_insert(ct, hash, repl_hash);
453 NF_CT_STAT_INC(net, insert);
454 spin_unlock_bh(&nf_conntrack_lock);
459 NF_CT_STAT_INC(net, insert_failed);
460 spin_unlock_bh(&nf_conntrack_lock);
463 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
465 /* Confirm a connection given skb; places it in hash table */
467 __nf_conntrack_confirm(struct sk_buff *skb)
469 unsigned int hash, repl_hash;
470 struct nf_conntrack_tuple_hash *h;
472 struct nf_conn_help *help;
473 struct nf_conn_tstamp *tstamp;
474 struct hlist_nulls_node *n;
475 enum ip_conntrack_info ctinfo;
479 ct = nf_ct_get(skb, &ctinfo);
482 /* ipt_REJECT uses nf_conntrack_attach to attach related
483 ICMP/TCP RST packets in other direction. Actual packet
484 which created connection will be IP_CT_NEW or for an
485 expected connection, IP_CT_RELATED. */
486 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
489 zone = nf_ct_zone(ct);
490 /* reuse the hash saved before */
491 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
492 hash = hash_bucket(hash, net);
493 repl_hash = hash_conntrack(net, zone,
494 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
496 /* We're not in hash table, and we refuse to set up related
497 connections for unconfirmed conns. But packet copies and
498 REJECT will give spurious warnings here. */
499 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
501 /* No external references means no one else could have
503 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
504 pr_debug("Confirming conntrack %p\n", ct);
506 spin_lock_bh(&nf_conntrack_lock);
508 /* We have to check the DYING flag inside the lock to prevent
509 a race against nf_ct_get_next_corpse() possibly called from
510 user context, else we insert an already 'dead' hash, blocking
511 further use of that particular connection -JM */
513 if (unlikely(nf_ct_is_dying(ct))) {
514 spin_unlock_bh(&nf_conntrack_lock);
518 /* See if there's one in the list already, including reverse:
519 NAT could have grabbed it without realizing, since we're
520 not in the hash. If there is, we lost race. */
521 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
522 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
524 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
526 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
527 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
529 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
532 /* Remove from unconfirmed list */
533 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
535 /* Timer relative to confirmation time, not original
536 setting time, otherwise we'd get timer wrap in
537 weird delay cases. */
538 ct->timeout.expires += jiffies;
539 add_timer(&ct->timeout);
540 atomic_inc(&ct->ct_general.use);
541 ct->status |= IPS_CONFIRMED;
543 /* set conntrack timestamp, if enabled. */
544 tstamp = nf_conn_tstamp_find(ct);
546 if (skb->tstamp.tv64 == 0)
547 __net_timestamp(skb);
549 tstamp->start = ktime_to_ns(skb->tstamp);
551 /* Since the lookup is lockless, hash insertion must be done after
552 * starting the timer and setting the CONFIRMED bit. The RCU barriers
553 * guarantee that no other CPU can find the conntrack before the above
554 * stores are visible.
556 __nf_conntrack_hash_insert(ct, hash, repl_hash);
557 NF_CT_STAT_INC(net, insert);
558 spin_unlock_bh(&nf_conntrack_lock);
560 help = nfct_help(ct);
561 if (help && help->helper)
562 nf_conntrack_event_cache(IPCT_HELPER, ct);
564 nf_conntrack_event_cache(master_ct(ct) ?
565 IPCT_RELATED : IPCT_NEW, ct);
569 NF_CT_STAT_INC(net, insert_failed);
570 spin_unlock_bh(&nf_conntrack_lock);
573 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
575 /* Returns true if a connection correspondings to the tuple (required
578 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
579 const struct nf_conn *ignored_conntrack)
581 struct net *net = nf_ct_net(ignored_conntrack);
582 struct nf_conntrack_tuple_hash *h;
583 struct hlist_nulls_node *n;
585 u16 zone = nf_ct_zone(ignored_conntrack);
586 unsigned int hash = hash_conntrack(net, zone, tuple);
588 /* Disable BHs the entire time since we need to disable them at
589 * least once for the stats anyway.
