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>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/proc_fs.h>
19 #include <linux/vmalloc.h>
20 #include <linux/stddef.h>
21 #include <linux/slab.h>
22 #include <linux/random.h>
23 #include <linux/jhash.h>
24 #include <linux/err.h>
25 #include <linux/percpu.h>
26 #include <linux/moduleparam.h>
27 #include <linux/notifier.h>
28 #include <linux/kernel.h>
29 #include <linux/netdevice.h>
30 #include <linux/socket.h>
32 #include <linux/rculist_nulls.h>
34 #include <net/netfilter/nf_conntrack.h>
35 #include <net/netfilter/nf_conntrack_l3proto.h>
36 #include <net/netfilter/nf_conntrack_l4proto.h>
37 #include <net/netfilter/nf_conntrack_expect.h>
38 #include <net/netfilter/nf_conntrack_helper.h>
39 #include <net/netfilter/nf_conntrack_core.h>
40 #include <net/netfilter/nf_conntrack_extend.h>
41 #include <net/netfilter/nf_conntrack_acct.h>
42 #include <net/netfilter/nf_conntrack_ecache.h>
43 #include <net/netfilter/nf_nat.h>
44 #include <net/netfilter/nf_nat_core.h>
46 #define NF_CONNTRACK_VERSION "0.5.0"
48 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
49 enum nf_nat_manip_type manip,
50 const struct nlattr *attr) __read_mostly;
51 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
53 DEFINE_SPINLOCK(nf_conntrack_lock);
54 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
56 unsigned int nf_conntrack_htable_size __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
59 unsigned int nf_conntrack_max __read_mostly;
60 EXPORT_SYMBOL_GPL(nf_conntrack_max);
62 struct nf_conn nf_conntrack_untracked __read_mostly;
63 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
67 static int nf_conntrack_hash_rnd_initted;
68 static unsigned int nf_conntrack_hash_rnd;
70 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
71 unsigned int size, unsigned int rnd)
76 /* The direction must be ignored, so we hash everything up to the
77 * destination ports (which is a multiple of 4) and treat the last
78 * three bytes manually.
80 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
81 h = jhash2((u32 *)tuple, n,
82 rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
83 tuple->dst.protonum));
85 return ((u64)h * size) >> 32;
88 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
90 return __hash_conntrack(tuple, nf_conntrack_htable_size,
91 nf_conntrack_hash_rnd);
95 nf_ct_get_tuple(const struct sk_buff *skb,
100 struct nf_conntrack_tuple *tuple,
101 const struct nf_conntrack_l3proto *l3proto,
102 const struct nf_conntrack_l4proto *l4proto)
104 memset(tuple, 0, sizeof(*tuple));
106 tuple->src.l3num = l3num;
107 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
110 tuple->dst.protonum = protonum;
111 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
113 return l4proto->pkt_to_tuple(skb, dataoff, tuple);
115 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
117 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
118 u_int16_t l3num, struct nf_conntrack_tuple *tuple)
120 struct nf_conntrack_l3proto *l3proto;
121 struct nf_conntrack_l4proto *l4proto;
122 unsigned int protoff;
128 l3proto = __nf_ct_l3proto_find(l3num);
129 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
130 if (ret != NF_ACCEPT) {
135 l4proto = __nf_ct_l4proto_find(l3num, protonum);
137 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
143 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
146 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
147 const struct nf_conntrack_tuple *orig,
148 const struct nf_conntrack_l3proto *l3proto,
149 const struct nf_conntrack_l4proto *l4proto)
151 memset(inverse, 0, sizeof(*inverse));
153 inverse->src.l3num = orig->src.l3num;
154 if (l3proto->invert_tuple(inverse, orig) == 0)
157 inverse->dst.dir = !orig->dst.dir;
159 inverse->dst.protonum = orig->dst.protonum;
160 return l4proto->invert_tuple(inverse, orig);
162 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
165 clean_from_lists(struct nf_conn *ct)
167 pr_debug("clean_from_lists(%p)\n", ct);
168 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
169 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
171 /* Destroy all pending expectations */
