1 /* NAT for netfilter; shared with compatibility layer. */
3 /* (C) 1999-2001 Paul `Rusty' Russell
4 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
11 #include <linux/module.h>
12 #include <linux/types.h>
13 #include <linux/timer.h>
14 #include <linux/skbuff.h>
15 #include <net/checksum.h>
18 #include <net/tcp.h> /* For tcp_prot in getorigdst */
19 #include <linux/icmp.h>
20 #include <linux/udp.h>
21 #include <linux/jhash.h>
23 #include <linux/netfilter_ipv4.h>
24 #include <net/netfilter/nf_conntrack.h>
25 #include <net/netfilter/nf_conntrack_core.h>
26 #include <net/netfilter/nf_nat.h>
27 #include <net/netfilter/nf_nat_protocol.h>
28 #include <net/netfilter/nf_nat_core.h>
29 #include <net/netfilter/nf_nat_helper.h>
30 #include <net/netfilter/nf_conntrack_helper.h>
31 #include <net/netfilter/nf_conntrack_l3proto.h>
32 #include <net/netfilter/nf_conntrack_l4proto.h>
34 static DEFINE_RWLOCK(nf_nat_lock);
36 static struct nf_conntrack_l3proto *l3proto __read_mostly;
38 /* Calculated at init based on memory size */
39 static unsigned int nf_nat_htable_size __read_mostly;
40 static int nf_nat_vmalloced;
42 static struct hlist_head *bysource __read_mostly;
44 #define MAX_IP_NAT_PROTO 256
45 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
48 static inline const struct nf_nat_protocol *
49 __nf_nat_proto_find(u_int8_t protonum)
51 return rcu_dereference(nf_nat_protos[protonum]);
54 const struct nf_nat_protocol *
55 nf_nat_proto_find_get(u_int8_t protonum)
57 const struct nf_nat_protocol *p;
60 p = __nf_nat_proto_find(protonum);
61 if (!try_module_get(p->me))
62 p = &nf_nat_unknown_protocol;
67 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
70 nf_nat_proto_put(const struct nf_nat_protocol *p)
74 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
76 /* We keep an extra hash for each conntrack, for fast searching. */
77 static inline unsigned int
78 hash_by_src(const struct nf_conntrack_tuple *tuple)
80 /* Original src, to ensure we map it consistently if poss. */
81 return jhash_3words((__force u32)tuple->src.u3.ip,
82 (__force u32)tuple->src.u.all,
83 tuple->dst.protonum, 0) % nf_nat_htable_size;
86 /* Is this tuple already taken? (not by us) */
88 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
89 const struct nf_conn *ignored_conntrack)
91 /* Conntrack tracking doesn't keep track of outgoing tuples; only
92 incoming ones. NAT means they don't have a fixed mapping,
93 so we invert the tuple and look for the incoming reply.
95 We could keep a separate hash if this proves too slow. */
96 struct nf_conntrack_tuple reply;
98 nf_ct_invert_tuplepr(&reply, tuple);
99 return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
101 EXPORT_SYMBOL(nf_nat_used_tuple);
103 /* If we source map this tuple so reply looks like reply_tuple, will
104 * that meet the constraints of range. */
106 in_range(const struct nf_conntrack_tuple *tuple,
107 const struct nf_nat_range *range)
109 const struct nf_nat_protocol *proto;
112 /* If we are supposed to map IPs, then we must be in the
113 range specified, otherwise let this drag us onto a new src IP. */
114 if (range->flags & IP_NAT_RANGE_MAP_IPS) {
115 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
116 ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
121 proto = __nf_nat_proto_find(tuple->dst.protonum);
122 if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
123 proto->in_range(tuple, IP_NAT_MANIP_SRC,
124 &range->min, &range->max))
132 same_src(const struct nf_conn *ct,
133 const struct nf_conntrack_tuple *tuple)
135 const struct nf_conntrack_tuple *t;
137 t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
138 return (t->dst.protonum == tuple->dst.protonum &&
139 t->src.u3.ip == tuple->src.u3.ip &&
140 t->src.u.all == tuple->src.u.all);
143 /* Only called for SRC manip */
145 find_appropriate_src(const struct nf_conntrack_tuple *tuple,
146 struct nf_conntrack_tuple *result,
147 const struct nf_nat_range *range)
149 unsigned int h = hash_by_src(tuple);
150 struct nf_conn_nat *nat;
152 struct hlist_node *n;
154 read_lock_bh(&nf_nat_lock);
155 hlist_for_each_entry(nat, n, &bysource[h], bysource) {
157 if (same_src(ct, tuple)) {
158 /* Copy source part from reply tuple. */
159 nf_ct_invert_tuplepr(result,
160 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
161 result->dst = tuple->dst;
163 if (in_range(result, range)) {
164 read_unlock_bh(&nf_nat_lock);
169 read_unlock_bh(&nf_nat_lock);
173 /* For [FUTURE] fragmentation handling, we want the least-used
174 src-ip/dst-ip/proto triple. Fairness doesn't come into it. Thus
175 if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
176 1-65535, we don't do pro-rata allocation based on ports; we choose
177 the ip with the lowest src-ip/dst-ip/proto usage.
