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);
452 /* The caller holds a reference to this object */
453 atomic_set(&ct->ct_general.use, 2);
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 /* deletion from this larval template list happens via nf_ct_put() */
468 void nf_conntrack_tmpl_insert(struct net *net, struct nf_conn *tmpl)
470 __set_bit(IPS_TEMPLATE_BIT, &tmpl->status);
471 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
472 nf_conntrack_get(&tmpl->ct_general);
474 spin_lock_bh(&nf_conntrack_lock);
475 /* Overload tuple linked list to put us in template list. */
476 hlist_nulls_add_head_rcu(&tmpl->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
478 spin_unlock_bh(&nf_conntrack_lock);
480 EXPORT_SYMBOL_GPL(nf_conntrack_tmpl_insert);
482 /* Confirm a connection given skb; places it in hash table */
484 __nf_conntrack_confirm(struct sk_buff *skb)
486 unsigned int hash, repl_hash;
487 struct nf_conntrack_tuple_hash *h;
489 struct nf_conn_help *help;
490 struct nf_conn_tstamp *tstamp;
491 struct hlist_nulls_node *n;
492 enum ip_conntrack_info ctinfo;
496 ct = nf_ct_get(skb, &ctinfo);
499 /* ipt_REJECT uses nf_conntrack_attach to attach related
500 ICMP/TCP RST packets in other direction. Actual packet
501 which created connection will be IP_CT_NEW or for an
502 expected connection, IP_CT_RELATED. */
503 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
506 zone = nf_ct_zone(ct);
507 /* reuse the hash saved before */
508 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
509 hash = hash_bucket(hash, net);
510 repl_hash = hash_conntrack(net, zone,
511 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
513 /* We're not in hash table, and we refuse to set up related
514 connections for unconfirmed conns. But packet copies and
515 REJECT will give spurious warnings here. */
516 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
518 /* No external references means no one else could have
520 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
521 pr_debug("Confirming conntrack %p\n", ct);
523 spin_lock_bh(&nf_conntrack_lock);
525 /* We have to check the DYING flag inside the lock to prevent
526 a race against nf_ct_get_next_corpse() possibly called from
527 user context, else we insert an already 'dead' hash, blocking
528 further use of that particular connection -JM */
530 if (unlikely(nf_ct_is_dying(ct))) {
531 spin_unlock_bh(&nf_conntrack_lock);
535 /* See if there's one in the list already, including reverse:
536 NAT could have grabbed it without realizing, since we're
537 not in the hash. If there is, we lost race. */
538 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
539 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
541 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
543 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
544 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
546 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
549 /* Remove from unconfirmed list */
550 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
552 /* Timer relative to confirmation time, not original
553 setting time, otherwise we'd get timer wrap in
554 weird delay cases. */
555 ct->timeout.expires += jiffies;
556 add_timer(&ct->timeout);
557 atomic_inc(&ct->ct_general.use);
558 ct->status |= IPS_CONFIRMED;
560 /* set conntrack timestamp, if enabled. */
561 tstamp = nf_conn_tstamp_find(ct);
563 if (skb->tstamp.tv64 == 0)
564 __net_timestamp(skb);
566 tstamp->start = ktime_to_ns(skb->tstamp);
568 /* Since the lookup is lockless, hash insertion must be done after
569 * starting the timer and setting the CONFIRMED bit. The RCU barriers
570 * guarantee that no other CPU can find the conntrack before the above
571 * stores are visible.
573 __nf_conntrack_hash_insert(ct, hash, repl_hash);
574 NF_CT_STAT_INC(net, insert);
575 spin_unlock_bh(&nf_conntrack_lock);
577 help = nfct_help(ct);
578 if (help && help->helper)
579 nf_conntrack_event_cache(IPCT_HELPER, ct);
581 nf_conntrack_event_cache(master_ct(ct) ?
