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) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
12 * 23 Apr 2001: Harald Welte <laforge@gnumonks.org>
13 * - new API and handling of conntrack/nat helpers
14 * - now capable of multiple expectations for one master
15 * 16 Jul 2002: Harald Welte <laforge@gnumonks.org>
16 * - add usage/reference counts to ip_conntrack_expect
17 * - export ip_conntrack[_expect]_{find_get,put} functions
20 #include <linux/config.h>
21 #include <linux/types.h>
22 #include <linux/icmp.h>
24 #include <linux/netfilter.h>
25 #include <linux/netfilter_ipv4.h>
26 #include <linux/module.h>
27 #include <linux/skbuff.h>
28 #include <linux/proc_fs.h>
29 #include <linux/vmalloc.h>
30 #include <net/checksum.h>
32 #include <linux/stddef.h>
33 #include <linux/sysctl.h>
34 #include <linux/slab.h>
35 #include <linux/random.h>
36 #include <linux/jhash.h>
37 #include <linux/err.h>
38 #include <linux/percpu.h>
39 #include <linux/moduleparam.h>
41 /* This rwlock protects the main hash table, protocol/helper/expected
42 registrations, conntrack timers*/
43 #define ASSERT_READ_LOCK(x) MUST_BE_READ_LOCKED(&ip_conntrack_lock)
44 #define ASSERT_WRITE_LOCK(x) MUST_BE_WRITE_LOCKED(&ip_conntrack_lock)
46 #include <linux/netfilter_ipv4/ip_conntrack.h>
47 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
48 #include <linux/netfilter_ipv4/ip_conntrack_helper.h>
49 #include <linux/netfilter_ipv4/ip_conntrack_core.h>
50 #include <linux/netfilter_ipv4/listhelp.h>
52 #define IP_CONNTRACK_VERSION "2.1"
57 #define DEBUGP(format, args...)
60 DECLARE_RWLOCK(ip_conntrack_lock);
62 /* ip_conntrack_standalone needs this */
63 atomic_t ip_conntrack_count = ATOMIC_INIT(0);
65 void (*ip_conntrack_destroyed)(struct ip_conntrack *conntrack) = NULL;
66 LIST_HEAD(ip_conntrack_expect_list);
67 struct ip_conntrack_protocol *ip_ct_protos[MAX_IP_CT_PROTO];
68 static LIST_HEAD(helpers);
69 unsigned int ip_conntrack_htable_size = 0;
71 struct list_head *ip_conntrack_hash;
72 static kmem_cache_t *ip_conntrack_cachep;
73 static kmem_cache_t *ip_conntrack_expect_cachep;
74 struct ip_conntrack ip_conntrack_untracked;
75 unsigned int ip_ct_log_invalid;
76 static LIST_HEAD(unconfirmed);
77 static int ip_conntrack_vmalloc;
79 DEFINE_PER_CPU(struct ip_conntrack_stat, ip_conntrack_stat);
82 ip_conntrack_put(struct ip_conntrack *ct)
85 nf_conntrack_put(&ct->ct_general);
88 static int ip_conntrack_hash_rnd_initted;
89 static unsigned int ip_conntrack_hash_rnd;
92 hash_conntrack(const struct ip_conntrack_tuple *tuple)
97 return (jhash_3words(tuple->src.ip,
98 (tuple->dst.ip ^ tuple->dst.protonum),
99 (tuple->src.u.all | (tuple->dst.u.all << 16)),
100 ip_conntrack_hash_rnd) % ip_conntrack_htable_size);
104 ip_ct_get_tuple(const struct iphdr *iph,
105 const struct sk_buff *skb,
106 unsigned int dataoff,
107 struct ip_conntrack_tuple *tuple,
108 const struct ip_conntrack_protocol *protocol)
111 if (iph->frag_off & htons(IP_OFFSET)) {
112 printk("ip_conntrack_core: Frag of proto %u.\n",
117 tuple->src.ip = iph->saddr;
118 tuple->dst.ip = iph->daddr;
119 tuple->dst.protonum = iph->protocol;
120 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
122 return protocol->pkt_to_tuple(skb, dataoff, tuple);
126 ip_ct_invert_tuple(struct ip_conntrack_tuple *inverse,
127 const struct ip_conntrack_tuple *orig,
128 const struct ip_conntrack_protocol *protocol)
130 inverse->src.ip = orig->dst.ip;
131 inverse->dst.ip = orig->src.ip;
132 inverse->dst.protonum = orig->dst.protonum;
133 inverse->dst.dir = !orig->dst.dir;
135 return protocol->invert_tuple(inverse, orig);
139 /* ip_conntrack_expect helper functions */
140 static void destroy_expect(struct ip_conntrack_expect *exp)
142 ip_conntrack_put(exp->master);
143 IP_NF_ASSERT(!timer_pending(&exp->timeout));
144 kmem_cache_free(ip_conntrack_expect_cachep, exp);
145 CONNTRACK_STAT_INC(expect_delete);
148 static void unlink_expect(struct ip_conntrack_expect *exp)
150 MUST_BE_WRITE_LOCKED(&ip_conntrack_lock);
151 list_del(&exp->list);
152 /* Logically in destroy_expect, but we hold the lock here. */
153 exp->master->expecting--;
156 static void expectation_timed_out(unsigned long ul_expect)
158 struct ip_conntrack_expect *exp = (void *)ul_expect;
160 WRITE_LOCK(&ip_conntrack_lock);
162 WRITE_UNLOCK(&ip_conntrack_lock);
166 /* If an expectation for this connection is found, it gets delete from
167 * global list then returned. */
168 static struct ip_conntrack_expect *
169 find_expectation(const struct ip_conntrack_tuple *tuple)
171 struct ip_conntrack_expect *i;
173 list_for_each_entry(i, &ip_conntrack_expect_list, list) {
174 /* If master is not in hash table yet (ie. packet hasn't left
175 this machine yet), how can other end know about expected?