180 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
181 const struct nf_nat_range *range,
182 const struct nf_conn *ct,
183 enum nf_nat_manip_type maniptype)
187 u_int32_t minip, maxip, j;
189 /* No IP mapping? Do nothing. */
190 if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
193 if (maniptype == IP_NAT_MANIP_SRC)
194 var_ipp = &tuple->src.u3.ip;
196 var_ipp = &tuple->dst.u3.ip;
198 /* Fast path: only one choice. */
199 if (range->min_ip == range->max_ip) {
200 *var_ipp = range->min_ip;
204 /* Hashing source and destination IPs gives a fairly even
205 * spread in practice (if there are a small number of IPs
206 * involved, there usually aren't that many connections
207 * anyway). The consistency means that servers see the same
208 * client coming from the same IP (some Internet Banking sites
209 * like this), even across reboots. */
210 minip = ntohl(range->min_ip);
211 maxip = ntohl(range->max_ip);
212 j = jhash_2words((__force u32)tuple->src.u3.ip,
213 (__force u32)tuple->dst.u3.ip, 0);
214 *var_ipp = htonl(minip + j % (maxip - minip + 1));
217 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
218 * we change the source to map into the range. For NF_INET_PRE_ROUTING
219 * and NF_INET_LOCAL_OUT, we change the destination to map into the
220 * range. It might not be possible to get a unique tuple, but we try.
221 * At worst (or if we race), we will end up with a final duplicate in
222 * __ip_conntrack_confirm and drop the packet. */
224 get_unique_tuple(struct nf_conntrack_tuple *tuple,
225 const struct nf_conntrack_tuple *orig_tuple,
226 const struct nf_nat_range *range,
228 enum nf_nat_manip_type maniptype)
230 const struct nf_nat_protocol *proto;
232 /* 1) If this srcip/proto/src-proto-part is currently mapped,
233 and that same mapping gives a unique tuple within the given
236 This is only required for source (ie. NAT/masq) mappings.
237 So far, we don't do local source mappings, so multiple
238 manips not an issue. */
239 if (maniptype == IP_NAT_MANIP_SRC) {
240 if (find_appropriate_src(orig_tuple, tuple, range)) {
241 pr_debug("get_unique_tuple: Found current src map\n");
242 if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM))
243 if (!nf_nat_used_tuple(tuple, ct))
248 /* 2) Select the least-used IP/proto combination in the given
250 *tuple = *orig_tuple;
251 find_best_ips_proto(tuple, range, ct, maniptype);
253 /* 3) The per-protocol part of the manip is made to map into
254 the range to make a unique tuple. */
257 proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
259 /* Change protocol info to have some randomization */
260 if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
261 proto->unique_tuple(tuple, range, maniptype, ct);
265 /* Only bother mapping if it's not already in range and unique */
266 if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
267 proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
268 !nf_nat_used_tuple(tuple, ct))
271 /* Last change: get protocol to try to obtain unique tuple. */
272 proto->unique_tuple(tuple, range, maniptype, ct);
278 nf_nat_setup_info(struct nf_conn *ct,
279 const struct nf_nat_range *range,
280 unsigned int hooknum)
282 struct nf_conntrack_tuple curr_tuple, new_tuple;
283 struct nf_conn_nat *nat;
284 int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
285 enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
287 /* nat helper or nfctnetlink also setup binding */
290 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
292 pr_debug("failed to add NAT extension\n");
297 NF_CT_ASSERT(hooknum == NF_INET_PRE_ROUTING ||
298 hooknum == NF_INET_POST_ROUTING ||
299 hooknum == NF_INET_LOCAL_IN ||
300 hooknum == NF_INET_LOCAL_OUT);
301 BUG_ON(nf_nat_initialized(ct, maniptype));
303 /* What we've got will look like inverse of reply. Normally
304 this is what is in the conntrack, except for prior
305 manipulations (future optimization: if num_manips == 0,
307 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
308 nf_ct_invert_tuplepr(&curr_tuple,
309 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
311 get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
313 if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
314 struct nf_conntrack_tuple reply;
316 /* Alter conntrack table so will recognize replies. */
317 nf_ct_invert_tuplepr(&reply, &new_tuple);
318 nf_conntrack_alter_reply(ct, &reply);
320 /* Non-atomic: we own this at the moment. */
321 if (maniptype == IP_NAT_MANIP_SRC)
322 ct->status |= IPS_SRC_NAT;
324 ct->status |= IPS_DST_NAT;