592 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
593 ct = nf_ct_tuplehash_to_ctrack(h);
594 if (ct != ignored_conntrack &&
595 nf_ct_tuple_equal(tuple, &h->tuple) &&
596 nf_ct_zone(ct) == zone) {
597 NF_CT_STAT_INC(net, found);
598 rcu_read_unlock_bh();
601 NF_CT_STAT_INC(net, searched);
603 rcu_read_unlock_bh();
607 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
609 #define NF_CT_EVICTION_RANGE 8
611 /* There's a small race here where we may free a just-assured
612 connection. Too bad: we're in trouble anyway. */
613 static noinline int early_drop(struct net *net, unsigned int hash)
615 /* Use oldest entry, which is roughly LRU */
616 struct nf_conntrack_tuple_hash *h;
617 struct nf_conn *ct = NULL, *tmp;
618 struct hlist_nulls_node *n;
619 unsigned int i, cnt = 0;
623 for (i = 0; i < net->ct.htable_size; i++) {
624 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
626 tmp = nf_ct_tuplehash_to_ctrack(h);
627 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
633 if (likely(!nf_ct_is_dying(ct) &&
634 atomic_inc_not_zero(&ct->ct_general.use)))
640 if (cnt >= NF_CT_EVICTION_RANGE)
643 hash = (hash + 1) % net->ct.htable_size;
650 if (del_timer(&ct->timeout)) {
651 if (nf_ct_delete(ct, 0, 0)) {
653 NF_CT_STAT_INC_ATOMIC(net, early_drop);
660 void init_nf_conntrack_hash_rnd(void)
665 * Why not initialize nf_conntrack_rnd in a "init()" function ?
666 * Because there isn't enough entropy when system initializing,
667 * and we initialize it as late as possible.
670 get_random_bytes(&rand, sizeof(rand));
672 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
675 static struct nf_conn *
676 __nf_conntrack_alloc(struct net *net, u16 zone,
677 const struct nf_conntrack_tuple *orig,
678 const struct nf_conntrack_tuple *repl,
683 if (unlikely(!nf_conntrack_hash_rnd)) {
684 init_nf_conntrack_hash_rnd();
685 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
686 hash = hash_conntrack_raw(orig, zone);
689 /* We don't want any race condition at early drop stage */
690 atomic_inc(&net->ct.count);
692 if (nf_conntrack_max &&
693 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
694 if (!early_drop(net, hash_bucket(hash, net))) {
695 atomic_dec(&net->ct.count);
696 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
697 return ERR_PTR(-ENOMEM);
702 * Do not use kmem_cache_zalloc(), as this cache uses
703 * SLAB_DESTROY_BY_RCU.
705 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
707 atomic_dec(&net->ct.count);
708 return ERR_PTR(-ENOMEM);
711 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
712 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
714 memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
715 offsetof(struct nf_conn, proto) -
716 offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
717 spin_lock_init(&ct->lock);
718 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
719 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
720 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
721 /* save hash for reusing when confirming */
722 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
723 /* Don't set timer yet: wait for confirmation */
724 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
725 write_pnet(&ct->ct_net, net);
726 #ifdef CONFIG_NF_CONNTRACK_ZONES
728 struct nf_conntrack_zone *nf_ct_zone;
730 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
733 nf_ct_zone->id = zone;
737 * changes to lookup keys must be done before setting refcnt to 1
740 atomic_set(&ct->ct_general.use, 1);
743 #ifdef CONFIG_NF_CONNTRACK_ZONES
745 atomic_dec(&net->ct.count);
746 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
747 return ERR_PTR(-ENOMEM);
751 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
752 const struct nf_conntrack_tuple *orig,
753 const struct nf_conntrack_tuple *repl,
756 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
758 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
760 void nf_conntrack_free(struct nf_conn *ct)
762 struct net *net = nf_ct_net(ct);
764 nf_ct_ext_destroy(ct);
766 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
767 smp_mb__before_atomic_dec();
768 atomic_dec(&net->ct.count);
770 EXPORT_SYMBOL_GPL(nf_conntrack_free);
773 /* Allocate a new conntrack: we return -ENOMEM if classification
774 failed due to stress. Otherwise it really is unclassifiable. */
775 static struct nf_conntrack_tuple_hash *
776 init_conntrack(struct net *net, struct nf_conn *tmpl,
777 const struct nf_conntrack_tuple *tuple,
778 struct nf_conntrack_l3proto *l3proto,
779 struct nf_conntrack_l4proto *l4proto,
781 unsigned int dataoff, u32 hash)