176 Hence these are not the droids you are looking for (if
177 master ct never got confirmed, we'd hold a reference to it
178 and weird things would happen to future packets). */
179 if (ip_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask)
180 && is_confirmed(i->master)
181 && del_timer(&i->timeout)) {
189 /* delete all expectations for this conntrack */
190 static void remove_expectations(struct ip_conntrack *ct)
192 struct ip_conntrack_expect *i, *tmp;
194 /* Optimization: most connection never expect any others. */
195 if (ct->expecting == 0)
198 list_for_each_entry_safe(i, tmp, &ip_conntrack_expect_list, list) {
199 if (i->master == ct && del_timer(&i->timeout)) {
207 clean_from_lists(struct ip_conntrack *ct)
211 DEBUGP("clean_from_lists(%p)\n", ct);
212 MUST_BE_WRITE_LOCKED(&ip_conntrack_lock);
214 ho = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
215 hr = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
216 LIST_DELETE(&ip_conntrack_hash[ho], &ct->tuplehash[IP_CT_DIR_ORIGINAL]);
217 LIST_DELETE(&ip_conntrack_hash[hr], &ct->tuplehash[IP_CT_DIR_REPLY]);
219 /* Destroy all pending expectations */
220 remove_expectations(ct);
224 destroy_conntrack(struct nf_conntrack *nfct)
226 struct ip_conntrack *ct = (struct ip_conntrack *)nfct;
227 struct ip_conntrack_protocol *proto;
229 DEBUGP("destroy_conntrack(%p)\n", ct);
230 IP_NF_ASSERT(atomic_read(&nfct->use) == 0);
231 IP_NF_ASSERT(!timer_pending(&ct->timeout));
233 /* To make sure we don't get any weird locking issues here:
234 * destroy_conntrack() MUST NOT be called with a write lock
235 * to ip_conntrack_lock!!! -HW */
236 proto = ip_ct_find_proto(ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
237 if (proto && proto->destroy)
240 if (ip_conntrack_destroyed)
241 ip_conntrack_destroyed(ct);
243 WRITE_LOCK(&ip_conntrack_lock);
244 /* Expectations will have been removed in clean_from_lists,
245 * except TFTP can create an expectation on the first packet,
246 * before connection is in the list, so we need to clean here,
248 remove_expectations(ct);
250 /* We overload first tuple to link into unconfirmed list. */
251 if (!is_confirmed(ct)) {
252 BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list));
253 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
256 CONNTRACK_STAT_INC(delete);
257 WRITE_UNLOCK(&ip_conntrack_lock);
260 ip_conntrack_put(ct->master);
262 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct);
263 kmem_cache_free(ip_conntrack_cachep, ct);
264 atomic_dec(&ip_conntrack_count);
267 static void death_by_timeout(unsigned long ul_conntrack)
269 struct ip_conntrack *ct = (void *)ul_conntrack;
271 WRITE_LOCK(&ip_conntrack_lock);
272 /* Inside lock so preempt is disabled on module removal path.