582 IPCT_RELATED : IPCT_NEW, ct);
586 NF_CT_STAT_INC(net, insert_failed);
587 spin_unlock_bh(&nf_conntrack_lock);
590 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
592 /* Returns true if a connection correspondings to the tuple (required
595 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
596 const struct nf_conn *ignored_conntrack)
598 struct net *net = nf_ct_net(ignored_conntrack);
599 struct nf_conntrack_tuple_hash *h;
600 struct hlist_nulls_node *n;
602 u16 zone = nf_ct_zone(ignored_conntrack);
603 unsigned int hash = hash_conntrack(net, zone, tuple);
605 /* Disable BHs the entire time since we need to disable them at
606 * least once for the stats anyway.
609 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
610 ct = nf_ct_tuplehash_to_ctrack(h);
611 if (ct != ignored_conntrack &&
612 nf_ct_tuple_equal(tuple, &h->tuple) &&
613 nf_ct_zone(ct) == zone) {
614 NF_CT_STAT_INC(net, found);
615 rcu_read_unlock_bh();
618 NF_CT_STAT_INC(net, searched);
620 rcu_read_unlock_bh();
624 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
626 #define NF_CT_EVICTION_RANGE 8
628 /* There's a small race here where we may free a just-assured
629 connection. Too bad: we're in trouble anyway. */
630 static noinline int early_drop(struct net *net, unsigned int hash)
632 /* Use oldest entry, which is roughly LRU */
633 struct nf_conntrack_tuple_hash *h;
634 struct nf_conn *ct = NULL, *tmp;
635 struct hlist_nulls_node *n;
636 unsigned int i, cnt = 0;
640 for (i = 0; i < net->ct.htable_size; i++) {
641 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
643 tmp = nf_ct_tuplehash_to_ctrack(h);
644 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
650 if (likely(!nf_ct_is_dying(ct) &&
651 atomic_inc_not_zero(&ct->ct_general.use)))
657 if (cnt >= NF_CT_EVICTION_RANGE)
660 hash = (hash + 1) % net->ct.htable_size;
667 if (del_timer(&ct->timeout)) {
668 if (nf_ct_delete(ct, 0, 0)) {
670 NF_CT_STAT_INC_ATOMIC(net, early_drop);
677 void init_nf_conntrack_hash_rnd(void)
682 * Why not initialize nf_conntrack_rnd in a "init()" function ?
683 * Because there isn't enough entropy when system initializing,
684 * and we initialize it as late as possible.
687 get_random_bytes(&rand, sizeof(rand));
689 cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
692 static struct nf_conn *
693 __nf_conntrack_alloc(struct net *net, u16 zone,
694 const struct nf_conntrack_tuple *orig,
695 const struct nf_conntrack_tuple *repl,
700 if (unlikely(!nf_conntrack_hash_rnd)) {
701 init_nf_conntrack_hash_rnd();
702 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
703 hash = hash_conntrack_raw(orig, zone);
706 /* We don't want any race condition at early drop stage */
707 atomic_inc(&net->ct.count);
709 if (nf_conntrack_max &&
710 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
711 if (!early_drop(net, hash_bucket(hash, net))) {
712 atomic_dec(&net->ct.count);
713 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
714 return ERR_PTR(-ENOMEM);
719 * Do not use kmem_cache_zalloc(), as this cache uses
720 * SLAB_DESTROY_BY_RCU.
722 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
724 atomic_dec(&net->ct.count);
725 return ERR_PTR(-ENOMEM);
728 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
729 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
731 memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
732 offsetof(struct nf_conn, proto) -
733 offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
734 spin_lock_init(&ct->lock);
735 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
736 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
737 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
738 /* save hash for reusing when confirming */
739 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
740 /* Don't set timer yet: wait for confirmation */
741 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
742 write_pnet(&ct->ct_net, net);
743 #ifdef CONFIG_NF_CONNTRACK_ZONES
745 struct nf_conntrack_zone *nf_ct_zone;
747 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
750 nf_ct_zone->id = zone;
753 /* Because we use RCU lookups, we set ct_general.use to zero before
754 * this is inserted in any list.
