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 <linux/vmalloc.h>
16 #include <net/checksum.h>
19 #include <net/tcp.h> /* For tcp_prot in getorigdst */
20 #include <linux/icmp.h>
21 #include <linux/udp.h>
22 #include <linux/jhash.h>
24 #include <linux/netfilter_ipv4.h>
25 #include <net/netfilter/nf_conntrack.h>
26 #include <net/netfilter/nf_conntrack_core.h>
27 #include <net/netfilter/nf_nat.h>
28 #include <net/netfilter/nf_nat_protocol.h>
29 #include <net/netfilter/nf_nat_core.h>
30 #include <net/netfilter/nf_nat_helper.h>
31 #include <net/netfilter/nf_conntrack_helper.h>
32 #include <net/netfilter/nf_conntrack_l3proto.h>
33 #include <net/netfilter/nf_conntrack_l4proto.h>
38 #define DEBUGP(format, args...)
41 static DEFINE_RWLOCK(nf_nat_lock);
43 static struct nf_conntrack_l3proto *l3proto = NULL;
45 /* Calculated at init based on memory size */
46 static unsigned int nf_nat_htable_size;
48 static struct list_head *bysource;
50 #define MAX_IP_NAT_PROTO 256
51 static struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO];
53 static inline struct nf_nat_protocol *
54 __nf_nat_proto_find(u_int8_t protonum)
56 return rcu_dereference(nf_nat_protos[protonum]);
59 struct nf_nat_protocol *
60 nf_nat_proto_find_get(u_int8_t protonum)
62 struct nf_nat_protocol *p;
65 p = __nf_nat_proto_find(protonum);
66 if (!try_module_get(p->me))
67 p = &nf_nat_unknown_protocol;
72 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
75 nf_nat_proto_put(struct nf_nat_protocol *p)
79 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
81 /* We keep an extra hash for each conntrack, for fast searching. */
82 static inline unsigned int
83 hash_by_src(const struct nf_conntrack_tuple *tuple)
85 /* Original src, to ensure we map it consistently if poss. */
86 return jhash_3words((__force u32)tuple->src.u3.ip, tuple->src.u.all,
87 tuple->dst.protonum, 0) % nf_nat_htable_size;
90 /* Is this tuple already taken? (not by us) */
92 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
93 const struct nf_conn *ignored_conntrack)
95 /* Conntrack tracking doesn't keep track of outgoing tuples; only
96 incoming ones. NAT means they don't have a fixed mapping,
97 so we invert the tuple and look for the incoming reply.
99 We could keep a separate hash if this proves too slow. */
100 struct nf_conntrack_tuple reply;
102 nf_ct_invert_tuplepr(&reply, tuple);
103 return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
105 EXPORT_SYMBOL(nf_nat_used_tuple);
107 /* If we source map this tuple so reply looks like reply_tuple, will
108 * that meet the constraints of range. */
110 in_range(const struct nf_conntrack_tuple *tuple,
111 const struct nf_nat_range *range)
113 struct nf_nat_protocol *proto;
116 /* If we are supposed to map IPs, then we must be in the
117 range specified, otherwise let this drag us onto a new src IP. */
118 if (range->flags & IP_NAT_RANGE_MAP_IPS) {
119 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
120 ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
125 proto = __nf_nat_proto_find(tuple->dst.protonum);
126 if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
127 proto->in_range(tuple, IP_NAT_MANIP_SRC,
128 &range->min, &range->max))
136 same_src(const struct nf_conn *ct,
137 const struct nf_conntrack_tuple *tuple)
139 const struct nf_conntrack_tuple *t;
141 t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
142 return (t->dst.protonum == tuple->dst.protonum &&
143 t->src.u3.ip == tuple->src.u3.ip &&
144 t->src.u.all == tuple->src.u.all);
147 /* Only called for SRC manip */
149 find_appropriate_src(const struct nf_conntrack_tuple *tuple,
150 struct nf_conntrack_tuple *result,
151 const struct nf_nat_range *range)
153 unsigned int h = hash_by_src(tuple);
154 struct nf_conn_nat *nat;
157 read_lock_bh(&nf_nat_lock);
158 list_for_each_entry(nat, &bysource[h], bysource) {
160 if (same_src(ct, tuple)) {
161 /* Copy source part from reply tuple. */
162 nf_ct_invert_tuplepr(result,
163 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
164 result->dst = tuple->dst;
166 if (in_range(result, range)) {
167 read_unlock_bh(&nf_nat_lock);
172 read_unlock_bh(&nf_nat_lock);
176 /* For [FUTURE] fragmentation handling, we want the least-used
177 src-ip/dst-ip/proto triple. Fairness doesn't come into it. Thus
178 if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
179 1-65535, we don't do pro-rata allocation based on ports; we choose
180 the ip with the lowest src-ip/dst-ip/proto usage.