273 * Otherwise we can get spurious warnings. */
274 CONNTRACK_STAT_INC(delete_list);
275 clean_from_lists(ct);
276 WRITE_UNLOCK(&ip_conntrack_lock);
277 ip_conntrack_put(ct);
281 conntrack_tuple_cmp(const struct ip_conntrack_tuple_hash *i,
282 const struct ip_conntrack_tuple *tuple,
283 const struct ip_conntrack *ignored_conntrack)
285 MUST_BE_READ_LOCKED(&ip_conntrack_lock);
286 return tuplehash_to_ctrack(i) != ignored_conntrack
287 && ip_ct_tuple_equal(tuple, &i->tuple);
290 static struct ip_conntrack_tuple_hash *
291 __ip_conntrack_find(const struct ip_conntrack_tuple *tuple,
292 const struct ip_conntrack *ignored_conntrack)
294 struct ip_conntrack_tuple_hash *h;
295 unsigned int hash = hash_conntrack(tuple);
297 MUST_BE_READ_LOCKED(&ip_conntrack_lock);
298 list_for_each_entry(h, &ip_conntrack_hash[hash], list) {
299 if (conntrack_tuple_cmp(h, tuple, ignored_conntrack)) {
300 CONNTRACK_STAT_INC(found);
303 CONNTRACK_STAT_INC(searched);
309 /* Find a connection corresponding to a tuple. */
310 struct ip_conntrack_tuple_hash *
311 ip_conntrack_find_get(const struct ip_conntrack_tuple *tuple,
312 const struct ip_conntrack *ignored_conntrack)
314 struct ip_conntrack_tuple_hash *h;
316 READ_LOCK(&ip_conntrack_lock);
317 h = __ip_conntrack_find(tuple, ignored_conntrack);
319 atomic_inc(&tuplehash_to_ctrack(h)->ct_general.use);
320 READ_UNLOCK(&ip_conntrack_lock);
325 /* Confirm a connection given skb; places it in hash table */
327 __ip_conntrack_confirm(struct sk_buff **pskb)
329 unsigned int hash, repl_hash;
330 struct ip_conntrack *ct;
331 enum ip_conntrack_info ctinfo;
333 ct = ip_conntrack_get(*pskb, &ctinfo);
335 /* ipt_REJECT uses ip_conntrack_attach to attach related
336 ICMP/TCP RST packets in other direction. Actual packet
337 which created connection will be IP_CT_NEW or for an
338 expected connection, IP_CT_RELATED. */
339 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
342 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
343 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
345 /* We're not in hash table, and we refuse to set up related
346 connections for unconfirmed conns. But packet copies and
347 REJECT will give spurious warnings here. */
348 /* IP_NF_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
350 /* No external references means noone else could have
352 IP_NF_ASSERT(!is_confirmed(ct));
353 DEBUGP("Confirming conntrack %p\n", ct);
355 WRITE_LOCK(&ip_conntrack_lock);
357 /* See if there's one in the list already, including reverse:
358 NAT could have grabbed it without realizing, since we're
359 not in the hash. If there is, we lost race. */
360 if (!LIST_FIND(&ip_conntrack_hash[hash],
362 struct ip_conntrack_tuple_hash *,
363 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, NULL)
364 && !LIST_FIND(&ip_conntrack_hash[repl_hash],
366 struct ip_conntrack_tuple_hash *,
367 &ct->tuplehash[IP_CT_DIR_REPLY].tuple, NULL)) {
368 /* Remove from unconfirmed list */
369 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
371 list_prepend(&ip_conntrack_hash[hash],
372 &ct->tuplehash[IP_CT_DIR_ORIGINAL]);
373 list_prepend(&ip_conntrack_hash[repl_hash],
374 &ct->tuplehash[IP_CT_DIR_REPLY]);
375 /* Timer relative to confirmation time, not original
376 setting time, otherwise we'd get timer wrap in
377 weird delay cases. */
378 ct->timeout.expires += jiffies;
379 add_timer(&ct->timeout);
380 atomic_inc(&ct->ct_general.use);
381 set_bit(IPS_CONFIRMED_BIT, &ct->status);
382 CONNTRACK_STAT_INC(insert);
383 WRITE_UNLOCK(&ip_conntrack_lock);
387 CONNTRACK_STAT_INC(insert_failed);
388 WRITE_UNLOCK(&ip_conntrack_lock);
393 /* Returns true if a connection correspondings to the tuple (required
396 ip_conntrack_tuple_taken(const struct ip_conntrack_tuple *tuple,
397 const struct ip_conntrack *ignored_conntrack)
399 struct ip_conntrack_tuple_hash *h;
401 READ_LOCK(&ip_conntrack_lock);
402 h = __ip_conntrack_find(tuple, ignored_conntrack);
403 READ_UNLOCK(&ip_conntrack_lock);
408 /* There's a small race here where we may free a just-assured
409 connection. Too bad: we're in trouble anyway. */
410 static inline int unreplied(const struct ip_conntrack_tuple_hash *i)
412 return !(test_bit(IPS_ASSURED_BIT, &tuplehash_to_ctrack(i)->status));
415 static int early_drop(struct list_head *chain)
417 /* Traverse backwards: gives us oldest, which is roughly LRU */
418 struct ip_conntrack_tuple_hash *h;
419 struct ip_conntrack *ct = NULL;
422 READ_LOCK(&ip_conntrack_lock);
423 h = LIST_FIND_B(chain, unreplied, struct ip_conntrack_tuple_hash *);
425 ct = tuplehash_to_ctrack(h);
426 atomic_inc(&ct->ct_general.use);
428 READ_UNLOCK(&ip_conntrack_lock);
433 if (del_timer(&ct->timeout)) {
434 death_by_timeout((unsigned long)ct);
436 CONNTRACK_STAT_INC(early_drop);
438 ip_conntrack_put(ct);
442 static inline int helper_cmp(const struct ip_conntrack_helper *i,
443 const struct ip_conntrack_tuple *rtuple)
445 return ip_ct_tuple_mask_cmp(rtuple, &i->tuple, &i->mask);
448 static struct ip_conntrack_helper *ip_ct_find_helper(const struct ip_conntrack_tuple *tuple)