172 nf_ct_remove_expectations(ct);
176 destroy_conntrack(struct nf_conntrack *nfct)
178 struct nf_conn *ct = (struct nf_conn *)nfct;
179 struct net *net = nf_ct_net(ct);
180 struct nf_conntrack_l4proto *l4proto;
182 pr_debug("destroy_conntrack(%p)\n", ct);
183 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
184 NF_CT_ASSERT(!timer_pending(&ct->timeout));
186 /* To make sure we don't get any weird locking issues here:
187 * destroy_conntrack() MUST NOT be called with a write lock
188 * to nf_conntrack_lock!!! -HW */
190 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
191 if (l4proto && l4proto->destroy)
192 l4proto->destroy(ct);
196 spin_lock_bh(&nf_conntrack_lock);
197 /* Expectations will have been removed in clean_from_lists,
198 * except TFTP can create an expectation on the first packet,
199 * before connection is in the list, so we need to clean here,
201 nf_ct_remove_expectations(ct);
203 /* We overload first tuple to link into unconfirmed list. */
204 if (!nf_ct_is_confirmed(ct)) {
205 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
206 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
209 NF_CT_STAT_INC(net, delete);
210 spin_unlock_bh(&nf_conntrack_lock);
213 nf_ct_put(ct->master);
215 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
216 nf_conntrack_free(ct);
219 void nf_ct_delete_from_lists(struct nf_conn *ct)
221 struct net *net = nf_ct_net(ct);
223 nf_ct_helper_destroy(ct);
224 spin_lock_bh(&nf_conntrack_lock);
225 /* Inside lock so preempt is disabled on module removal path.
226 * Otherwise we can get spurious warnings. */
227 NF_CT_STAT_INC(net, delete_list);
228 clean_from_lists(ct);
229 spin_unlock_bh(&nf_conntrack_lock);
231 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
233 static void death_by_event(unsigned long ul_conntrack)
235 struct nf_conn *ct = (void *)ul_conntrack;
236 struct net *net = nf_ct_net(ct);
238 if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
239 /* bad luck, let's retry again */
240 ct->timeout.expires = jiffies +
241 (random32() % net->ct.sysctl_events_retry_timeout);
242 add_timer(&ct->timeout);
245 /* we've got the event delivered, now it's dying */
246 set_bit(IPS_DYING_BIT, &ct->status);
247 spin_lock(&nf_conntrack_lock);
248 hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
249 spin_unlock(&nf_conntrack_lock);
253 void nf_ct_insert_dying_list(struct nf_conn *ct)
255 struct net *net = nf_ct_net(ct);
257 /* add this conntrack to the dying list */
258 spin_lock_bh(&nf_conntrack_lock);
259 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
261 spin_unlock_bh(&nf_conntrack_lock);
262 /* set a new timer to retry event delivery */
263 setup_timer(&ct->timeout, death_by_event, (unsigned long)ct);
264 ct->timeout.expires = jiffies +
265 (random32() % net->ct.sysctl_events_retry_timeout);
266 add_timer(&ct->timeout);
268 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list);
270 static void death_by_timeout(unsigned long ul_conntrack)
272 struct nf_conn *ct = (void *)ul_conntrack;
274 if (!test_bit(IPS_DYING_BIT, &ct->status) &&
275 unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
276 /* destroy event was not delivered */
277 nf_ct_delete_from_lists(ct);
278 nf_ct_insert_dying_list(ct);
281 set_bit(IPS_DYING_BIT, &ct->status);
282 nf_ct_delete_from_lists(ct);
288 * - Caller must take a reference on returned object
289 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
291 * - Caller must lock nf_conntrack_lock before calling this function
293 struct nf_conntrack_tuple_hash *
294 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
296 struct nf_conntrack_tuple_hash *h;
297 struct hlist_nulls_node *n;
298 unsigned int hash = hash_conntrack(tuple);
300 /* Disable BHs the entire time since we normally need to disable them
301 * at least once for the stats anyway.
305 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
306 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
307 NF_CT_STAT_INC(net, found);
311 NF_CT_STAT_INC(net, searched);
314 * if the nulls value we got at the end of this lookup is
315 * not the expected one, we must restart lookup.