756 atomic_set(&ct->ct_general.use, 0);
759 #ifdef CONFIG_NF_CONNTRACK_ZONES
761 atomic_dec(&net->ct.count);
762 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
763 return ERR_PTR(-ENOMEM);
767 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
768 const struct nf_conntrack_tuple *orig,
769 const struct nf_conntrack_tuple *repl,
772 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
774 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
776 void nf_conntrack_free(struct nf_conn *ct)
778 struct net *net = nf_ct_net(ct);
780 /* A freed object has refcnt == 0, that's
781 * the golden rule for SLAB_DESTROY_BY_RCU
783 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
785 nf_ct_ext_destroy(ct);
787 kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
788 smp_mb__before_atomic_dec();
789 atomic_dec(&net->ct.count);
791 EXPORT_SYMBOL_GPL(nf_conntrack_free);
794 /* Allocate a new conntrack: we return -ENOMEM if classification
795 failed due to stress. Otherwise it really is unclassifiable. */
796 static struct nf_conntrack_tuple_hash *
797 init_conntrack(struct net *net, struct nf_conn *tmpl,
798 const struct nf_conntrack_tuple *tuple,
799 struct nf_conntrack_l3proto *l3proto,
800 struct nf_conntrack_l4proto *l4proto,
802 unsigned int dataoff, u32 hash)
805 struct nf_conn_help *help;
806 struct nf_conntrack_tuple repl_tuple;
807 struct nf_conntrack_ecache *ecache;
808 struct nf_conntrack_expect *exp;
809 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
810 struct nf_conn_timeout *timeout_ext;
811 unsigned int *timeouts;
813 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
814 pr_debug("Can't invert tuple.\n");
818 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
821 return (struct nf_conntrack_tuple_hash *)ct;
823 if (tmpl && nfct_synproxy(tmpl)) {
824 nfct_seqadj_ext_add(ct);
825 nfct_synproxy_ext_add(ct);
828 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
830 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
832 timeouts = l4proto->get_timeouts(net);
834 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
835 nf_conntrack_free(ct);
836 pr_debug("init conntrack: can't track with proto module\n");
841 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
843 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
844 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
845 nf_ct_labels_ext_add(ct);
847 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
848 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
849 ecache ? ecache->expmask : 0,
852 spin_lock_bh(&nf_conntrack_lock);
853 exp = nf_ct_find_expectation(net, zone, tuple);
855 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
857 /* Welcome, Mr. Bond. We've been expecting you... */
858 __set_bit(IPS_EXPECTED_BIT, &ct->status);
859 ct->master = exp->master;
861 help = nf_ct_helper_ext_add(ct, exp->helper,
864 rcu_assign_pointer(help->helper, exp->helper);
867 #ifdef CONFIG_NF_CONNTRACK_MARK
868 ct->mark = exp->master->mark;
870 #ifdef CONFIG_NF_CONNTRACK_SECMARK
871 ct->secmark = exp->master->secmark;
873 nf_conntrack_get(&ct->master->ct_general);
874 NF_CT_STAT_INC(net, expect_new);
876 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
877 NF_CT_STAT_INC(net, new);
880 /* Now it is inserted into the unconfirmed list, bump refcount */
881 nf_conntrack_get(&ct->ct_general);
883 /* Overload tuple linked list to put us in unconfirmed list. */
884 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
885 &net->ct.unconfirmed);
887 spin_unlock_bh(&nf_conntrack_lock);
891 exp->expectfn(ct, exp);
892 nf_ct_expect_put(exp);
895 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
898 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
899 static inline struct nf_conn *
900 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
902 unsigned int dataoff,
905 struct nf_conntrack_l3proto *l3proto,
906 struct nf_conntrack_l4proto *l4proto,
908 enum ip_conntrack_info *ctinfo)
910 struct nf_conntrack_tuple tuple;
911 struct nf_conntrack_tuple_hash *h;
913 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
916 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
917 dataoff, l3num, protonum, &tuple, l3proto,
919 pr_debug("resolve_normal_ct: Can't get tuple\n");
923 /* look for tuple match */
924 hash = hash_conntrack_raw(&tuple, zone);
925 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