450 return LIST_FIND(&helpers, helper_cmp,
451 struct ip_conntrack_helper *,
455 /* Allocate a new conntrack: we return -ENOMEM if classification
456 failed due to stress. Otherwise it really is unclassifiable. */
457 static struct ip_conntrack_tuple_hash *
458 init_conntrack(const struct ip_conntrack_tuple *tuple,
459 struct ip_conntrack_protocol *protocol,
462 struct ip_conntrack *conntrack;
463 struct ip_conntrack_tuple repl_tuple;
465 struct ip_conntrack_expect *exp;
467 if (!ip_conntrack_hash_rnd_initted) {
468 get_random_bytes(&ip_conntrack_hash_rnd, 4);
469 ip_conntrack_hash_rnd_initted = 1;
472 hash = hash_conntrack(tuple);
475 && atomic_read(&ip_conntrack_count) >= ip_conntrack_max) {
476 /* Try dropping from this hash chain. */
477 if (!early_drop(&ip_conntrack_hash[hash])) {
480 "ip_conntrack: table full, dropping"
482 return ERR_PTR(-ENOMEM);
486 if (!ip_ct_invert_tuple(&repl_tuple, tuple, protocol)) {
487 DEBUGP("Can't invert tuple.\n");
491 conntrack = kmem_cache_alloc(ip_conntrack_cachep, GFP_ATOMIC);
493 DEBUGP("Can't allocate conntrack.\n");
494 return ERR_PTR(-ENOMEM);
497 memset(conntrack, 0, sizeof(*conntrack));
498 atomic_set(&conntrack->ct_general.use, 1);
499 conntrack->ct_general.destroy = destroy_conntrack;
500 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *tuple;
501 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = repl_tuple;
502 if (!protocol->new(conntrack, skb)) {
503 kmem_cache_free(ip_conntrack_cachep, conntrack);
506 /* Don't set timer yet: wait for confirmation */
507 init_timer(&conntrack->timeout);
508 conntrack->timeout.data = (unsigned long)conntrack;
509 conntrack->timeout.function = death_by_timeout;
511 WRITE_LOCK(&ip_conntrack_lock);
512 exp = find_expectation(tuple);
515 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
517 /* Welcome, Mr. Bond. We've been expecting you... */
518 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
519 conntrack->master = exp->master;
520 #if CONFIG_IP_NF_CONNTRACK_MARK
521 conntrack->mark = exp->master->mark;
523 nf_conntrack_get(&conntrack->master->ct_general);
524 CONNTRACK_STAT_INC(expect_new);
526 conntrack->helper = ip_ct_find_helper(&repl_tuple);
528 CONNTRACK_STAT_INC(new);
531 /* Overload tuple linked list to put us in unconfirmed list. */
532 list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed);
534 atomic_inc(&ip_conntrack_count);
535 WRITE_UNLOCK(&ip_conntrack_lock);
539 exp->expectfn(conntrack, exp);
543 return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
546 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
547 static inline struct ip_conntrack *
548 resolve_normal_ct(struct sk_buff *skb,
549 struct ip_conntrack_protocol *proto,
551 unsigned int hooknum,
552 enum ip_conntrack_info *ctinfo)
554 struct ip_conntrack_tuple tuple;
555 struct ip_conntrack_tuple_hash *h;
556 struct ip_conntrack *ct;
558 IP_NF_ASSERT((skb->nh.iph->frag_off & htons(IP_OFFSET)) == 0);
560 if (!ip_ct_get_tuple(skb->nh.iph, skb, skb->nh.iph->ihl*4,
564 /* look for tuple match */
565 h = ip_conntrack_find_get(&tuple, NULL);
567 h = init_conntrack(&tuple, proto, skb);
573 ct = tuplehash_to_ctrack(h);
575 /* It exists; we have (non-exclusive) reference. */
576 if (DIRECTION(h) == IP_CT_DIR_REPLY) {
577 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
578 /* Please set reply bit if this packet OK */
581 /* Once we've had two way comms, always ESTABLISHED. */
582 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
583 DEBUGP("ip_conntrack_in: normal packet for %p\n",
585 *ctinfo = IP_CT_ESTABLISHED;
586 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
587 DEBUGP("ip_conntrack_in: related packet for %p\n",
589 *ctinfo = IP_CT_RELATED;
591 DEBUGP("ip_conntrack_in: new packet for %p\n",
597 skb->nfct = &ct->ct_general;
598 skb->nfctinfo = *ctinfo;
602 /* Netfilter hook itself. */
603 unsigned int ip_conntrack_in(unsigned int hooknum,
604 struct sk_buff **pskb,
605 const struct net_device *in,
606 const struct net_device *out,
607 int (*okfn)(struct sk_buff *))
609 struct ip_conntrack *ct;
610 enum ip_conntrack_info ctinfo;
611 struct ip_conntrack_protocol *proto;
615 /* Previously seen (loopback or untracked)? Ignore. */
617 CONNTRACK_STAT_INC(ignore);
622 if ((*pskb)->nh.iph->frag_off & htons(IP_OFFSET)) {
623 if (net_ratelimit()) {
624 printk(KERN_ERR "ip_conntrack_in: Frag of proto %u (hook=%u)\n",
625 (*pskb)->nh.iph->protocol, hooknum);
630 /* FIXME: Do this right please. --RR */
631 (*pskb)->nfcache |= NFC_UNKNOWN;
633 /* Doesn't cover locally-generated broadcast, so not worth it. */
635 /* Ignore broadcast: no `connection'. */
636 if ((*pskb)->pkt_type == PACKET_BROADCAST) {
637 printk("Broadcast packet!\n");
639 } else if (((*pskb)->nh.iph->daddr & htonl(0x000000FF))
640 == htonl(0x000000FF)) {
641 printk("Should bcast: %u.%u.%u.%u->%u.%u.%u.%u (sk=%p, ptype=%u)\n",
642 NIPQUAD((*pskb)->nh.iph->saddr),
643 NIPQUAD((*pskb)->nh.iph->daddr),
644 (*pskb)->sk, (*pskb)->pkt_type);
648 proto = ip_ct_find_proto((*pskb)->nh.iph->protocol);
650 /* It may be an special packet, error, unclean...
