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>
33 #include <net/netfilter/nf_conntrack.h>
34 #include <net/netfilter/nf_conntrack_l3proto.h>
35 #include <net/netfilter/nf_conntrack_l4proto.h>
36 #include <net/netfilter/nf_conntrack_expect.h>
37 #include <net/netfilter/nf_conntrack_helper.h>
38 #include <net/netfilter/nf_conntrack_core.h>
39 #include <net/netfilter/nf_conntrack_extend.h>
41 #define NF_CONNTRACK_VERSION "0.5.0"
43 DEFINE_RWLOCK(nf_conntrack_lock);
44 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
46 /* nf_conntrack_standalone needs this */
47 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
48 EXPORT_SYMBOL_GPL(nf_conntrack_count);
50 unsigned int nf_conntrack_htable_size __read_mostly;
51 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
53 int nf_conntrack_max __read_mostly;
54 EXPORT_SYMBOL_GPL(nf_conntrack_max);
56 struct hlist_head *nf_conntrack_hash __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
59 struct nf_conn nf_conntrack_untracked __read_mostly;
60 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
62 unsigned int nf_ct_log_invalid __read_mostly;
63 HLIST_HEAD(unconfirmed);
64 static int nf_conntrack_vmalloc __read_mostly;
65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
66 static unsigned int nf_conntrack_next_id;
68 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
69 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
71 static int nf_conntrack_hash_rnd_initted;
72 static unsigned int nf_conntrack_hash_rnd;
74 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
75 unsigned int size, unsigned int rnd)
79 a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
80 (tuple->src.l3num << 16) | tuple->dst.protonum);
81 b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
82 ((__force __u16)tuple->src.u.all << 16) |
83 (__force __u16)tuple->dst.u.all);
85 return jhash_2words(a, b, rnd) % size;
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 NF_CT_TUPLE_U_BLANK(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 int nf_ct_get_tuplepr(const struct sk_buff *skb,
120 struct nf_conntrack_tuple *tuple)
122 struct nf_conntrack_l3proto *l3proto;
123 struct nf_conntrack_l4proto *l4proto;
124 unsigned int protoff;
130 l3proto = __nf_ct_l3proto_find(l3num);
131 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
132 if (ret != NF_ACCEPT) {
137 l4proto = __nf_ct_l4proto_find(l3num, protonum);
139 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
145 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
148 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
149 const struct nf_conntrack_tuple *orig,
150 const struct nf_conntrack_l3proto *l3proto,
151 const struct nf_conntrack_l4proto *l4proto)
153 NF_CT_TUPLE_U_BLANK(inverse);
155 inverse->src.l3num = orig->src.l3num;
156 if (l3proto->invert_tuple(inverse, orig) == 0)
159 inverse->dst.dir = !orig->dst.dir;
161 inverse->dst.protonum = orig->dst.protonum;
162 return l4proto->invert_tuple(inverse, orig);
164 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
167 clean_from_lists(struct nf_conn *ct)
169 pr_debug("clean_from_lists(%p)\n", ct);
170 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
171 hlist_del(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
173 /* Destroy all pending expectations */
174 nf_ct_remove_expectations(ct);
178 destroy_conntrack(struct nf_conntrack *nfct)
180 struct nf_conn *ct = (struct nf_conn *)nfct;
181 struct nf_conntrack_l4proto *l4proto;
183 pr_debug("destroy_conntrack(%p)\n", ct);
184 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
185 NF_CT_ASSERT(!timer_pending(&ct->timeout));
187 nf_conntrack_event(IPCT_DESTROY, ct);
188 set_bit(IPS_DYING_BIT, &ct->status);
190 /* To make sure we don't get any weird locking issues here:
191 * destroy_conntrack() MUST NOT be called with a write lock
192 * to nf_conntrack_lock!!! -HW */
194 l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
195 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
196 if (l4proto && l4proto->destroy)
197 l4proto->destroy(ct);
199 nf_ct_ext_destroy(ct);
203 write_lock_bh(&nf_conntrack_lock);
204 /* Expectations will have been removed in clean_from_lists,
205 * except TFTP can create an expectation on the first packet,
206 * before connection is in the list, so we need to clean here,
208 nf_ct_remove_expectations(ct);
210 /* We overload first tuple to link into unconfirmed list. */
211 if (!nf_ct_is_confirmed(ct)) {
212 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
213 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
216 NF_CT_STAT_INC(delete);
217 write_unlock_bh(&nf_conntrack_lock);
220 nf_ct_put(ct->master);
222 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
223 nf_conntrack_free(ct);
226 static void death_by_timeout(unsigned long ul_conntrack)
228 struct nf_conn *ct = (void *)ul_conntrack;
229 struct nf_conn_help *help = nfct_help(ct);
230 struct nf_conntrack_helper *helper;
234 helper = rcu_dereference(help->helper);
235 if (helper && helper->destroy)
240 write_lock_bh(&nf_conntrack_lock);
241 /* Inside lock so preempt is disabled on module removal path.