316 * We probably met an item that was moved to another chain.
318 if (get_nulls_value(n) != hash)
324 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
326 /* Find a connection corresponding to a tuple. */
327 struct nf_conntrack_tuple_hash *
328 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
330 struct nf_conntrack_tuple_hash *h;
335 h = __nf_conntrack_find(net, tuple);
337 ct = nf_ct_tuplehash_to_ctrack(h);
338 if (unlikely(nf_ct_is_dying(ct) ||
339 !atomic_inc_not_zero(&ct->ct_general.use)))
342 if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
352 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
354 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
356 unsigned int repl_hash)
358 struct net *net = nf_ct_net(ct);
360 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
361 &net->ct.hash[hash]);
362 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
363 &net->ct.hash[repl_hash]);
366 void nf_conntrack_hash_insert(struct nf_conn *ct)
368 unsigned int hash, repl_hash;
370 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
371 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
373 __nf_conntrack_hash_insert(ct, hash, repl_hash);
375 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
377 /* Confirm a connection given skb; places it in hash table */
379 __nf_conntrack_confirm(struct sk_buff *skb)
381 unsigned int hash, repl_hash;
382 struct nf_conntrack_tuple_hash *h;
384 struct nf_conn_help *help;
385 struct hlist_nulls_node *n;
386 enum ip_conntrack_info ctinfo;
389 ct = nf_ct_get(skb, &ctinfo);
392 /* ipt_REJECT uses nf_conntrack_attach to attach related
393 ICMP/TCP RST packets in other direction. Actual packet
394 which created connection will be IP_CT_NEW or for an
395 expected connection, IP_CT_RELATED. */
396 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
399 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
400 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
402 /* We're not in hash table, and we refuse to set up related
403 connections for unconfirmed conns. But packet copies and
404 REJECT will give spurious warnings here. */
405 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
407 /* No external references means noone else could have
409 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
410 pr_debug("Confirming conntrack %p\n", ct);
412 spin_lock_bh(&nf_conntrack_lock);
414 /* See if there's one in the list already, including reverse:
415 NAT could have grabbed it without realizing, since we're
416 not in the hash. If there is, we lost race. */
417 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
418 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
421 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
422 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
426 /* Remove from unconfirmed list */
427 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
429 /* Timer relative to confirmation time, not original
430 setting time, otherwise we'd get timer wrap in
431 weird delay cases. */
432 ct->timeout.expires += jiffies;
433 add_timer(&ct->timeout);
434 atomic_inc(&ct->ct_general.use);
435 set_bit(IPS_CONFIRMED_BIT, &ct->status);
437 /* Since the lookup is lockless, hash insertion must be done after
438 * starting the timer and setting the CONFIRMED bit. The RCU barriers
439 * guarantee that no other CPU can find the conntrack before the above
440 * stores are visible.
442 __nf_conntrack_hash_insert(ct, hash, repl_hash);
443 NF_CT_STAT_INC(net, insert);
444 spin_unlock_bh(&nf_conntrack_lock);
446 help = nfct_help(ct);
447 if (help && help->helper)
448 nf_conntrack_event_cache(IPCT_HELPER, ct);
450 nf_conntrack_event_cache(master_ct(ct) ?
451 IPCT_RELATED : IPCT_NEW, ct);
455 NF_CT_STAT_INC(net, insert_failed);
456 spin_unlock_bh(&nf_conntrack_lock);
459 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
461 /* Returns true if a connection correspondings to the tuple (required
464 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
465 const struct nf_conn *ignored_conntrack)
467 struct net *net = nf_ct_net(ignored_conntrack);
468 struct nf_conntrack_tuple_hash *h;
469 struct hlist_nulls_node *n;
470 unsigned int hash = hash_conntrack(tuple);
472 /* Disable BHs the entire time since we need to disable them at
473 * least once for the stats anyway.