651 * inverse of the return code tells to the netfilter
652 * core what to do with the packet. */
653 if (proto->error != NULL
654 && (ret = proto->error(*pskb, &ctinfo, hooknum)) <= 0) {
655 CONNTRACK_STAT_INC(error);
656 CONNTRACK_STAT_INC(invalid);
660 if (!(ct = resolve_normal_ct(*pskb, proto,&set_reply,hooknum,&ctinfo))) {
661 /* Not valid part of a connection */
662 CONNTRACK_STAT_INC(invalid);
667 /* Too stressed to deal. */
668 CONNTRACK_STAT_INC(drop);
672 IP_NF_ASSERT((*pskb)->nfct);
674 ret = proto->packet(ct, *pskb, ctinfo);
676 /* Invalid: inverse of the return code tells
677 * the netfilter core what to do*/
678 nf_conntrack_put((*pskb)->nfct);
679 (*pskb)->nfct = NULL;
680 CONNTRACK_STAT_INC(invalid);
685 set_bit(IPS_SEEN_REPLY_BIT, &ct->status);
690 int invert_tuplepr(struct ip_conntrack_tuple *inverse,
691 const struct ip_conntrack_tuple *orig)
693 return ip_ct_invert_tuple(inverse, orig,
694 ip_ct_find_proto(orig->dst.protonum));
697 /* Would two expected things clash? */
698 static inline int expect_clash(const struct ip_conntrack_expect *a,
699 const struct ip_conntrack_expect *b)
701 /* Part covered by intersection of masks must be unequal,
702 otherwise they clash */
703 struct ip_conntrack_tuple intersect_mask
704 = { { a->mask.src.ip & b->mask.src.ip,
705 { a->mask.src.u.all & b->mask.src.u.all } },
706 { a->mask.dst.ip & b->mask.dst.ip,
707 { a->mask.dst.u.all & b->mask.dst.u.all },
708 a->mask.dst.protonum & b->mask.dst.protonum } };
710 return ip_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask);
713 static inline int expect_matches(const struct ip_conntrack_expect *a,
714 const struct ip_conntrack_expect *b)
716 return a->master == b->master
717 && ip_ct_tuple_equal(&a->tuple, &b->tuple)
718 && ip_ct_tuple_equal(&a->mask, &b->mask);
721 /* Generally a bad idea to call this: could have matched already. */
722 void ip_conntrack_unexpect_related(struct ip_conntrack_expect *exp)
724 struct ip_conntrack_expect *i;
726 WRITE_LOCK(&ip_conntrack_lock);
727 /* choose the the oldest expectation to evict */
728 list_for_each_entry_reverse(i, &ip_conntrack_expect_list, list) {
729 if (expect_matches(i, exp) && del_timer(&i->timeout)) {
731 WRITE_UNLOCK(&ip_conntrack_lock);
736 WRITE_UNLOCK(&ip_conntrack_lock);
739 struct ip_conntrack_expect *ip_conntrack_expect_alloc(void)
741 struct ip_conntrack_expect *new;
743 new = kmem_cache_alloc(ip_conntrack_expect_cachep, GFP_ATOMIC);
745 DEBUGP("expect_related: OOM allocating expect\n");
752 void ip_conntrack_expect_free(struct ip_conntrack_expect *expect)
754 kmem_cache_free(ip_conntrack_expect_cachep, expect);
757 static void ip_conntrack_expect_insert(struct ip_conntrack_expect *exp)
759 atomic_inc(&exp->master->ct_general.use);
760 exp->master->expecting++;
761 list_add(&exp->list, &ip_conntrack_expect_list);
763 if (exp->master->helper->timeout) {
764 init_timer(&exp->timeout);
765 exp->timeout.data = (unsigned long)exp;
766 exp->timeout.function = expectation_timed_out;
768 = jiffies + exp->master->helper->timeout * HZ;
769 add_timer(&exp->timeout);
771 exp->timeout.function = NULL;
773 CONNTRACK_STAT_INC(expect_create);
776 /* Race with expectations being used means we could have none to find; OK. */
777 static void evict_oldest_expect(struct ip_conntrack *master)
779 struct ip_conntrack_expect *i;
781 list_for_each_entry_reverse(i, &ip_conntrack_expect_list, list) {
782 if (i->master == master) {
783 if (del_timer(&i->timeout)) {
792 static inline int refresh_timer(struct ip_conntrack_expect *i)
794 if (!del_timer(&i->timeout))
797 i->timeout.expires = jiffies + i->master->helper->timeout*HZ;
798 add_timer(&i->timeout);
802 int ip_conntrack_expect_related(struct ip_conntrack_expect *expect)
804 struct ip_conntrack_expect *i;
807 DEBUGP("ip_conntrack_expect_related %p\n", related_to);