242 * Otherwise we can get spurious warnings. */
243 NF_CT_STAT_INC(delete_list);
244 clean_from_lists(ct);
245 write_unlock_bh(&nf_conntrack_lock);
249 struct nf_conntrack_tuple_hash *
250 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
251 const struct nf_conn *ignored_conntrack)
253 struct nf_conntrack_tuple_hash *h;
254 struct hlist_node *n;
255 unsigned int hash = hash_conntrack(tuple);
257 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
258 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
259 nf_ct_tuple_equal(tuple, &h->tuple)) {
260 NF_CT_STAT_INC(found);
263 NF_CT_STAT_INC(searched);
268 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
270 /* Find a connection corresponding to a tuple. */
271 struct nf_conntrack_tuple_hash *
272 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
274 struct nf_conntrack_tuple_hash *h;
276 read_lock_bh(&nf_conntrack_lock);
277 h = __nf_conntrack_find(tuple, NULL);
279 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
280 read_unlock_bh(&nf_conntrack_lock);
284 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
286 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
288 unsigned int repl_hash)
290 ct->id = ++nf_conntrack_next_id;
291 hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
292 &nf_conntrack_hash[hash]);
293 hlist_add_head(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
294 &nf_conntrack_hash[repl_hash]);
297 void nf_conntrack_hash_insert(struct nf_conn *ct)
299 unsigned int hash, repl_hash;
301 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
302 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
304 write_lock_bh(&nf_conntrack_lock);
305 __nf_conntrack_hash_insert(ct, hash, repl_hash);
306 write_unlock_bh(&nf_conntrack_lock);
308 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
310 /* Confirm a connection given skb; places it in hash table */
312 __nf_conntrack_confirm(struct sk_buff **pskb)
314 unsigned int hash, repl_hash;
315 struct nf_conntrack_tuple_hash *h;
317 struct nf_conn_help *help;
318 struct hlist_node *n;
319 enum ip_conntrack_info ctinfo;
321 ct = nf_ct_get(*pskb, &ctinfo);
323 /* ipt_REJECT uses nf_conntrack_attach to attach related
324 ICMP/TCP RST packets in other direction. Actual packet
325 which created connection will be IP_CT_NEW or for an
326 expected connection, IP_CT_RELATED. */
327 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
330 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
331 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
333 /* We're not in hash table, and we refuse to set up related
334 connections for unconfirmed conns. But packet copies and
335 REJECT will give spurious warnings here. */
336 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
338 /* No external references means noone else could have
340 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
341 pr_debug("Confirming conntrack %p\n", ct);
343 write_lock_bh(&nf_conntrack_lock);
345 /* See if there's one in the list already, including reverse:
346 NAT could have grabbed it without realizing, since we're
347 not in the hash. If there is, we lost race. */
348 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
349 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
352 hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
353 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
357 /* Remove from unconfirmed list */
358 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
360 __nf_conntrack_hash_insert(ct, hash, repl_hash);
361 /* Timer relative to confirmation time, not original
362 setting time, otherwise we'd get timer wrap in
363 weird delay cases. */
364 ct->timeout.expires += jiffies;
365 add_timer(&ct->timeout);
366 atomic_inc(&ct->ct_general.use);
367 set_bit(IPS_CONFIRMED_BIT, &ct->status);
368 NF_CT_STAT_INC(insert);
369 write_unlock_bh(&nf_conntrack_lock);
370 help = nfct_help(ct);
371 if (help && help->helper)
372 nf_conntrack_event_cache(IPCT_HELPER, *pskb);
373 #ifdef CONFIG_NF_NAT_NEEDED
374 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
375 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
376 nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
378 nf_conntrack_event_cache(master_ct(ct) ?