476 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
477 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
478 nf_ct_tuple_equal(tuple, &h->tuple)) {
479 NF_CT_STAT_INC(net, found);
480 rcu_read_unlock_bh();
483 NF_CT_STAT_INC(net, searched);
485 rcu_read_unlock_bh();
489 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
491 #define NF_CT_EVICTION_RANGE 8
493 /* There's a small race here where we may free a just-assured
494 connection. Too bad: we're in trouble anyway. */
495 static noinline int early_drop(struct net *net, unsigned int hash)
497 /* Use oldest entry, which is roughly LRU */
498 struct nf_conntrack_tuple_hash *h;
499 struct nf_conn *ct = NULL, *tmp;
500 struct hlist_nulls_node *n;
501 unsigned int i, cnt = 0;
505 for (i = 0; i < nf_conntrack_htable_size; i++) {
506 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
508 tmp = nf_ct_tuplehash_to_ctrack(h);
509 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
515 if (likely(!nf_ct_is_dying(ct) &&
516 atomic_inc_not_zero(&ct->ct_general.use)))
522 if (cnt >= NF_CT_EVICTION_RANGE)
525 hash = (hash + 1) % nf_conntrack_htable_size;
532 if (del_timer(&ct->timeout)) {
533 death_by_timeout((unsigned long)ct);
535 NF_CT_STAT_INC_ATOMIC(net, early_drop);
541 struct nf_conn *nf_conntrack_alloc(struct net *net,
542 const struct nf_conntrack_tuple *orig,
543 const struct nf_conntrack_tuple *repl,
548 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
549 get_random_bytes(&nf_conntrack_hash_rnd,
550 sizeof(nf_conntrack_hash_rnd));
551 nf_conntrack_hash_rnd_initted = 1;
554 /* We don't want any race condition at early drop stage */
555 atomic_inc(&net->ct.count);
557 if (nf_conntrack_max &&
558 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
559 unsigned int hash = hash_conntrack(orig);
560 if (!early_drop(net, hash)) {
561 atomic_dec(&net->ct.count);
564 "nf_conntrack: table full, dropping"
566 return ERR_PTR(-ENOMEM);
571 * Do not use kmem_cache_zalloc(), as this cache uses
572 * SLAB_DESTROY_BY_RCU.
574 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
576 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
577 atomic_dec(&net->ct.count);
578 return ERR_PTR(-ENOMEM);
581 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
582 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
584 memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
585 sizeof(*ct) - offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
586 spin_lock_init(&ct->lock);
587 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
588 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
589 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
590 ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev = NULL;
591 /* Don't set timer yet: wait for confirmation */
592 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
598 * changes to lookup keys must be done before setting refcnt to 1
601 atomic_set(&ct->ct_general.use, 1);
604 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
606 void nf_conntrack_free(struct nf_conn *ct)
608 struct net *net = nf_ct_net(ct);
610 nf_ct_ext_destroy(ct);
611 atomic_dec(&net->ct.count);
613 kmem_cache_free(nf_conntrack_cachep, ct);
615 EXPORT_SYMBOL_GPL(nf_conntrack_free);
617 /* Allocate a new conntrack: we return -ENOMEM if classification
618 failed due to stress. Otherwise it really is unclassifiable. */
619 static struct nf_conntrack_tuple_hash *
620 init_conntrack(struct net *net,
621 const struct nf_conntrack_tuple *tuple,
622 struct nf_conntrack_l3proto *l3proto,
623 struct nf_conntrack_l4proto *l4proto,
625 unsigned int dataoff)
628 struct nf_conn_help *help;