808 DEBUGP("tuple: "); DUMP_TUPLE(&expect->tuple);
809 DEBUGP("mask: "); DUMP_TUPLE(&expect->mask);
811 WRITE_LOCK(&ip_conntrack_lock);
812 list_for_each_entry(i, &ip_conntrack_expect_list, list) {
813 if (expect_matches(i, expect)) {
814 /* Refresh timer: if it's dying, ignore.. */
815 if (refresh_timer(i)) {
817 /* We don't need the one they've given us. */
818 ip_conntrack_expect_free(expect);
821 } else if (expect_clash(i, expect)) {
827 /* Will be over limit? */
828 if (expect->master->helper->max_expected &&
829 expect->master->expecting >= expect->master->helper->max_expected)
830 evict_oldest_expect(expect->master);
832 ip_conntrack_expect_insert(expect);
835 WRITE_UNLOCK(&ip_conntrack_lock);
839 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
840 implicitly racy: see __ip_conntrack_confirm */
841 void ip_conntrack_alter_reply(struct ip_conntrack *conntrack,
842 const struct ip_conntrack_tuple *newreply)
844 WRITE_LOCK(&ip_conntrack_lock);
845 /* Should be unconfirmed, so not in hash table yet */
846 IP_NF_ASSERT(!is_confirmed(conntrack));
848 DEBUGP("Altering reply tuple of %p to ", conntrack);
849 DUMP_TUPLE(newreply);
851 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
852 if (!conntrack->master && conntrack->expecting == 0)
853 conntrack->helper = ip_ct_find_helper(newreply);
854 WRITE_UNLOCK(&ip_conntrack_lock);
857 int ip_conntrack_helper_register(struct ip_conntrack_helper *me)
859 BUG_ON(me->timeout == 0);
860 WRITE_LOCK(&ip_conntrack_lock);
861 list_prepend(&helpers, me);
862 WRITE_UNLOCK(&ip_conntrack_lock);
867 static inline int unhelp(struct ip_conntrack_tuple_hash *i,
868 const struct ip_conntrack_helper *me)
870 if (tuplehash_to_ctrack(i)->helper == me)
871 tuplehash_to_ctrack(i)->helper = NULL;
875 void ip_conntrack_helper_unregister(struct ip_conntrack_helper *me)
878 struct ip_conntrack_expect *exp, *tmp;
880 /* Need write lock here, to delete helper. */
881 WRITE_LOCK(&ip_conntrack_lock);
882 LIST_DELETE(&helpers, me);
884 /* Get rid of expectations */
885 list_for_each_entry_safe(exp, tmp, &ip_conntrack_expect_list, list) {
886 if (exp->master->helper == me && del_timer(&exp->timeout)) {
891 /* Get rid of expecteds, set helpers to NULL. */
892 LIST_FIND_W(&unconfirmed, unhelp, struct ip_conntrack_tuple_hash*, me);
893 for (i = 0; i < ip_conntrack_htable_size; i++)
894 LIST_FIND_W(&ip_conntrack_hash[i], unhelp,
895 struct ip_conntrack_tuple_hash *, me);
896 WRITE_UNLOCK(&ip_conntrack_lock);
898 /* Someone could be still looking at the helper in a bh. */
902 static inline void ct_add_counters(struct ip_conntrack *ct,
903 enum ip_conntrack_info ctinfo,
904 const struct sk_buff *skb)
906 #ifdef CONFIG_IP_NF_CT_ACCT
908 ct->counters[CTINFO2DIR(ctinfo)].packets++;
909 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
910 ntohs(skb->nh.iph->tot_len);
915 /* Refresh conntrack for this many jiffies and do accounting (if skb != NULL) */
916 void ip_ct_refresh_acct(struct ip_conntrack *ct,
917 enum ip_conntrack_info ctinfo,
918 const struct sk_buff *skb,
919 unsigned long extra_jiffies)
921 IP_NF_ASSERT(ct->timeout.data == (unsigned long)ct);
923 /* If not in hash table, timer will not be active yet */
924 if (!is_confirmed(ct)) {
925 ct->timeout.expires = extra_jiffies;
926 ct_add_counters(ct, ctinfo, skb);
928 WRITE_LOCK(&ip_conntrack_lock);
929 /* Need del_timer for race avoidance (may already be dying). */
930 if (del_timer(&ct->timeout)) {
931 ct->timeout.expires = jiffies + extra_jiffies;
932 add_timer(&ct->timeout);
934 ct_add_counters(ct, ctinfo, skb);
935 WRITE_UNLOCK(&ip_conntrack_lock);
939 /* Returns new sk_buff, or NULL */
941 ip_ct_gather_frags(struct sk_buff *skb, u_int32_t user)
943 #ifdef CONFIG_NETFILTER_DEBUG
944 unsigned int olddebug = skb->nf_debug;
950 skb = ip_defrag(skb, user);