379 IPCT_RELATED : IPCT_NEW, *pskb);
383 NF_CT_STAT_INC(insert_failed);
384 write_unlock_bh(&nf_conntrack_lock);
387 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
389 /* Returns true if a connection correspondings to the tuple (required
392 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
393 const struct nf_conn *ignored_conntrack)
395 struct nf_conntrack_tuple_hash *h;
397 read_lock_bh(&nf_conntrack_lock);
398 h = __nf_conntrack_find(tuple, ignored_conntrack);
399 read_unlock_bh(&nf_conntrack_lock);
403 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
405 #define NF_CT_EVICTION_RANGE 8
407 /* There's a small race here where we may free a just-assured
408 connection. Too bad: we're in trouble anyway. */
409 static int early_drop(unsigned int hash)
411 /* Use oldest entry, which is roughly LRU */
412 struct nf_conntrack_tuple_hash *h;
413 struct nf_conn *ct = NULL, *tmp;
414 struct hlist_node *n;
415 unsigned int i, cnt = 0;
418 read_lock_bh(&nf_conntrack_lock);
419 for (i = 0; i < nf_conntrack_htable_size; i++) {
420 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
421 tmp = nf_ct_tuplehash_to_ctrack(h);
422 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
426 if (ct || cnt >= NF_CT_EVICTION_RANGE)
428 hash = (hash + 1) % nf_conntrack_htable_size;
431 atomic_inc(&ct->ct_general.use);
432 read_unlock_bh(&nf_conntrack_lock);
437 if (del_timer(&ct->timeout)) {
438 death_by_timeout((unsigned long)ct);
440 NF_CT_STAT_INC_ATOMIC(early_drop);
446 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
447 const struct nf_conntrack_tuple *repl)
449 struct nf_conn *conntrack = NULL;
451 if (unlikely(!nf_conntrack_hash_rnd_initted)) {
452 get_random_bytes(&nf_conntrack_hash_rnd, 4);
453 nf_conntrack_hash_rnd_initted = 1;
456 /* We don't want any race condition at early drop stage */
457 atomic_inc(&nf_conntrack_count);
460 && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
461 unsigned int hash = hash_conntrack(orig);
462 if (!early_drop(hash)) {
463 atomic_dec(&nf_conntrack_count);
466 "nf_conntrack: table full, dropping"
468 return ERR_PTR(-ENOMEM);
472 conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
473 if (conntrack == NULL) {
474 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
475 atomic_dec(&nf_conntrack_count);
476 return ERR_PTR(-ENOMEM);
479 atomic_set(&conntrack->ct_general.use, 1);
480 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
481 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
482 /* Don't set timer yet: wait for confirmation */
483 setup_timer(&conntrack->timeout, death_by_timeout,
484 (unsigned long)conntrack);
488 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
490 void nf_conntrack_free(struct nf_conn *conntrack)
492 nf_ct_ext_free(conntrack);
493 kmem_cache_free(nf_conntrack_cachep, conntrack);
494 atomic_dec(&nf_conntrack_count);
496 EXPORT_SYMBOL_GPL(nf_conntrack_free);
498 /* Allocate a new conntrack: we return -ENOMEM if classification
499 failed due to stress. Otherwise it really is unclassifiable. */
500 static struct nf_conntrack_tuple_hash *
501 init_conntrack(const struct nf_conntrack_tuple *tuple,
502 struct nf_conntrack_l3proto *l3proto,
503 struct nf_conntrack_l4proto *l4proto,
505 unsigned int dataoff)
507 struct nf_conn *conntrack;
508 struct nf_conn_help *help;
509 struct nf_conntrack_tuple repl_tuple;
510 struct nf_conntrack_expect *exp;
512 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
513 pr_debug("Can't invert tuple.\n");
517 conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