629 struct nf_conntrack_tuple repl_tuple;
630 struct nf_conntrack_expect *exp;
632 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
633 pr_debug("Can't invert tuple.\n");
637 ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
639 pr_debug("Can't allocate conntrack.\n");
640 return (struct nf_conntrack_tuple_hash *)ct;
643 if (!l4proto->new(ct, skb, dataoff)) {
644 nf_conntrack_free(ct);
645 pr_debug("init conntrack: can't track with proto module\n");
649 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
650 nf_ct_ecache_ext_add(ct, GFP_ATOMIC);
652 spin_lock_bh(&nf_conntrack_lock);
653 exp = nf_ct_find_expectation(net, tuple);
655 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
657 /* Welcome, Mr. Bond. We've been expecting you... */
658 __set_bit(IPS_EXPECTED_BIT, &ct->status);
659 ct->master = exp->master;
661 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
663 rcu_assign_pointer(help->helper, exp->helper);
666 #ifdef CONFIG_NF_CONNTRACK_MARK
667 ct->mark = exp->master->mark;
669 #ifdef CONFIG_NF_CONNTRACK_SECMARK
670 ct->secmark = exp->master->secmark;
672 nf_conntrack_get(&ct->master->ct_general);
673 NF_CT_STAT_INC(net, expect_new);
675 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
676 NF_CT_STAT_INC(net, new);
679 /* Overload tuple linked list to put us in unconfirmed list. */
680 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
681 &net->ct.unconfirmed);
683 spin_unlock_bh(&nf_conntrack_lock);
687 exp->expectfn(ct, exp);
688 nf_ct_expect_put(exp);
691 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
694 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
695 static inline struct nf_conn *
696 resolve_normal_ct(struct net *net,
698 unsigned int dataoff,
701 struct nf_conntrack_l3proto *l3proto,
702 struct nf_conntrack_l4proto *l4proto,
704 enum ip_conntrack_info *ctinfo)
706 struct nf_conntrack_tuple tuple;
707 struct nf_conntrack_tuple_hash *h;
710 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
711 dataoff, l3num, protonum, &tuple, l3proto,
713 pr_debug("resolve_normal_ct: Can't get tuple\n");
717 /* look for tuple match */
718 h = nf_conntrack_find_get(net, &tuple);
720 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
726 ct = nf_ct_tuplehash_to_ctrack(h);
728 /* It exists; we have (non-exclusive) reference. */
729 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
730 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
731 /* Please set reply bit if this packet OK */
734 /* Once we've had two way comms, always ESTABLISHED. */
735 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
736 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
737 *ctinfo = IP_CT_ESTABLISHED;
738 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
739 pr_debug("nf_conntrack_in: related packet for %p\n",
741 *ctinfo = IP_CT_RELATED;
743 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
748 skb->nfct = &ct->ct_general;
749 skb->nfctinfo = *ctinfo;
754 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
758 enum ip_conntrack_info ctinfo;
759 struct nf_conntrack_l3proto *l3proto;
760 struct nf_conntrack_l4proto *l4proto;
761 unsigned int dataoff;
766 /* Previously seen (loopback or untracked)? Ignore. */
768 NF_CT_STAT_INC_ATOMIC(net, ignore);
772 /* rcu_read_lock()ed by nf_hook_slow */
773 l3proto = __nf_ct_l3proto_find(pf);
774 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
775 &dataoff, &protonum);
777 pr_debug("not prepared to track yet or error occured\n");
778 NF_CT_STAT_INC_ATOMIC(net, error);
779 NF_CT_STAT_INC_ATOMIC(net, invalid);
783 l4proto = __nf_ct_l4proto_find(pf, protonum);
785 /* It may be an special packet, error, unclean...