327 /* Place in source hash if this is the first time. */
329 unsigned int srchash;
331 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
332 write_lock_bh(&nf_nat_lock);
333 /* nf_conntrack_alter_reply might re-allocate exntension aera */
336 hlist_add_head(&nat->bysource, &bysource[srchash]);
337 write_unlock_bh(&nf_nat_lock);
341 if (maniptype == IP_NAT_MANIP_DST)
342 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
344 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
348 EXPORT_SYMBOL(nf_nat_setup_info);
350 /* Returns true if succeeded. */
352 manip_pkt(u_int16_t proto,
354 unsigned int iphdroff,
355 const struct nf_conntrack_tuple *target,
356 enum nf_nat_manip_type maniptype)
359 const struct nf_nat_protocol *p;
361 if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
364 iph = (void *)skb->data + iphdroff;
366 /* Manipulate protcol part. */
368 /* rcu_read_lock()ed by nf_hook_slow */
369 p = __nf_nat_proto_find(proto);
370 if (!p->manip_pkt(skb, iphdroff, target, maniptype))
373 iph = (void *)skb->data + iphdroff;
375 if (maniptype == IP_NAT_MANIP_SRC) {
376 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
377 iph->saddr = target->src.u3.ip;
379 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
380 iph->daddr = target->dst.u3.ip;
385 /* Do packet manipulations according to nf_nat_setup_info. */
386 unsigned int nf_nat_packet(struct nf_conn *ct,
387 enum ip_conntrack_info ctinfo,
388 unsigned int hooknum,
391 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
392 unsigned long statusbit;
393 enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
395 if (mtype == IP_NAT_MANIP_SRC)
396 statusbit = IPS_SRC_NAT;
398 statusbit = IPS_DST_NAT;
400 /* Invert if this is reply dir. */
401 if (dir == IP_CT_DIR_REPLY)
402 statusbit ^= IPS_NAT_MASK;
404 /* Non-atomic: these bits don't change. */
405 if (ct->status & statusbit) {
406 struct nf_conntrack_tuple target;
408 /* We are aiming to look like inverse of other direction. */
409 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
411 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
416 EXPORT_SYMBOL_GPL(nf_nat_packet);
418 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
419 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
420 enum ip_conntrack_info ctinfo,
421 unsigned int hooknum,
428 struct nf_conntrack_l4proto *l4proto;
429 struct nf_conntrack_tuple inner, target;
430 int hdrlen = ip_hdrlen(skb);
431 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
432 unsigned long statusbit;
433 enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
435 if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
438 inside = (void *)skb->data + ip_hdrlen(skb);
440 /* We're actually going to mangle it beyond trivial checksum
441 adjustment, so make sure the current checksum is correct. */
442 if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
445 /* Must be RELATED */
446 NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
447 skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
449 /* Redirects on non-null nats must be dropped, else they'll
450 start talking to each other without our translation, and be
452 if (inside->icmp.type == ICMP_REDIRECT) {
453 /* If NAT isn't finished, assume it and drop. */
454 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
457 if (ct->status & IPS_NAT_MASK)
461 pr_debug("icmp_reply_translation: translating error %p manip %u "
462 "dir %s\n", skb, manip,
463 dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
465 /* rcu_read_lock()ed by nf_hook_slow */
466 l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
468 if (!nf_ct_get_tuple(skb,
469 ip_hdrlen(skb) + sizeof(struct icmphdr),
471 sizeof(struct icmphdr) + inside->ip.ihl * 4),
474 &inner, l3proto, l4proto))
477 /* Change inner back to look like incoming packet. We do the
478 opposite manip on this hook to normal, because it might not
479 pass all hooks (locally-generated ICMP). Consider incoming
480 packet: PREROUTING (DST manip), routing produces ICMP, goes
481 through POSTROUTING (which must correct the DST manip). */
482 if (!manip_pkt(inside->ip.protocol, skb,
483 ip_hdrlen(skb) + sizeof(inside->icmp),
484 &ct->tuplehash[!dir].tuple,
488 if (skb->ip_summed != CHECKSUM_PARTIAL) {
489 /* Reloading "inside" here since manip_pkt inner. */
490 inside = (void *)skb->data + ip_hdrlen(skb);
491 inside->icmp.checksum = 0;
492 inside->icmp.checksum =
493 csum_fold(skb_checksum(skb, hdrlen,
494 skb->len - hdrlen, 0));
497 /* Change outer to look the reply to an incoming packet