183 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
184 const struct nf_nat_range *range,
185 const struct nf_conn *ct,
186 enum nf_nat_manip_type maniptype)
190 u_int32_t minip, maxip, j;
192 /* No IP mapping? Do nothing. */
193 if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
196 if (maniptype == IP_NAT_MANIP_SRC)
197 var_ipp = &tuple->src.u3.ip;
199 var_ipp = &tuple->dst.u3.ip;
201 /* Fast path: only one choice. */
202 if (range->min_ip == range->max_ip) {
203 *var_ipp = range->min_ip;
207 /* Hashing source and destination IPs gives a fairly even
208 * spread in practice (if there are a small number of IPs
209 * involved, there usually aren't that many connections
210 * anyway). The consistency means that servers see the same
211 * client coming from the same IP (some Internet Banking sites
212 * like this), even across reboots. */
213 minip = ntohl(range->min_ip);
214 maxip = ntohl(range->max_ip);
215 j = jhash_2words((__force u32)tuple->src.u3.ip,
216 (__force u32)tuple->dst.u3.ip, 0);
217 *var_ipp = htonl(minip + j % (maxip - minip + 1));
220 /* Manipulate the tuple into the range given. For NF_IP_POST_ROUTING,
221 * we change the source to map into the range. For NF_IP_PRE_ROUTING
222 * and NF_IP_LOCAL_OUT, we change the destination to map into the
223 * range. It might not be possible to get a unique tuple, but we try.
224 * At worst (or if we race), we will end up with a final duplicate in
225 * __ip_conntrack_confirm and drop the packet. */
227 get_unique_tuple(struct nf_conntrack_tuple *tuple,
228 const struct nf_conntrack_tuple *orig_tuple,
229 const struct nf_nat_range *range,
231 enum nf_nat_manip_type maniptype)
233 struct nf_nat_protocol *proto;
235 /* 1) If this srcip/proto/src-proto-part is currently mapped,
236 and that same mapping gives a unique tuple within the given
239 This is only required for source (ie. NAT/masq) mappings.
240 So far, we don't do local source mappings, so multiple
241 manips not an issue. */
242 if (maniptype == IP_NAT_MANIP_SRC) {
243 if (find_appropriate_src(orig_tuple, tuple, range)) {
244 DEBUGP("get_unique_tuple: Found current src map\n");
245 if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM))
246 if (!nf_nat_used_tuple(tuple, ct))
251 /* 2) Select the least-used IP/proto combination in the given
253 *tuple = *orig_tuple;
254 find_best_ips_proto(tuple, range, ct, maniptype);
256 /* 3) The per-protocol part of the manip is made to map into
257 the range to make a unique tuple. */
260 proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
262 /* Change protocol info to have some randomization */
263 if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
264 proto->unique_tuple(tuple, range, maniptype, ct);
268 /* Only bother mapping if it's not already in range and unique */
269 if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
270 proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
271 !nf_nat_used_tuple(tuple, ct))
274 /* Last change: get protocol to try to obtain unique tuple. */
275 proto->unique_tuple(tuple, range, maniptype, ct);
281 nf_nat_setup_info(struct nf_conn *ct,
282 const struct nf_nat_range *range,
283 unsigned int hooknum)
285 struct nf_conntrack_tuple curr_tuple, new_tuple;
286 struct nf_conn_nat *nat;
287 int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
288 enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
290 /* nat helper or nfctnetlink also setup binding */
293 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
295 DEBUGP("failed to add NAT extension\n");
300 NF_CT_ASSERT(hooknum == NF_IP_PRE_ROUTING ||
301 hooknum == NF_IP_POST_ROUTING ||
302 hooknum == NF_IP_LOCAL_IN ||
303 hooknum == NF_IP_LOCAL_OUT);
304 BUG_ON(nf_nat_initialized(ct, maniptype));
306 /* What we've got will look like inverse of reply. Normally
307 this is what is in the conntrack, except for prior
308 manipulations (future optimization: if num_manips == 0,
310 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
311 nf_ct_invert_tuplepr(&curr_tuple,
312 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