786 * inverse of the return code tells to the netfilter
787 * core what to do with the packet. */
788 if (l4proto->error != NULL) {
789 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
791 NF_CT_STAT_INC_ATOMIC(net, error);
792 NF_CT_STAT_INC_ATOMIC(net, invalid);
797 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
798 l3proto, l4proto, &set_reply, &ctinfo);
800 /* Not valid part of a connection */
801 NF_CT_STAT_INC_ATOMIC(net, invalid);
806 /* Too stressed to deal. */
807 NF_CT_STAT_INC_ATOMIC(net, drop);
811 NF_CT_ASSERT(skb->nfct);
813 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
815 /* Invalid: inverse of the return code tells
816 * the netfilter core what to do */
817 pr_debug("nf_conntrack_in: Can't track with proto module\n");
818 nf_conntrack_put(skb->nfct);
820 NF_CT_STAT_INC_ATOMIC(net, invalid);
822 NF_CT_STAT_INC_ATOMIC(net, drop);
826 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
827 nf_conntrack_event_cache(IPCT_STATUS, ct);
831 EXPORT_SYMBOL_GPL(nf_conntrack_in);
833 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
834 const struct nf_conntrack_tuple *orig)
839 ret = nf_ct_invert_tuple(inverse, orig,
840 __nf_ct_l3proto_find(orig->src.l3num),
841 __nf_ct_l4proto_find(orig->src.l3num,
842 orig->dst.protonum));
846 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
848 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
849 implicitly racy: see __nf_conntrack_confirm */
850 void nf_conntrack_alter_reply(struct nf_conn *ct,
851 const struct nf_conntrack_tuple *newreply)
853 struct nf_conn_help *help = nfct_help(ct);
855 /* Should be unconfirmed, so not in hash table yet */
856 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
858 pr_debug("Altering reply tuple of %p to ", ct);
859 nf_ct_dump_tuple(newreply);
861 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
862 if (ct->master || (help && !hlist_empty(&help->expectations)))
866 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
869 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
871 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
872 void __nf_ct_refresh_acct(struct nf_conn *ct,
873 enum ip_conntrack_info ctinfo,
874 const struct sk_buff *skb,
875 unsigned long extra_jiffies,
878 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
881 /* Only update if this is not a fixed timeout */
882 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
885 /* If not in hash table, timer will not be active yet */
886 if (!nf_ct_is_confirmed(ct)) {
887 ct->timeout.expires = extra_jiffies;
889 unsigned long newtime = jiffies + extra_jiffies;
891 /* Only update the timeout if the new timeout is at least
892 HZ jiffies from the old timeout. Need del_timer for race
893 avoidance (may already be dying). */
894 if (newtime - ct->timeout.expires >= HZ)
895 mod_timer_pending(&ct->timeout, newtime);
900 struct nf_conn_counter *acct;
902 acct = nf_conn_acct_find(ct);
904 spin_lock_bh(&ct->lock);
905 acct[CTINFO2DIR(ctinfo)].packets++;
906 acct[CTINFO2DIR(ctinfo)].bytes +=
907 skb->len - skb_network_offset(skb);
908 spin_unlock_bh(&ct->lock);
912 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
914 bool __nf_ct_kill_acct(struct nf_conn *ct,
915 enum ip_conntrack_info ctinfo,
916 const struct sk_buff *skb,
920 struct nf_conn_counter *acct;
922 acct = nf_conn_acct_find(ct);
924 spin_lock_bh(&ct->lock);
925 acct[CTINFO2DIR(ctinfo)].packets++;
926 acct[CTINFO2DIR(ctinfo)].bytes +=
927 skb->len - skb_network_offset(skb);
928 spin_unlock_bh(&ct->lock);
932 if (del_timer(&ct->timeout)) {
933 ct->timeout.function((unsigned long)ct);
938 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
940 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
942 #include <linux/netfilter/nfnetlink.h>
943 #include <linux/netfilter/nfnetlink_conntrack.h>
944 #include <linux/mutex.h>
946 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
947 * in ip_conntrack_core, since we don't want the protocols to autoload
948 * or depend on ctnetlink */
949 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
950 const struct nf_conntrack_tuple *tuple)
952 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
953 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
959 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
961 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
962 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
963 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
965 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
967 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
968 struct nf_conntrack_tuple *t)
970 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
973 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
974 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
978 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
980 int nf_ct_port_nlattr_tuple_size(void)
982 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
984 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
987 /* Used by ipt_REJECT and ip6t_REJECT. */
988 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
991 enum ip_conntrack_info ctinfo;
993 /* This ICMP is in reverse direction to the packet which caused it */
994 ct = nf_ct_get(skb, &ctinfo);
995 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