784 struct nf_conn_help *help;
785 struct nf_conntrack_tuple repl_tuple;
786 struct nf_conntrack_ecache *ecache;
787 struct nf_conntrack_expect *exp;
788 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
789 struct nf_conn_timeout *timeout_ext;
790 unsigned int *timeouts;
792 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
793 pr_debug("Can't invert tuple.\n");
797 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
800 return (struct nf_conntrack_tuple_hash *)ct;
802 if (tmpl && nfct_synproxy(tmpl)) {
803 nfct_seqadj_ext_add(ct);
804 nfct_synproxy_ext_add(ct);
807 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
809 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
811 timeouts = l4proto->get_timeouts(net);
813 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
814 nf_conntrack_free(ct);
815 pr_debug("init conntrack: can't track with proto module\n");
820 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
822 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
823 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
824 nf_ct_labels_ext_add(ct);
826 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
827 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
828 ecache ? ecache->expmask : 0,
831 spin_lock_bh(&nf_conntrack_lock);
832 exp = nf_ct_find_expectation(net, zone, tuple);
834 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
836 /* Welcome, Mr. Bond. We've been expecting you... */
837 __set_bit(IPS_EXPECTED_BIT, &ct->status);
838 ct->master = exp->master;
840 help = nf_ct_helper_ext_add(ct, exp->helper,
843 rcu_assign_pointer(help->helper, exp->helper);
846 #ifdef CONFIG_NF_CONNTRACK_MARK
847 ct->mark = exp->master->mark;
849 #ifdef CONFIG_NF_CONNTRACK_SECMARK
850 ct->secmark = exp->master->secmark;
852 nf_conntrack_get(&ct->master->ct_general);
853 NF_CT_STAT_INC(net, expect_new);
855 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
856 NF_CT_STAT_INC(net, new);
859 /* Overload tuple linked list to put us in unconfirmed list. */
860 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
861 &net->ct.unconfirmed);
863 spin_unlock_bh(&nf_conntrack_lock);
867 exp->expectfn(ct, exp);
868 nf_ct_expect_put(exp);
871 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
874 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
875 static inline struct nf_conn *
876 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
878 unsigned int dataoff,
881 struct nf_conntrack_l3proto *l3proto,
882 struct nf_conntrack_l4proto *l4proto,
884 enum ip_conntrack_info *ctinfo)
886 struct nf_conntrack_tuple tuple;
887 struct nf_conntrack_tuple_hash *h;
889 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
892 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
893 dataoff, l3num, protonum, &tuple, l3proto,
895 pr_debug("resolve_normal_ct: Can't get tuple\n");
899 /* look for tuple match */
900 hash = hash_conntrack_raw(&tuple, zone);
901 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
903 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
910 ct = nf_ct_tuplehash_to_ctrack(h);
912 /* It exists; we have (non-exclusive) reference. */
913 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
914 *ctinfo = IP_CT_ESTABLISHED_REPLY;
915 /* Please set reply bit if this packet OK */
918 /* Once we've had two way comms, always ESTABLISHED. */
919 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
920 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
921 *ctinfo = IP_CT_ESTABLISHED;
922 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
923 pr_debug("nf_conntrack_in: related packet for %p\n",
925 *ctinfo = IP_CT_RELATED;
927 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
932 skb->nfct = &ct->ct_general;
933 skb->nfctinfo = *ctinfo;
938 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
941 struct nf_conn *ct, *tmpl = NULL;
942 enum ip_conntrack_info ctinfo;
943 struct nf_conntrack_l3proto *l3proto;
944 struct nf_conntrack_l4proto *l4proto;
945 unsigned int *timeouts;
946 unsigned int dataoff;
952 /* Previously seen (loopback or untracked)? Ignore. */
953 tmpl = (struct nf_conn *)skb->nfct;
954 if (!nf_ct_is_template(tmpl)) {
955 NF_CT_STAT_INC_ATOMIC(net, ignore);
961 /* rcu_read_lock()ed by nf_hook_slow */
962 l3proto = __nf_ct_l3proto_find(pf);
963 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
964 &dataoff, &protonum);
966 pr_debug("not prepared to track yet or error occurred\n");
967 NF_CT_STAT_INC_ATOMIC(net, error);
968 NF_CT_STAT_INC_ATOMIC(net, invalid);
973 l4proto = __nf_ct_l4proto_find(pf, protonum);
975 /* It may be an special packet, error, unclean...