927 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
934 ct = nf_ct_tuplehash_to_ctrack(h);
936 /* It exists; we have (non-exclusive) reference. */
937 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
938 *ctinfo = IP_CT_ESTABLISHED_REPLY;
939 /* Please set reply bit if this packet OK */
942 /* Once we've had two way comms, always ESTABLISHED. */
943 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
944 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
945 *ctinfo = IP_CT_ESTABLISHED;
946 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
947 pr_debug("nf_conntrack_in: related packet for %p\n",
949 *ctinfo = IP_CT_RELATED;
951 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
956 skb->nfct = &ct->ct_general;
957 skb->nfctinfo = *ctinfo;
962 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
965 struct nf_conn *ct, *tmpl = NULL;
966 enum ip_conntrack_info ctinfo;
967 struct nf_conntrack_l3proto *l3proto;
968 struct nf_conntrack_l4proto *l4proto;
969 unsigned int *timeouts;
970 unsigned int dataoff;
976 /* Previously seen (loopback or untracked)? Ignore. */
977 tmpl = (struct nf_conn *)skb->nfct;
978 if (!nf_ct_is_template(tmpl)) {
979 NF_CT_STAT_INC_ATOMIC(net, ignore);
985 /* rcu_read_lock()ed by nf_hook_slow */
986 l3proto = __nf_ct_l3proto_find(pf);
987 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
988 &dataoff, &protonum);
990 pr_debug("not prepared to track yet or error occurred\n");
991 NF_CT_STAT_INC_ATOMIC(net, error);
992 NF_CT_STAT_INC_ATOMIC(net, invalid);
997 l4proto = __nf_ct_l4proto_find(pf, protonum);
999 /* It may be an special packet, error, unclean...
1000 * inverse of the return code tells to the netfilter
1001 * core what to do with the packet. */
1002 if (l4proto->error != NULL) {
1003 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1006 NF_CT_STAT_INC_ATOMIC(net, error);
1007 NF_CT_STAT_INC_ATOMIC(net, invalid);
1011 /* ICMP[v6] protocol trackers may assign one conntrack. */
1016 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1017 l3proto, l4proto, &set_reply, &ctinfo);
1019 /* Not valid part of a connection */
1020 NF_CT_STAT_INC_ATOMIC(net, invalid);
1026 /* Too stressed to deal. */
1027 NF_CT_STAT_INC_ATOMIC(net, drop);
1032 NF_CT_ASSERT(skb->nfct);
1034 /* Decide what timeout policy we want to apply to this flow. */
1035 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1037 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1039 /* Invalid: inverse of the return code tells
1040 * the netfilter core what to do */
1041 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1042 nf_conntrack_put(skb->nfct);
1044 NF_CT_STAT_INC_ATOMIC(net, invalid);
1045 if (ret == -NF_DROP)
1046 NF_CT_STAT_INC_ATOMIC(net, drop);
1051 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1052 nf_conntrack_event_cache(IPCT_REPLY, ct);
1055 /* Special case: we have to repeat this hook, assign the
1056 * template again to this packet. We assume that this packet
1057 * has no conntrack assigned. This is used by nf_ct_tcp. */
1058 if (ret == NF_REPEAT)
1059 skb->nfct = (struct nf_conntrack *)tmpl;
1066 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1068 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1069 const struct nf_conntrack_tuple *orig)
1074 ret = nf_ct_invert_tuple(inverse, orig,
1075 __nf_ct_l3proto_find(orig->src.l3num),
1076 __nf_ct_l4proto_find(orig->src.l3num,
1077 orig->dst.protonum));
1081 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1083 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1084 implicitly racy: see __nf_conntrack_confirm */
1085 void nf_conntrack_alter_reply(struct nf_conn *ct,
1086 const struct nf_conntrack_tuple *newreply)
1088 struct nf_conn_help *help = nfct_help(ct);
1090 /* Should be unconfirmed, so not in hash table yet */
1091 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1093 pr_debug("Altering reply tuple of %p to ", ct);
1094 nf_ct_dump_tuple(newreply);
1096 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1097 if (ct->master || (help && !hlist_empty(&help->expectations)))
1101 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1104 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1106 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1107 void __nf_ct_refresh_acct(struct nf_conn *ct,
1108 enum ip_conntrack_info ctinfo,
1109 const struct sk_buff *skb,
1110 unsigned long extra_jiffies,