498 * (proto 0 means don't invert per-proto part). */
499 if (manip == IP_NAT_MANIP_SRC)
500 statusbit = IPS_SRC_NAT;
502 statusbit = IPS_DST_NAT;
504 /* Invert if this is reply dir. */
505 if (dir == IP_CT_DIR_REPLY)
506 statusbit ^= IPS_NAT_MASK;
508 if (ct->status & statusbit) {
509 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
510 if (!manip_pkt(0, skb, 0, &target, manip))
516 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
518 /* Protocol registration. */
519 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
523 write_lock_bh(&nf_nat_lock);
524 if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
528 rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
530 write_unlock_bh(&nf_nat_lock);
533 EXPORT_SYMBOL(nf_nat_protocol_register);
535 /* Noone stores the protocol anywhere; simply delete it. */
536 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
538 write_lock_bh(&nf_nat_lock);
539 rcu_assign_pointer(nf_nat_protos[proto->protonum],
540 &nf_nat_unknown_protocol);
541 write_unlock_bh(&nf_nat_lock);
544 EXPORT_SYMBOL(nf_nat_protocol_unregister);
546 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
548 nf_nat_port_range_to_nlattr(struct sk_buff *skb,
549 const struct nf_nat_range *range)
551 NLA_PUT_BE16(skb, CTA_PROTONAT_PORT_MIN, range->min.tcp.port);
552 NLA_PUT_BE16(skb, CTA_PROTONAT_PORT_MAX, range->max.tcp.port);
559 EXPORT_SYMBOL_GPL(nf_nat_port_nlattr_to_range);
562 nf_nat_port_nlattr_to_range(struct nlattr *tb[], struct nf_nat_range *range)
566 /* we have to return whether we actually parsed something or not */
568 if (tb[CTA_PROTONAT_PORT_MIN]) {
570 range->min.tcp.port = nla_get_be16(tb[CTA_PROTONAT_PORT_MIN]);
573 if (!tb[CTA_PROTONAT_PORT_MAX]) {
575 range->max.tcp.port = range->min.tcp.port;
578 range->max.tcp.port = nla_get_be16(tb[CTA_PROTONAT_PORT_MAX]);
583 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nlattr);
586 /* Noone using conntrack by the time this called. */
587 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
589 struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
591 if (nat == NULL || nat->ct == NULL)
594 NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
596 write_lock_bh(&nf_nat_lock);
597 hlist_del(&nat->bysource);
599 write_unlock_bh(&nf_nat_lock);
602 static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
604 struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
605 struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
606 struct nf_conn *ct = old_nat->ct;
608 if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
611 write_lock_bh(&nf_nat_lock);
612 hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
614 write_unlock_bh(&nf_nat_lock);
617 static struct nf_ct_ext_type nat_extend __read_mostly = {
618 .len = sizeof(struct nf_conn_nat),
619 .align = __alignof__(struct nf_conn_nat),
620 .destroy = nf_nat_cleanup_conntrack,
621 .move = nf_nat_move_storage,
623 .flags = NF_CT_EXT_F_PREALLOC,
626 static int __init nf_nat_init(void)
631 ret = nf_ct_extend_register(&nat_extend);
633 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
637 /* Leave them the same for the moment. */
638 nf_nat_htable_size = nf_conntrack_htable_size;
640 bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
647 /* Sew in builtin protocols. */
648 write_lock_bh(&nf_nat_lock);
649 for (i = 0; i < MAX_IP_NAT_PROTO; i++)
650 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
651 rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
652 rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
653 rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
654 write_unlock_bh(&nf_nat_lock);
656 for (i = 0; i < nf_nat_htable_size; i++) {
657 INIT_HLIST_HEAD(&bysource[i]);
660 /* Initialize fake conntrack so that NAT will skip it */
661 nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
663 l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
667 nf_ct_extend_unregister(&nat_extend);
671 /* Clear NAT section of all conntracks, in case we're loaded again. */
672 static int clean_nat(struct nf_conn *i, void *data)
674 struct nf_conn_nat *nat = nfct_nat(i);
678 memset(nat, 0, sizeof(*nat));
679 i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
683 static void __exit nf_nat_cleanup(void)
685 nf_ct_iterate_cleanup(&clean_nat, NULL);
687 nf_ct_free_hashtable(bysource, nf_nat_vmalloced, nf_nat_htable_size);
688 nf_ct_l3proto_put(l3proto);
689 nf_ct_extend_unregister(&nat_extend);
692 MODULE_LICENSE("GPL");
694 module_init(nf_nat_init);
695 module_exit(nf_nat_cleanup);