976 * inverse of the return code tells to the netfilter
977 * core what to do with the packet. */
978 if (l4proto->error != NULL) {
979 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
982 NF_CT_STAT_INC_ATOMIC(net, error);
983 NF_CT_STAT_INC_ATOMIC(net, invalid);
987 /* ICMP[v6] protocol trackers may assign one conntrack. */
992 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
993 l3proto, l4proto, &set_reply, &ctinfo);
995 /* Not valid part of a connection */
996 NF_CT_STAT_INC_ATOMIC(net, invalid);
1002 /* Too stressed to deal. */
1003 NF_CT_STAT_INC_ATOMIC(net, drop);
1008 NF_CT_ASSERT(skb->nfct);
1010 /* Decide what timeout policy we want to apply to this flow. */
1011 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1013 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1015 /* Invalid: inverse of the return code tells
1016 * the netfilter core what to do */
1017 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1018 nf_conntrack_put(skb->nfct);
1020 NF_CT_STAT_INC_ATOMIC(net, invalid);
1021 if (ret == -NF_DROP)
1022 NF_CT_STAT_INC_ATOMIC(net, drop);
1027 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1028 nf_conntrack_event_cache(IPCT_REPLY, ct);
1031 /* Special case: we have to repeat this hook, assign the
1032 * template again to this packet. We assume that this packet
1033 * has no conntrack assigned. This is used by nf_ct_tcp. */
1034 if (ret == NF_REPEAT)
1035 skb->nfct = (struct nf_conntrack *)tmpl;
1042 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1044 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1045 const struct nf_conntrack_tuple *orig)
1050 ret = nf_ct_invert_tuple(inverse, orig,
1051 __nf_ct_l3proto_find(orig->src.l3num),
1052 __nf_ct_l4proto_find(orig->src.l3num,
1053 orig->dst.protonum));
1057 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1059 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1060 implicitly racy: see __nf_conntrack_confirm */
1061 void nf_conntrack_alter_reply(struct nf_conn *ct,
1062 const struct nf_conntrack_tuple *newreply)
1064 struct nf_conn_help *help = nfct_help(ct);
1066 /* Should be unconfirmed, so not in hash table yet */
1067 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1069 pr_debug("Altering reply tuple of %p to ", ct);
1070 nf_ct_dump_tuple(newreply);
1072 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1073 if (ct->master || (help && !hlist_empty(&help->expectations)))
1077 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1080 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1082 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1083 void __nf_ct_refresh_acct(struct nf_conn *ct,
1084 enum ip_conntrack_info ctinfo,
1085 const struct sk_buff *skb,
1086 unsigned long extra_jiffies,
1089 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1092 /* Only update if this is not a fixed timeout */
1093 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1096 /* If not in hash table, timer will not be active yet */
1097 if (!nf_ct_is_confirmed(ct)) {
1098 ct->timeout.expires = extra_jiffies;
1100 unsigned long newtime = jiffies + extra_jiffies;
1102 /* Only update the timeout if the new timeout is at least
1103 HZ jiffies from the old timeout. Need del_timer for race
1104 avoidance (may already be dying). */
1105 if (newtime - ct->timeout.expires >= HZ)
1106 mod_timer_pending(&ct->timeout, newtime);
1111 struct nf_conn_acct *acct;
1113 acct = nf_conn_acct_find(ct);
1115 struct nf_conn_counter *counter = acct->counter;
1117 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1118 atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1122 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1124 bool __nf_ct_kill_acct(struct nf_conn *ct,
1125 enum ip_conntrack_info ctinfo,
1126 const struct sk_buff *skb,
1130 struct nf_conn_acct *acct;
1132 acct = nf_conn_acct_find(ct);