1113 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1116 /* Only update if this is not a fixed timeout */
1117 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1120 /* If not in hash table, timer will not be active yet */
1121 if (!nf_ct_is_confirmed(ct)) {
1122 ct->timeout.expires = extra_jiffies;
1124 unsigned long newtime = jiffies + extra_jiffies;
1126 /* Only update the timeout if the new timeout is at least
1127 HZ jiffies from the old timeout. Need del_timer for race
1128 avoidance (may already be dying). */
1129 if (newtime - ct->timeout.expires >= HZ)
1130 mod_timer_pending(&ct->timeout, newtime);
1135 struct nf_conn_acct *acct;
1137 acct = nf_conn_acct_find(ct);
1139 struct nf_conn_counter *counter = acct->counter;
1141 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1142 atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1146 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1148 bool __nf_ct_kill_acct(struct nf_conn *ct,
1149 enum ip_conntrack_info ctinfo,
1150 const struct sk_buff *skb,
1154 struct nf_conn_acct *acct;
1156 acct = nf_conn_acct_find(ct);
1158 struct nf_conn_counter *counter = acct->counter;
1160 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1161 atomic64_add(skb->len - skb_network_offset(skb),
1162 &counter[CTINFO2DIR(ctinfo)].bytes);
1166 if (del_timer(&ct->timeout)) {
1167 ct->timeout.function((unsigned long)ct);
1172 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1174 #ifdef CONFIG_NF_CONNTRACK_ZONES
1175 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1176 .len = sizeof(struct nf_conntrack_zone),
1177 .align = __alignof__(struct nf_conntrack_zone),
1178 .id = NF_CT_EXT_ZONE,
1182 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1184 #include <linux/netfilter/nfnetlink.h>
1185 #include <linux/netfilter/nfnetlink_conntrack.h>
1186 #include <linux/mutex.h>
1188 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1189 * in ip_conntrack_core, since we don't want the protocols to autoload
1190 * or depend on ctnetlink */
1191 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1192 const struct nf_conntrack_tuple *tuple)
1194 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1195 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1196 goto nla_put_failure;
1202 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1204 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1205 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1206 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1208 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1210 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1211 struct nf_conntrack_tuple *t)
1213 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1216 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1217 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1221 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1223 int nf_ct_port_nlattr_tuple_size(void)
1225 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1227 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1230 /* Used by ipt_REJECT and ip6t_REJECT. */
1231 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1234 enum ip_conntrack_info ctinfo;
1236 /* This ICMP is in reverse direction to the packet which caused it */
1237 ct = nf_ct_get(skb, &ctinfo);
1238 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1239 ctinfo = IP_CT_RELATED_REPLY;
1241 ctinfo = IP_CT_RELATED;
1243 /* Attach to new skbuff, and increment count */
1244 nskb->nfct = &ct->ct_general;
1245 nskb->nfctinfo = ctinfo;
1246 nf_conntrack_get(nskb->nfct);
1249 /* Bring out ya dead! */
1250 static struct nf_conn *
1251 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1252 void *data, unsigned int *bucket)
1254 struct nf_conntrack_tuple_hash *h;
1256 struct hlist_nulls_node *n;
1258 spin_lock_bh(&nf_conntrack_lock);
1259 for (; *bucket < net->ct.htable_size; (*bucket)++) {
1260 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1261 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1263 ct = nf_ct_tuplehash_to_ctrack(h);
1268 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1269 ct = nf_ct_tuplehash_to_ctrack(h);
1271 set_bit(IPS_DYING_BIT, &ct->status);
1273 spin_unlock_bh(&nf_conntrack_lock);
1276 atomic_inc(&ct->ct_general.use);
1277 spin_unlock_bh(&nf_conntrack_lock);
1281 void nf_ct_iterate_cleanup(struct net *net,
1282 int (*iter)(struct nf_conn *i, void *data),