954 ip_send_check(skb->nh.iph);
955 skb->nfcache |= NFC_ALTERED;
956 #ifdef CONFIG_NETFILTER_DEBUG
957 /* Packet path as if nothing had happened. */
958 skb->nf_debug = olddebug;
965 /* Used by ipt_REJECT. */
966 static void ip_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
968 struct ip_conntrack *ct;
969 enum ip_conntrack_info ctinfo;
971 /* This ICMP is in reverse direction to the packet which caused it */
972 ct = ip_conntrack_get(skb, &ctinfo);
974 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
975 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
977 ctinfo = IP_CT_RELATED;
979 /* Attach to new skbuff, and increment count */
980 nskb->nfct = &ct->ct_general;
981 nskb->nfctinfo = ctinfo;
982 nf_conntrack_get(nskb->nfct);
986 do_iter(const struct ip_conntrack_tuple_hash *i,
987 int (*iter)(struct ip_conntrack *i, void *data),
990 return iter(tuplehash_to_ctrack(i), data);
993 /* Bring out ya dead! */
994 static struct ip_conntrack_tuple_hash *
995 get_next_corpse(int (*iter)(struct ip_conntrack *i, void *data),
996 void *data, unsigned int *bucket)
998 struct ip_conntrack_tuple_hash *h = NULL;
1000 WRITE_LOCK(&ip_conntrack_lock);
1001 for (; *bucket < ip_conntrack_htable_size; (*bucket)++) {
1002 h = LIST_FIND_W(&ip_conntrack_hash[*bucket], do_iter,
1003 struct ip_conntrack_tuple_hash *, iter, data);
1008 h = LIST_FIND_W(&unconfirmed, do_iter,
1009 struct ip_conntrack_tuple_hash *, iter, data);
1011 atomic_inc(&tuplehash_to_ctrack(h)->ct_general.use);
1012 WRITE_UNLOCK(&ip_conntrack_lock);
1018 ip_ct_iterate_cleanup(int (*iter)(struct ip_conntrack *i, void *), void *data)
1020 struct ip_conntrack_tuple_hash *h;
1021 unsigned int bucket = 0;
1023 while ((h = get_next_corpse(iter, data, &bucket)) != NULL) {
1024 struct ip_conntrack *ct = tuplehash_to_ctrack(h);
1025 /* Time to push up daises... */
1026 if (del_timer(&ct->timeout))
1027 death_by_timeout((unsigned long)ct);
1028 /* ... else the timer will get him soon. */
1030 ip_conntrack_put(ct);
1034 /* Fast function for those who don't want to parse /proc (and I don't
1036 /* Reversing the socket's dst/src point of view gives us the reply
1039 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
1041 struct inet_sock *inet = inet_sk(sk);
1042 struct ip_conntrack_tuple_hash *h;
1043 struct ip_conntrack_tuple tuple;
1045 IP_CT_TUPLE_U_BLANK(&tuple);
1046 tuple.src.ip = inet->rcv_saddr;
1047 tuple.src.u.tcp.port = inet->sport;
1048 tuple.dst.ip = inet->daddr;
1049 tuple.dst.u.tcp.port = inet->dport;
1050 tuple.dst.protonum = IPPROTO_TCP;
1052 /* We only do TCP at the moment: is there a better way? */
1053 if (strcmp(sk->sk_prot->name, "TCP")) {
1054 DEBUGP("SO_ORIGINAL_DST: Not a TCP socket\n");
1055 return -ENOPROTOOPT;
1058 if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
1059 DEBUGP("SO_ORIGINAL_DST: len %u not %u\n",
1060 *len, sizeof(struct sockaddr_in));
1064 h = ip_conntrack_find_get(&tuple, NULL);
1066 struct sockaddr_in sin;
1067 struct ip_conntrack *ct = tuplehash_to_ctrack(h);
1069 sin.sin_family = AF_INET;
1070 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
1071 .tuple.dst.u.tcp.port;
1072 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
1075 DEBUGP("SO_ORIGINAL_DST: %u.%u.%u.%u %u\n",
1076 NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
1077 ip_conntrack_put(ct);
1078 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
1083 DEBUGP("SO_ORIGINAL_DST: Can't find %u.%u.%u.%u/%u-%u.%u.%u.%u/%u.\n",
1084 NIPQUAD(tuple.src.ip), ntohs(tuple.src.u.tcp.port),
1085 NIPQUAD(tuple.dst.ip), ntohs(tuple.dst.u.tcp.port));
1089 static struct nf_sockopt_ops so_getorigdst = {
1091 .get_optmin = SO_ORIGINAL_DST,
1092 .get_optmax = SO_ORIGINAL_DST+1,
1096 static int kill_all(struct ip_conntrack *i, void *data)
1101 static void free_conntrack_hash(void)
1103 if (ip_conntrack_vmalloc)