1134 struct nf_conn_counter *counter = acct->counter;
1136 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1137 atomic64_add(skb->len - skb_network_offset(skb),
1138 &counter[CTINFO2DIR(ctinfo)].bytes);
1142 if (del_timer(&ct->timeout)) {
1143 ct->timeout.function((unsigned long)ct);
1148 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1150 #ifdef CONFIG_NF_CONNTRACK_ZONES
1151 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1152 .len = sizeof(struct nf_conntrack_zone),
1153 .align = __alignof__(struct nf_conntrack_zone),
1154 .id = NF_CT_EXT_ZONE,
1158 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1160 #include <linux/netfilter/nfnetlink.h>
1161 #include <linux/netfilter/nfnetlink_conntrack.h>
1162 #include <linux/mutex.h>
1164 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1165 * in ip_conntrack_core, since we don't want the protocols to autoload
1166 * or depend on ctnetlink */
1167 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1168 const struct nf_conntrack_tuple *tuple)
1170 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1171 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1172 goto nla_put_failure;
1178 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1180 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1181 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1182 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1184 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1186 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1187 struct nf_conntrack_tuple *t)
1189 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1192 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1193 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1197 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1199 int nf_ct_port_nlattr_tuple_size(void)
1201 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1203 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1206 /* Used by ipt_REJECT and ip6t_REJECT. */
1207 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1210 enum ip_conntrack_info ctinfo;
1212 /* This ICMP is in reverse direction to the packet which caused it */
1213 ct = nf_ct_get(skb, &ctinfo);
1214 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1215 ctinfo = IP_CT_RELATED_REPLY;
1217 ctinfo = IP_CT_RELATED;
1219 /* Attach to new skbuff, and increment count */
1220 nskb->nfct = &ct->ct_general;
1221 nskb->nfctinfo = ctinfo;
1222 nf_conntrack_get(nskb->nfct);
1225 /* Bring out ya dead! */
1226 static struct nf_conn *
1227 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1228 void *data, unsigned int *bucket)
1230 struct nf_conntrack_tuple_hash *h;
1232 struct hlist_nulls_node *n;
1234 spin_lock_bh(&nf_conntrack_lock);
1235 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1236 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1237 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1239 ct = nf_ct_tuplehash_to_ctrack(h);
1244 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1245 ct = nf_ct_tuplehash_to_ctrack(h);
1247 set_bit(IPS_DYING_BIT, &ct->status);
1249 spin_unlock_bh(&nf_conntrack_lock);
1252 atomic_inc(&ct->ct_general.use);
1253 spin_unlock_bh(&nf_conntrack_lock);
1257 void nf_ct_iterate_cleanup(struct net *net,
1258 int (*iter)(struct nf_conn *i, void *data),
1259 void *data, u32 portid, int report)
1262 unsigned int bucket = 0;
1264 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1265 /* Time to push up daises... */
1266 if (del_timer(&ct->timeout))
1267 nf_ct_delete(ct, portid, report);
1269 /* ... else the timer will get him soon. */
1274 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1276 static int kill_all(struct nf_conn *i, void *data)
1281 void nf_ct_free_hashtable(void *hash, unsigned int size)
1283 if (is_vmalloc_addr(hash))
1286 free_pages((unsigned long)hash,