1283 void *data, u32 portid, int report)
1286 unsigned int bucket = 0;
1288 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1289 /* Time to push up daises... */
1290 if (del_timer(&ct->timeout))
1291 nf_ct_delete(ct, portid, report);
1293 /* ... else the timer will get him soon. */
1298 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1300 static int kill_all(struct nf_conn *i, void *data)
1305 void nf_ct_free_hashtable(void *hash, unsigned int size)
1307 if (is_vmalloc_addr(hash))
1310 free_pages((unsigned long)hash,
1311 get_order(sizeof(struct hlist_head) * size));
1313 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1315 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1317 nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1319 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1321 static void nf_ct_release_dying_list(struct net *net)
1323 struct nf_conntrack_tuple_hash *h;
1325 struct hlist_nulls_node *n;
1327 spin_lock_bh(&nf_conntrack_lock);
1328 hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1329 ct = nf_ct_tuplehash_to_ctrack(h);
1330 /* never fails to remove them, no listeners at this point */
1333 spin_unlock_bh(&nf_conntrack_lock);
1336 static int untrack_refs(void)
1340 for_each_possible_cpu(cpu) {
1341 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1343 cnt += atomic_read(&ct->ct_general.use) - 1;
1348 void nf_conntrack_cleanup_start(void)
1350 RCU_INIT_POINTER(ip_ct_attach, NULL);
1353 void nf_conntrack_cleanup_end(void)
1355 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1356 while (untrack_refs() > 0)
1359 #ifdef CONFIG_NF_CONNTRACK_ZONES
1360 nf_ct_extend_unregister(&nf_ct_zone_extend);
1362 nf_conntrack_proto_fini();
1363 nf_conntrack_seqadj_fini();
1364 nf_conntrack_labels_fini();
1365 nf_conntrack_helper_fini();
1366 nf_conntrack_timeout_fini();
1367 nf_conntrack_ecache_fini();
1368 nf_conntrack_tstamp_fini();
1369 nf_conntrack_acct_fini();
1370 nf_conntrack_expect_fini();
1374 * Mishearing the voices in his head, our hero wonders how he's
1375 * supposed to kill the mall.
1377 void nf_conntrack_cleanup_net(struct net *net)
1381 list_add(&net->exit_list, &single);
1382 nf_conntrack_cleanup_net_list(&single);
1385 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1391 * This makes sure all current packets have passed through
1392 * netfilter framework. Roll on, two-stage module
1398 list_for_each_entry(net, net_exit_list, exit_list) {
1399 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1400 nf_ct_release_dying_list(net);
1401 if (atomic_read(&net->ct.count) != 0)
1406 goto i_see_dead_people;
1409 list_for_each_entry(net, net_exit_list, exit_list) {
1410 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1411 nf_conntrack_proto_pernet_fini(net);
1412 nf_conntrack_helper_pernet_fini(net);
1413 nf_conntrack_ecache_pernet_fini(net);
1414 nf_conntrack_tstamp_pernet_fini(net);
1415 nf_conntrack_acct_pernet_fini(net);
1416 nf_conntrack_expect_pernet_fini(net);
1417 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1418 kfree(net->ct.slabname);
1419 free_percpu(net->ct.stat);
1423 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1425 struct hlist_nulls_head *hash;
1426 unsigned int nr_slots, i;
1429 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1430 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1431 sz = nr_slots * sizeof(struct hlist_nulls_head);
1432 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1435 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1440 for (i = 0; i < nr_slots; i++)
1441 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1445 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1447 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1450 unsigned int hashsize, old_size;
1451 struct hlist_nulls_head *hash, *old_hash;
1452 struct nf_conntrack_tuple_hash *h;
1455 if (current->nsproxy->net_ns != &init_net)
1458 /* On boot, we can set this without any fancy locking. */
1459 if (!nf_conntrack_htable_size)
1460 return param_set_uint(val, kp);
1462 rc = kstrtouint(val, 0, &hashsize);
1468 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1472 /* Lookups in the old hash might happen in parallel, which means we
1473 * might get false negatives during connection lookup. New connections
1474 * created because of a false negative won't make it into the hash
1475 * though since that required taking the lock.