518 if (conntrack == NULL || IS_ERR(conntrack)) {
519 pr_debug("Can't allocate conntrack.\n");
520 return (struct nf_conntrack_tuple_hash *)conntrack;
523 if (!l4proto->new(conntrack, skb, dataoff)) {
524 nf_conntrack_free(conntrack);
525 pr_debug("init conntrack: can't track with proto module\n");
529 write_lock_bh(&nf_conntrack_lock);
530 exp = nf_ct_find_expectation(tuple);
532 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
534 /* Welcome, Mr. Bond. We've been expecting you... */
535 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
536 conntrack->master = exp->master;
538 help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
540 rcu_assign_pointer(help->helper, exp->helper);
543 #ifdef CONFIG_NF_CONNTRACK_MARK
544 conntrack->mark = exp->master->mark;
546 #ifdef CONFIG_NF_CONNTRACK_SECMARK
547 conntrack->secmark = exp->master->secmark;
549 nf_conntrack_get(&conntrack->master->ct_general);
550 NF_CT_STAT_INC(expect_new);
552 struct nf_conntrack_helper *helper;
554 helper = __nf_ct_helper_find(&repl_tuple);
556 help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
558 rcu_assign_pointer(help->helper, helper);
563 /* Overload tuple linked list to put us in unconfirmed list. */
564 hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
567 write_unlock_bh(&nf_conntrack_lock);
571 exp->expectfn(conntrack, exp);
572 nf_ct_expect_put(exp);
575 return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
578 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
579 static inline struct nf_conn *
580 resolve_normal_ct(struct sk_buff *skb,
581 unsigned int dataoff,
584 struct nf_conntrack_l3proto *l3proto,
585 struct nf_conntrack_l4proto *l4proto,
587 enum ip_conntrack_info *ctinfo)
589 struct nf_conntrack_tuple tuple;
590 struct nf_conntrack_tuple_hash *h;
593 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
594 dataoff, l3num, protonum, &tuple, l3proto,
596 pr_debug("resolve_normal_ct: Can't get tuple\n");
600 /* look for tuple match */
601 h = nf_conntrack_find_get(&tuple);
603 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
609 ct = nf_ct_tuplehash_to_ctrack(h);
611 /* It exists; we have (non-exclusive) reference. */
612 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
613 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
614 /* Please set reply bit if this packet OK */
617 /* Once we've had two way comms, always ESTABLISHED. */
618 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
619 pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
620 *ctinfo = IP_CT_ESTABLISHED;
621 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
622 pr_debug("nf_conntrack_in: related packet for %p\n",
624 *ctinfo = IP_CT_RELATED;
626 pr_debug("nf_conntrack_in: new packet for %p\n", ct);
631 skb->nfct = &ct->ct_general;
632 skb->nfctinfo = *ctinfo;
637 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
640 enum ip_conntrack_info ctinfo;
641 struct nf_conntrack_l3proto *l3proto;
642 struct nf_conntrack_l4proto *l4proto;
643 unsigned int dataoff;
648 /* Previously seen (loopback or untracked)? Ignore. */
650 NF_CT_STAT_INC_ATOMIC(ignore);
654 /* rcu_read_lock()ed by nf_hook_slow */
655 l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
656 ret = l3proto->get_l4proto(*pskb, skb_network_offset(*pskb),
657 &dataoff, &protonum);
659 pr_debug("not prepared to track yet or error occured\n");
660 NF_CT_STAT_INC_ATOMIC(error);
661 NF_CT_STAT_INC_ATOMIC(invalid);
665 l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
667 /* It may be an special packet, error, unclean...