996 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
998 ctinfo = IP_CT_RELATED;
1000 /* Attach to new skbuff, and increment count */
1001 nskb->nfct = &ct->ct_general;
1002 nskb->nfctinfo = ctinfo;
1003 nf_conntrack_get(nskb->nfct);
1006 /* Bring out ya dead! */
1007 static struct nf_conn *
1008 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1009 void *data, unsigned int *bucket)
1011 struct nf_conntrack_tuple_hash *h;
1013 struct hlist_nulls_node *n;
1015 spin_lock_bh(&nf_conntrack_lock);
1016 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1017 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1018 ct = nf_ct_tuplehash_to_ctrack(h);
1023 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1024 ct = nf_ct_tuplehash_to_ctrack(h);
1026 set_bit(IPS_DYING_BIT, &ct->status);
1028 spin_unlock_bh(&nf_conntrack_lock);
1031 atomic_inc(&ct->ct_general.use);
1032 spin_unlock_bh(&nf_conntrack_lock);
1036 void nf_ct_iterate_cleanup(struct net *net,
1037 int (*iter)(struct nf_conn *i, void *data),
1041 unsigned int bucket = 0;
1043 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1044 /* Time to push up daises... */
1045 if (del_timer(&ct->timeout))
1046 death_by_timeout((unsigned long)ct);
1047 /* ... else the timer will get him soon. */
1052 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1054 struct __nf_ct_flush_report {
1059 static int kill_report(struct nf_conn *i, void *data)
1061 struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1063 /* If we fail to deliver the event, death_by_timeout() will retry */
1064 if (nf_conntrack_event_report(IPCT_DESTROY, i,
1065 fr->pid, fr->report) < 0)
1068 /* Avoid the delivery of the destroy event in death_by_timeout(). */
1069 set_bit(IPS_DYING_BIT, &i->status);
1073 static int kill_all(struct nf_conn *i, void *data)
1078 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1083 free_pages((unsigned long)hash,
1084 get_order(sizeof(struct hlist_head) * size));
1086 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1088 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1090 struct __nf_ct_flush_report fr = {
1094 nf_ct_iterate_cleanup(net, kill_report, &fr);
1096 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1098 static void nf_ct_release_dying_list(struct net *net)
1100 struct nf_conntrack_tuple_hash *h;
1102 struct hlist_nulls_node *n;
1104 spin_lock_bh(&nf_conntrack_lock);
1105 hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1106 ct = nf_ct_tuplehash_to_ctrack(h);
1107 /* never fails to remove them, no listeners at this point */
1110 spin_unlock_bh(&nf_conntrack_lock);
1113 static void nf_conntrack_cleanup_init_net(void)
1115 nf_conntrack_helper_fini();
1116 nf_conntrack_proto_fini();
1117 kmem_cache_destroy(nf_conntrack_cachep);
1120 static void nf_conntrack_cleanup_net(struct net *net)
1123 nf_ct_iterate_cleanup(net, kill_all, NULL);
1124 nf_ct_release_dying_list(net);
1125 if (atomic_read(&net->ct.count) != 0) {
1127 goto i_see_dead_people;
1129 /* wait until all references to nf_conntrack_untracked are dropped */
1130 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1133 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1134 nf_conntrack_htable_size);
1135 nf_conntrack_ecache_fini(net);
1136 nf_conntrack_acct_fini(net);
1137 nf_conntrack_expect_fini(net);
1138 free_percpu(net->ct.stat);
1141 /* Mishearing the voices in his head, our hero wonders how he's
1142 supposed to kill the mall. */
1143 void nf_conntrack_cleanup(struct net *net)
1145 if (net_eq(net, &init_net))
1146 rcu_assign_pointer(ip_ct_attach, NULL);
1148 /* This makes sure all current packets have passed through
1149 netfilter framework. Roll on, two-stage module
1153 nf_conntrack_cleanup_net(net);
1155 if (net_eq(net, &init_net)) {
1156 rcu_assign_pointer(nf_ct_destroy, NULL);
1157 nf_conntrack_cleanup_init_net();
1161 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1163 struct hlist_nulls_head *hash;
1164 unsigned int nr_slots, i;
1169 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1170 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1171 sz = nr_slots * sizeof(struct hlist_nulls_head);
1172 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1176 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1177 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1181 for (i = 0; i < nr_slots; i++)
1182 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1186 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1188 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1190 int i, bucket, vmalloced, old_vmalloced;
1191 unsigned int hashsize, old_size;
1193 struct hlist_nulls_head *hash, *old_hash;
1194 struct nf_conntrack_tuple_hash *h;
1196 /* On boot, we can set this without any fancy locking. */
1197 if (!nf_conntrack_htable_size)
1198 return param_set_uint(val, kp);
1200 hashsize = simple_strtoul(val, NULL, 0);
1204 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1208 /* We have to rehahs for the new table anyway, so we also can
1209 * use a newrandom seed */
1210 get_random_bytes(&rnd, sizeof(rnd));
1212 /* Lookups in the old hash might happen in parallel, which means we
1213 * might get false negatives during connection lookup. New connections
1214 * created because of a false negative won't make it into the hash
1215 * though since that required taking the lock.