1477 spin_lock_bh(&nf_conntrack_lock);
1478 for (i = 0; i < init_net.ct.htable_size; i++) {
1479 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1480 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1481 struct nf_conntrack_tuple_hash, hnnode);
1482 ct = nf_ct_tuplehash_to_ctrack(h);
1483 hlist_nulls_del_rcu(&h->hnnode);
1484 bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1486 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1489 old_size = init_net.ct.htable_size;
1490 old_hash = init_net.ct.hash;
1492 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1493 init_net.ct.hash = hash;
1494 spin_unlock_bh(&nf_conntrack_lock);
1496 nf_ct_free_hashtable(old_hash, old_size);
1499 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1501 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1502 &nf_conntrack_htable_size, 0600);
1504 void nf_ct_untracked_status_or(unsigned long bits)
1508 for_each_possible_cpu(cpu)
1509 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1511 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1513 int nf_conntrack_init_start(void)
1518 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1519 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1520 if (!nf_conntrack_htable_size) {
1521 nf_conntrack_htable_size
1522 = (((totalram_pages << PAGE_SHIFT) / 16384)
1523 / sizeof(struct hlist_head));
1524 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1525 nf_conntrack_htable_size = 16384;
1526 if (nf_conntrack_htable_size < 32)
1527 nf_conntrack_htable_size = 32;
1529 /* Use a max. factor of four by default to get the same max as
1530 * with the old struct list_heads. When a table size is given
1531 * we use the old value of 8 to avoid reducing the max.
1535 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1537 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1538 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1541 ret = nf_conntrack_expect_init();
1545 ret = nf_conntrack_acct_init();
1549 ret = nf_conntrack_tstamp_init();
1553 ret = nf_conntrack_ecache_init();
1557 ret = nf_conntrack_timeout_init();
1561 ret = nf_conntrack_helper_init();
1565 ret = nf_conntrack_labels_init();
1569 ret = nf_conntrack_seqadj_init();
1573 #ifdef CONFIG_NF_CONNTRACK_ZONES
1574 ret = nf_ct_extend_register(&nf_ct_zone_extend);
1578 ret = nf_conntrack_proto_init();
1582 /* Set up fake conntrack: to never be deleted, not in any hashes */
1583 for_each_possible_cpu(cpu) {
1584 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1585 write_pnet(&ct->ct_net, &init_net);
1586 atomic_set(&ct->ct_general.use, 1);
1588 /* - and look it like as a confirmed connection */
1589 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1593 #ifdef CONFIG_NF_CONNTRACK_ZONES
1594 nf_ct_extend_unregister(&nf_ct_zone_extend);
1597 nf_conntrack_seqadj_fini();
1599 nf_conntrack_labels_fini();
1601 nf_conntrack_helper_fini();
1603 nf_conntrack_timeout_fini();
1605 nf_conntrack_ecache_fini();
1607 nf_conntrack_tstamp_fini();
1609 nf_conntrack_acct_fini();
1611 nf_conntrack_expect_fini();
1616 void nf_conntrack_init_end(void)
1618 /* For use by REJECT target */
1619 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1620 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1624 * We need to use special "null" values, not used in hash table
1626 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1627 #define DYING_NULLS_VAL ((1<<30)+1)
1628 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1630 int nf_conntrack_init_net(struct net *net)
1634 atomic_set(&net->ct.count, 0);
1635 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1636 INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1637 INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1638 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1639 if (!net->ct.stat) {
1644 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1645 if (!net->ct.slabname) {
1650 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1651 sizeof(struct nf_conn), 0,
1652 SLAB_DESTROY_BY_RCU, NULL);
1653 if (!net->ct.nf_conntrack_cachep) {
1654 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1659 net->ct.htable_size = nf_conntrack_htable_size;
1660 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1661 if (!net->ct.hash) {
1663 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1666 ret = nf_conntrack_expect_pernet_init(net);
1669 ret = nf_conntrack_acct_pernet_init(net);
1672 ret = nf_conntrack_tstamp_pernet_init(net);
1675 ret = nf_conntrack_ecache_pernet_init(net);
1678 ret = nf_conntrack_helper_pernet_init(net);
1681 ret = nf_conntrack_proto_pernet_init(net);
1687 nf_conntrack_helper_pernet_fini(net);
1689 nf_conntrack_ecache_pernet_fini(net);
1691 nf_conntrack_tstamp_pernet_fini(net);
1693 nf_conntrack_acct_pernet_fini(net);
1695 nf_conntrack_expect_pernet_fini(net);
1697 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1699 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1701 kfree(net->ct.slabname);
1703 free_percpu(net->ct.stat);