668 * inverse of the return code tells to the netfilter
669 * core what to do with the packet. */
670 if (l4proto->error != NULL &&
671 (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
672 NF_CT_STAT_INC_ATOMIC(error);
673 NF_CT_STAT_INC_ATOMIC(invalid);
677 ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto,
678 &set_reply, &ctinfo);
680 /* Not valid part of a connection */
681 NF_CT_STAT_INC_ATOMIC(invalid);
686 /* Too stressed to deal. */
687 NF_CT_STAT_INC_ATOMIC(drop);
691 NF_CT_ASSERT((*pskb)->nfct);
693 ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum);
695 /* Invalid: inverse of the return code tells
696 * the netfilter core what to do */
697 pr_debug("nf_conntrack_in: Can't track with proto module\n");
698 nf_conntrack_put((*pskb)->nfct);
699 (*pskb)->nfct = NULL;
700 NF_CT_STAT_INC_ATOMIC(invalid);
704 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
705 nf_conntrack_event_cache(IPCT_STATUS, *pskb);
709 EXPORT_SYMBOL_GPL(nf_conntrack_in);
711 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
712 const struct nf_conntrack_tuple *orig)
717 ret = nf_ct_invert_tuple(inverse, orig,
718 __nf_ct_l3proto_find(orig->src.l3num),
719 __nf_ct_l4proto_find(orig->src.l3num,
720 orig->dst.protonum));
724 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
726 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
727 implicitly racy: see __nf_conntrack_confirm */
728 void nf_conntrack_alter_reply(struct nf_conn *ct,
729 const struct nf_conntrack_tuple *newreply)
731 struct nf_conn_help *help = nfct_help(ct);
732 struct nf_conntrack_helper *helper;
734 write_lock_bh(&nf_conntrack_lock);
735 /* Should be unconfirmed, so not in hash table yet */
736 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
738 pr_debug("Altering reply tuple of %p to ", ct);
739 NF_CT_DUMP_TUPLE(newreply);
741 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
742 if (ct->master || (help && help->expecting != 0))
745 helper = __nf_ct_helper_find(newreply);
746 if (helper == NULL) {
748 rcu_assign_pointer(help->helper, NULL);
753 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
757 memset(&help->help, 0, sizeof(help->help));
760 rcu_assign_pointer(help->helper, helper);
762 write_unlock_bh(&nf_conntrack_lock);
764 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
766 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
767 void __nf_ct_refresh_acct(struct nf_conn *ct,
768 enum ip_conntrack_info ctinfo,
769 const struct sk_buff *skb,
770 unsigned long extra_jiffies,
775 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
778 write_lock_bh(&nf_conntrack_lock);
780 /* Only update if this is not a fixed timeout */
781 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
782 write_unlock_bh(&nf_conntrack_lock);
786 /* If not in hash table, timer will not be active yet */
787 if (!nf_ct_is_confirmed(ct)) {
788 ct->timeout.expires = extra_jiffies;
789 event = IPCT_REFRESH;
791 unsigned long newtime = jiffies + extra_jiffies;
793 /* Only update the timeout if the new timeout is at least
794 HZ jiffies from the old timeout. Need del_timer for race
795 avoidance (may already be dying). */
796 if (newtime - ct->timeout.expires >= HZ
797 && del_timer(&ct->timeout)) {
798 ct->timeout.expires = newtime;
799 add_timer(&ct->timeout);
800 event = IPCT_REFRESH;
804 #ifdef CONFIG_NF_CT_ACCT
806 ct->counters[CTINFO2DIR(ctinfo)].packets++;
807 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
808 skb->len - skb_network_offset(skb);
810 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
811 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
812 event |= IPCT_COUNTER_FILLING;
816 write_unlock_bh(&nf_conntrack_lock);
818 /* must be unlocked when calling event cache */
820 nf_conntrack_event_cache(event, skb);
822 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
824 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
826 #include <linux/netfilter/nfnetlink.h>
827 #include <linux/netfilter/nfnetlink_conntrack.h>
828 #include <linux/mutex.h>