1104 vfree(ip_conntrack_hash);
1106 free_pages((unsigned long)ip_conntrack_hash,
1107 get_order(sizeof(struct list_head)
1108 * ip_conntrack_htable_size));
1111 /* Mishearing the voices in his head, our hero wonders how he's
1112 supposed to kill the mall. */
1113 void ip_conntrack_cleanup(void)
1115 ip_ct_attach = NULL;
1116 /* This makes sure all current packets have passed through
1117 netfilter framework. Roll on, two-stage module
1122 ip_ct_iterate_cleanup(kill_all, NULL);
1123 if (atomic_read(&ip_conntrack_count) != 0) {
1125 goto i_see_dead_people;
1128 kmem_cache_destroy(ip_conntrack_cachep);
1129 kmem_cache_destroy(ip_conntrack_expect_cachep);
1130 free_conntrack_hash();
1131 nf_unregister_sockopt(&so_getorigdst);
1134 static int hashsize;
1135 module_param(hashsize, int, 0400);
1137 int __init ip_conntrack_init(void)
1142 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB
1143 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */
1145 ip_conntrack_htable_size = hashsize;
1147 ip_conntrack_htable_size
1148 = (((num_physpages << PAGE_SHIFT) / 16384)
1149 / sizeof(struct list_head));
1150 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1151 ip_conntrack_htable_size = 8192;
1152 if (ip_conntrack_htable_size < 16)
1153 ip_conntrack_htable_size = 16;
1155 ip_conntrack_max = 8 * ip_conntrack_htable_size;
1157 printk("ip_conntrack version %s (%u buckets, %d max)"
1158 " - %Zd bytes per conntrack\n", IP_CONNTRACK_VERSION,
1159 ip_conntrack_htable_size, ip_conntrack_max,
1160 sizeof(struct ip_conntrack));
1162 ret = nf_register_sockopt(&so_getorigdst);
1164 printk(KERN_ERR "Unable to register netfilter socket option\n");
1168 /* AK: the hash table is twice as big than needed because it
1169 uses list_head. it would be much nicer to caches to use a
1170 single pointer list head here. */
1171 ip_conntrack_vmalloc = 0;
1173 =(void*)__get_free_pages(GFP_KERNEL,
1174 get_order(sizeof(struct list_head)
1175 *ip_conntrack_htable_size));
1176 if (!ip_conntrack_hash) {
1177 ip_conntrack_vmalloc = 1;
1178 printk(KERN_WARNING "ip_conntrack: falling back to vmalloc.\n");
1179 ip_conntrack_hash = vmalloc(sizeof(struct list_head)
1180 * ip_conntrack_htable_size);
1182 if (!ip_conntrack_hash) {
1183 printk(KERN_ERR "Unable to create ip_conntrack_hash\n");
1184 goto err_unreg_sockopt;
1187 ip_conntrack_cachep = kmem_cache_create("ip_conntrack",
1188 sizeof(struct ip_conntrack), 0,
1190 if (!ip_conntrack_cachep) {
1191 printk(KERN_ERR "Unable to create ip_conntrack slab cache\n");
1195 ip_conntrack_expect_cachep = kmem_cache_create("ip_conntrack_expect",
1196 sizeof(struct ip_conntrack_expect),
1198 if (!ip_conntrack_expect_cachep) {
1199 printk(KERN_ERR "Unable to create ip_expect slab cache\n");
1200 goto err_free_conntrack_slab;
1203 /* Don't NEED lock here, but good form anyway. */
1204 WRITE_LOCK(&ip_conntrack_lock);
1205 for (i = 0; i < MAX_IP_CT_PROTO; i++)
1206 ip_ct_protos[i] = &ip_conntrack_generic_protocol;
1207 /* Sew in builtin protocols. */
1208 ip_ct_protos[IPPROTO_TCP] = &ip_conntrack_protocol_tcp;
1209 ip_ct_protos[IPPROTO_UDP] = &ip_conntrack_protocol_udp;
1210 ip_ct_protos[IPPROTO_ICMP] = &ip_conntrack_protocol_icmp;
1211 WRITE_UNLOCK(&ip_conntrack_lock);
1213 for (i = 0; i < ip_conntrack_htable_size; i++)
1214 INIT_LIST_HEAD(&ip_conntrack_hash[i]);
1216 /* For use by ipt_REJECT */
1217 ip_ct_attach = ip_conntrack_attach;
1219 /* Set up fake conntrack:
1220 - to never be deleted, not in any hashes */
1221 atomic_set(&ip_conntrack_untracked.ct_general.use, 1);
1222 /* - and look it like as a confirmed connection */
1223 set_bit(IPS_CONFIRMED_BIT, &ip_conntrack_untracked.status);
1227 err_free_conntrack_slab:
1228 kmem_cache_destroy(ip_conntrack_cachep);
1230 free_conntrack_hash();
1232 nf_unregister_sockopt(&so_getorigdst);