1217 spin_lock_bh(&nf_conntrack_lock);
1218 for (i = 0; i < nf_conntrack_htable_size; i++) {
1219 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1220 h = hlist_nulls_entry(init_net.ct.hash[i].first,
1221 struct nf_conntrack_tuple_hash, hnnode);
1222 hlist_nulls_del_rcu(&h->hnnode);
1223 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1224 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1227 old_size = nf_conntrack_htable_size;
1228 old_vmalloced = init_net.ct.hash_vmalloc;
1229 old_hash = init_net.ct.hash;
1231 nf_conntrack_htable_size = hashsize;
1232 init_net.ct.hash_vmalloc = vmalloced;
1233 init_net.ct.hash = hash;
1234 nf_conntrack_hash_rnd = rnd;
1235 spin_unlock_bh(&nf_conntrack_lock);
1237 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1240 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1242 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1243 &nf_conntrack_htable_size, 0600);
1245 static int nf_conntrack_init_init_net(void)
1250 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1251 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1252 if (!nf_conntrack_htable_size) {
1253 nf_conntrack_htable_size
1254 = (((totalram_pages << PAGE_SHIFT) / 16384)
1255 / sizeof(struct hlist_head));
1256 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1257 nf_conntrack_htable_size = 16384;
1258 if (nf_conntrack_htable_size < 32)
1259 nf_conntrack_htable_size = 32;
1261 /* Use a max. factor of four by default to get the same max as
1262 * with the old struct list_heads. When a table size is given
1263 * we use the old value of 8 to avoid reducing the max.
1267 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1269 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1270 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1273 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1274 sizeof(struct nf_conn),
1275 0, SLAB_DESTROY_BY_RCU, NULL);
1276 if (!nf_conntrack_cachep) {
1277 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1282 ret = nf_conntrack_proto_init();
1286 ret = nf_conntrack_helper_init();
1293 nf_conntrack_proto_fini();
1295 kmem_cache_destroy(nf_conntrack_cachep);
1301 * We need to use special "null" values, not used in hash table
1303 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1304 #define DYING_NULLS_VAL ((1<<30)+1)
1306 static int nf_conntrack_init_net(struct net *net)
1310 atomic_set(&net->ct.count, 0);
1311 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1312 INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1313 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1314 if (!net->ct.stat) {
1318 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1319 &net->ct.hash_vmalloc, 1);
1320 if (!net->ct.hash) {
1322 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1325 ret = nf_conntrack_expect_init(net);
1328 ret = nf_conntrack_acct_init(net);
1331 ret = nf_conntrack_ecache_init(net);
1335 /* Set up fake conntrack:
1336 - to never be deleted, not in any hashes */
1337 #ifdef CONFIG_NET_NS
1338 nf_conntrack_untracked.ct_net = &init_net;
1340 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1341 /* - and look it like as a confirmed connection */
1342 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1347 nf_conntrack_acct_fini(net);
1349 nf_conntrack_expect_fini(net);
1351 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1352 nf_conntrack_htable_size);
1354 free_percpu(net->ct.stat);
1359 int nf_conntrack_init(struct net *net)
1363 if (net_eq(net, &init_net)) {
1364 ret = nf_conntrack_init_init_net();
1368 ret = nf_conntrack_init_net(net);
1372 if (net_eq(net, &init_net)) {
1373 /* For use by REJECT target */
1374 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1375 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1380 if (net_eq(net, &init_net))
1381 nf_conntrack_cleanup_init_net();