1287 get_order(sizeof(struct hlist_head) * size));
1289 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1291 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1293 nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1295 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1297 static void nf_ct_release_dying_list(struct net *net)
1299 struct nf_conntrack_tuple_hash *h;
1301 struct hlist_nulls_node *n;
1303 spin_lock_bh(&nf_conntrack_lock);
1304 hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1305 ct = nf_ct_tuplehash_to_ctrack(h);
1306 /* never fails to remove them, no listeners at this point */
1309 spin_unlock_bh(&nf_conntrack_lock);
1312 static int untrack_refs(void)
1316 for_each_possible_cpu(cpu) {
1317 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1319 cnt += atomic_read(&ct->ct_general.use) - 1;
1324 void nf_conntrack_cleanup_start(void)
1326 RCU_INIT_POINTER(ip_ct_attach, NULL);
1329 void nf_conntrack_cleanup_end(void)
1331 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1332 while (untrack_refs() > 0)
1335 #ifdef CONFIG_NF_CONNTRACK_ZONES
1336 nf_ct_extend_unregister(&nf_ct_zone_extend);
1338 nf_conntrack_proto_fini();
1339 nf_conntrack_seqadj_fini();
1340 nf_conntrack_labels_fini();
1341 nf_conntrack_helper_fini();
1342 nf_conntrack_timeout_fini();
1343 nf_conntrack_ecache_fini();
1344 nf_conntrack_tstamp_fini();
1345 nf_conntrack_acct_fini();
1346 nf_conntrack_expect_fini();
1350 * Mishearing the voices in his head, our hero wonders how he's
1351 * supposed to kill the mall.
1353 void nf_conntrack_cleanup_net(struct net *net)
1357 list_add(&net->exit_list, &single);
1358 nf_conntrack_cleanup_net_list(&single);
1361 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1367 * This makes sure all current packets have passed through
1368 * netfilter framework. Roll on, two-stage module
1374 list_for_each_entry(net, net_exit_list, exit_list) {
1375 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1376 nf_ct_release_dying_list(net);
1377 if (atomic_read(&net->ct.count) != 0)
1382 goto i_see_dead_people;
1385 list_for_each_entry(net, net_exit_list, exit_list) {
1386 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1387 nf_conntrack_proto_pernet_fini(net);
1388 nf_conntrack_helper_pernet_fini(net);
1389 nf_conntrack_ecache_pernet_fini(net);
1390 nf_conntrack_tstamp_pernet_fini(net);
1391 nf_conntrack_acct_pernet_fini(net);
1392 nf_conntrack_expect_pernet_fini(net);
1393 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1394 kfree(net->ct.slabname);
1395 free_percpu(net->ct.stat);
1399 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1401 struct hlist_nulls_head *hash;
1402 unsigned int nr_slots, i;
1405 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1406 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1407 sz = nr_slots * sizeof(struct hlist_nulls_head);
1408 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1411 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1416 for (i = 0; i < nr_slots; i++)
1417 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1421 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1423 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1426 unsigned int hashsize, old_size;
1427 struct hlist_nulls_head *hash, *old_hash;
1428 struct nf_conntrack_tuple_hash *h;
1431 if (current->nsproxy->net_ns != &init_net)
1434 /* On boot, we can set this without any fancy locking. */
1435 if (!nf_conntrack_htable_size)
1436 return param_set_uint(val, kp);
1438 rc = kstrtouint(val, 0, &hashsize);
1444 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1448 /* Lookups in the old hash might happen in parallel, which means we
1449 * might get false negatives during connection lookup. New connections
1450 * created because of a false negative won't make it into the hash
1451 * though since that required taking the lock.