830 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
831 * in ip_conntrack_core, since we don't want the protocols to autoload
832 * or depend on ctnetlink */
833 int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
834 const struct nf_conntrack_tuple *tuple)
836 NLA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
837 &tuple->src.u.tcp.port);
838 NLA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
839 &tuple->dst.u.tcp.port);
845 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr);
847 static const size_t cta_min_proto[CTA_PROTO_MAX+1] = {
848 [CTA_PROTO_SRC_PORT] = sizeof(u_int16_t),
849 [CTA_PROTO_DST_PORT] = sizeof(u_int16_t)
852 int nf_ct_port_nfattr_to_tuple(struct nlattr *tb[],
853 struct nf_conntrack_tuple *t)
855 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
858 if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
861 t->src.u.tcp.port = *(__be16 *)nla_data(tb[CTA_PROTO_SRC_PORT]);
862 t->dst.u.tcp.port = *(__be16 *)nla_data(tb[CTA_PROTO_DST_PORT]);
866 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple);
869 /* Used by ipt_REJECT and ip6t_REJECT. */
870 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
873 enum ip_conntrack_info ctinfo;
875 /* This ICMP is in reverse direction to the packet which caused it */
876 ct = nf_ct_get(skb, &ctinfo);
877 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
878 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
880 ctinfo = IP_CT_RELATED;
882 /* Attach to new skbuff, and increment count */
883 nskb->nfct = &ct->ct_general;
884 nskb->nfctinfo = ctinfo;
885 nf_conntrack_get(nskb->nfct);
887 EXPORT_SYMBOL_GPL(__nf_conntrack_attach);
890 do_iter(const struct nf_conntrack_tuple_hash *i,
891 int (*iter)(struct nf_conn *i, void *data),
894 return iter(nf_ct_tuplehash_to_ctrack(i), data);
897 /* Bring out ya dead! */
898 static struct nf_conn *
899 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
900 void *data, unsigned int *bucket)
902 struct nf_conntrack_tuple_hash *h;
904 struct hlist_node *n;
906 write_lock_bh(&nf_conntrack_lock);
907 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
908 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
909 ct = nf_ct_tuplehash_to_ctrack(h);
914 hlist_for_each_entry(h, n, &unconfirmed, hnode) {
915 ct = nf_ct_tuplehash_to_ctrack(h);
917 set_bit(IPS_DYING_BIT, &ct->status);
919 write_unlock_bh(&nf_conntrack_lock);
922 atomic_inc(&ct->ct_general.use);
923 write_unlock_bh(&nf_conntrack_lock);
928 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
931 unsigned int bucket = 0;
933 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
934 /* Time to push up daises... */
935 if (del_timer(&ct->timeout))
936 death_by_timeout((unsigned long)ct);
937 /* ... else the timer will get him soon. */
942 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
944 static int kill_all(struct nf_conn *i, void *data)
949 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, int size)
954 free_pages((unsigned long)hash,
955 get_order(sizeof(struct hlist_head) * size));
957 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
959 void nf_conntrack_flush(void)
961 nf_ct_iterate_cleanup(kill_all, NULL);
963 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
965 /* Mishearing the voices in his head, our hero wonders how he's
966 supposed to kill the mall. */
967 void nf_conntrack_cleanup(void)
969 rcu_assign_pointer(ip_ct_attach, NULL);
971 /* This makes sure all current packets have passed through
972 netfilter framework. Roll on, two-stage module
976 nf_ct_event_cache_flush();
978 nf_conntrack_flush();
979 if (atomic_read(&nf_conntrack_count) != 0) {
981 goto i_see_dead_people;
983 /* wait until all references to nf_conntrack_untracked are dropped */
984 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
987 rcu_assign_pointer(nf_ct_destroy, NULL);
989 kmem_cache_destroy(nf_conntrack_cachep);
990 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
991 nf_conntrack_htable_size);
993 nf_conntrack_proto_fini();
994 nf_conntrack_helper_fini();