1453 spin_lock_bh(&nf_conntrack_lock);
1454 for (i = 0; i < init_net.ct.htable_size; i++) {
1455 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1456 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1457 struct nf_conntrack_tuple_hash, hnnode);
1458 ct = nf_ct_tuplehash_to_ctrack(h);
1459 hlist_nulls_del_rcu(&h->hnnode);
1460 bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1462 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1465 old_size = init_net.ct.htable_size;
1466 old_hash = init_net.ct.hash;
1468 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1469 init_net.ct.hash = hash;
1470 spin_unlock_bh(&nf_conntrack_lock);
1472 nf_ct_free_hashtable(old_hash, old_size);
1475 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1477 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1478 &nf_conntrack_htable_size, 0600);
1480 void nf_ct_untracked_status_or(unsigned long bits)
1484 for_each_possible_cpu(cpu)
1485 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1487 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1489 int nf_conntrack_init_start(void)
1494 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1495 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1496 if (!nf_conntrack_htable_size) {
1497 nf_conntrack_htable_size
1498 = (((totalram_pages << PAGE_SHIFT) / 16384)
1499 / sizeof(struct hlist_head));
1500 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1501 nf_conntrack_htable_size = 16384;
1502 if (nf_conntrack_htable_size < 32)
1503 nf_conntrack_htable_size = 32;
1505 /* Use a max. factor of four by default to get the same max as
1506 * with the old struct list_heads. When a table size is given
1507 * we use the old value of 8 to avoid reducing the max.
1511 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1513 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1514 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1517 ret = nf_conntrack_expect_init();
1521 ret = nf_conntrack_acct_init();
1525 ret = nf_conntrack_tstamp_init();
1529 ret = nf_conntrack_ecache_init();
1533 ret = nf_conntrack_timeout_init();
1537 ret = nf_conntrack_helper_init();
1541 ret = nf_conntrack_labels_init();
1545 ret = nf_conntrack_seqadj_init();
1549 #ifdef CONFIG_NF_CONNTRACK_ZONES
1550 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1554 ret = nf_conntrack_proto_init();
1558 /* Set up fake conntrack: to never be deleted, not in any hashes */
1559 for_each_possible_cpu(cpu) {
1560 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1561 write_pnet(&ct->ct_net, &init_net);
1562 atomic_set(&ct->ct_general.use, 1);
1564 /* - and look it like as a confirmed connection */
1565 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1569 #ifdef CONFIG_NF_CONNTRACK_ZONES
1570 nf_ct_extend_unregister(&nf_ct_zone_extend);
1573 nf_conntrack_seqadj_fini();
1575 nf_conntrack_labels_fini();
1577 nf_conntrack_helper_fini();
1579 nf_conntrack_timeout_fini();
1581 nf_conntrack_ecache_fini();
1583 nf_conntrack_tstamp_fini();
1585 nf_conntrack_acct_fini();
1587 nf_conntrack_expect_fini();
1592 void nf_conntrack_init_end(void)
1594 /* For use by REJECT target */
1595 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1596 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1600 * We need to use special "null" values, not used in hash table
1602 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1603 #define DYING_NULLS_VAL ((1<<30)+1)
1604 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1606 int nf_conntrack_init_net(struct net *net)
1610 atomic_set(&net->ct.count, 0);
1611 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1612 INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1613 INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1614 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1615 if (!net->ct.stat) {
1620 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1621 if (!net->ct.slabname) {
1626 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1627 sizeof(struct nf_conn), 0,
1628 SLAB_DESTROY_BY_RCU, NULL);
1629 if (!net->ct.nf_conntrack_cachep) {
1630 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1635 net->ct.htable_size = nf_conntrack_htable_size;
1636 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1637 if (!net->ct.hash) {
1639 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1642 ret = nf_conntrack_expect_pernet_init(net);
1645 ret = nf_conntrack_acct_pernet_init(net);
1648 ret = nf_conntrack_tstamp_pernet_init(net);
1651 ret = nf_conntrack_ecache_pernet_init(net);
1654 ret = nf_conntrack_helper_pernet_init(net);
1657 ret = nf_conntrack_proto_pernet_init(net);
1663 nf_conntrack_helper_pernet_fini(net);
1665 nf_conntrack_ecache_pernet_fini(net);
1667 nf_conntrack_tstamp_pernet_fini(net);
1669 nf_conntrack_acct_pernet_fini(net);
1671 nf_conntrack_expect_pernet_fini(net);
1673 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1675 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1677 kfree(net->ct.slabname);
1679 free_percpu(net->ct.stat);