995 nf_conntrack_expect_fini();
998 struct hlist_head *nf_ct_alloc_hashtable(int *sizep, int *vmalloced)
1000 struct hlist_head *hash;
1001 unsigned int size, i;
1005 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1006 hash = (void*)__get_free_pages(GFP_KERNEL,
1007 get_order(sizeof(struct hlist_head)
1011 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1012 hash = vmalloc(sizeof(struct hlist_head) * size);
1016 for (i = 0; i < size; i++)
1017 INIT_HLIST_HEAD(&hash[i]);
1021 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1023 int set_hashsize(const char *val, struct kernel_param *kp)
1025 int i, bucket, hashsize, vmalloced;
1026 int old_vmalloced, old_size;
1028 struct hlist_head *hash, *old_hash;
1029 struct nf_conntrack_tuple_hash *h;
1031 /* On boot, we can set this without any fancy locking. */
1032 if (!nf_conntrack_htable_size)
1033 return param_set_uint(val, kp);
1035 hashsize = simple_strtol(val, NULL, 0);
1039 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1043 /* We have to rehahs for the new table anyway, so we also can
1044 * use a newrandom seed */
1045 get_random_bytes(&rnd, 4);
1047 write_lock_bh(&nf_conntrack_lock);
1048 for (i = 0; i < nf_conntrack_htable_size; i++) {
1049 while (!hlist_empty(&nf_conntrack_hash[i])) {
1050 h = hlist_entry(nf_conntrack_hash[i].first,
1051 struct nf_conntrack_tuple_hash, hnode);
1052 hlist_del(&h->hnode);
1053 bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1054 hlist_add_head(&h->hnode, &hash[bucket]);
1057 old_size = nf_conntrack_htable_size;
1058 old_vmalloced = nf_conntrack_vmalloc;
1059 old_hash = nf_conntrack_hash;
1061 nf_conntrack_htable_size = hashsize;
1062 nf_conntrack_vmalloc = vmalloced;
1063 nf_conntrack_hash = hash;
1064 nf_conntrack_hash_rnd = rnd;
1065 write_unlock_bh(&nf_conntrack_lock);
1067 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1071 module_param_call(hashsize, set_hashsize, param_get_uint,
1072 &nf_conntrack_htable_size, 0600);
1074 int __init nf_conntrack_init(void)
1079 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1080 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1081 if (!nf_conntrack_htable_size) {
1082 nf_conntrack_htable_size
1083 = (((num_physpages << PAGE_SHIFT) / 16384)
1084 / sizeof(struct hlist_head));
1085 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1086 nf_conntrack_htable_size = 16384;
1087 if (nf_conntrack_htable_size < 32)
1088 nf_conntrack_htable_size = 32;
1090 /* Use a max. factor of four by default to get the same max as
1091 * with the old struct list_heads. When a table size is given
1092 * we use the old value of 8 to avoid reducing the max.
1096 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1097 &nf_conntrack_vmalloc);
1098 if (!nf_conntrack_hash) {
1099 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1103 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1105 printk("nf_conntrack version %s (%u buckets, %d max)\n",
1106 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1109 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1110 sizeof(struct nf_conn),
1112 if (!nf_conntrack_cachep) {
1113 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1117 ret = nf_conntrack_proto_init();
1119 goto err_free_conntrack_slab;
1121 ret = nf_conntrack_expect_init();
1123 goto out_fini_proto;
1125 ret = nf_conntrack_helper_init();
1127 goto out_fini_expect;
1129 /* For use by REJECT target */
1130 rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
1131 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1133 /* Set up fake conntrack:
1134 - to never be deleted, not in any hashes */
1135 atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1136 /* - and look it like as a confirmed connection */
1137 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1142 nf_conntrack_expect_fini();
1144 nf_conntrack_proto_fini();
1145 err_free_conntrack_slab:
1146 kmem_cache_destroy(nf_conntrack_cachep);
1148 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1149 nf_conntrack_htable_size);