314 get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
316 if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
317 struct nf_conntrack_tuple reply;
319 /* Alter conntrack table so will recognize replies. */
320 nf_ct_invert_tuplepr(&reply, &new_tuple);
321 nf_conntrack_alter_reply(ct, &reply);
323 /* Non-atomic: we own this at the moment. */
324 if (maniptype == IP_NAT_MANIP_SRC)
325 ct->status |= IPS_SRC_NAT;
327 ct->status |= IPS_DST_NAT;
330 /* Place in source hash if this is the first time. */
332 unsigned int srchash;
334 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
335 write_lock_bh(&nf_nat_lock);
336 /* nf_conntrack_alter_reply might re-allocate exntension aera */
339 list_add(&nat->bysource, &bysource[srchash]);
340 write_unlock_bh(&nf_nat_lock);
344 if (maniptype == IP_NAT_MANIP_DST)
345 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
347 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
351 EXPORT_SYMBOL(nf_nat_setup_info);
353 /* Returns true if succeeded. */
355 manip_pkt(u_int16_t proto,
356 struct sk_buff **pskb,
357 unsigned int iphdroff,
358 const struct nf_conntrack_tuple *target,
359 enum nf_nat_manip_type maniptype)
362 struct nf_nat_protocol *p;
364 if (!skb_make_writable(pskb, iphdroff + sizeof(*iph)))
367 iph = (void *)(*pskb)->data + iphdroff;
369 /* Manipulate protcol part. */
371 /* rcu_read_lock()ed by nf_hook_slow */
372 p = __nf_nat_proto_find(proto);
373 if (!p->manip_pkt(pskb, iphdroff, target, maniptype))
376 iph = (void *)(*pskb)->data + iphdroff;
378 if (maniptype == IP_NAT_MANIP_SRC) {
379 nf_csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
380 iph->saddr = target->src.u3.ip;
382 nf_csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
383 iph->daddr = target->dst.u3.ip;
388 /* Do packet manipulations according to nf_nat_setup_info. */
389 unsigned int nf_nat_packet(struct nf_conn *ct,
390 enum ip_conntrack_info ctinfo,
391 unsigned int hooknum,
392 struct sk_buff **pskb)
394 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
395 unsigned long statusbit;
396 enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
398 if (mtype == IP_NAT_MANIP_SRC)
399 statusbit = IPS_SRC_NAT;
401 statusbit = IPS_DST_NAT;
403 /* Invert if this is reply dir. */
404 if (dir == IP_CT_DIR_REPLY)
405 statusbit ^= IPS_NAT_MASK;
407 /* Non-atomic: these bits don't change. */
408 if (ct->status & statusbit) {
409 struct nf_conntrack_tuple target;
411 /* We are aiming to look like inverse of other direction. */
412 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
414 if (!manip_pkt(target.dst.protonum, pskb, 0, &target, mtype))
419 EXPORT_SYMBOL_GPL(nf_nat_packet);
421 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
422 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
423 enum ip_conntrack_info ctinfo,
424 unsigned int hooknum,
425 struct sk_buff **pskb)
431 struct nf_conntrack_l4proto *l4proto;
432 struct nf_conntrack_tuple inner, target;
433 int hdrlen = ip_hdrlen(*pskb);
434 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
435 unsigned long statusbit;
436 enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
438 if (!skb_make_writable(pskb, hdrlen + sizeof(*inside)))
441 inside = (void *)(*pskb)->data + ip_hdrlen(*pskb);
443 /* We're actually going to mangle it beyond trivial checksum
444 adjustment, so make sure the current checksum is correct. */
445 if (nf_ip_checksum(*pskb, hooknum, hdrlen, 0))
448 /* Must be RELATED */
449 NF_CT_ASSERT((*pskb)->nfctinfo == IP_CT_RELATED ||
450 (*pskb)->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
452 /* Redirects on non-null nats must be dropped, else they'll
453 start talking to each other without our translation, and be
455 if (inside->icmp.type == ICMP_REDIRECT) {
456 /* If NAT isn't finished, assume it and drop. */
457 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
460 if (ct->status & IPS_NAT_MASK)
464 DEBUGP("icmp_reply_translation: translating error %p manp %u dir %s\n",
465 *pskb, manip, dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
467 /* rcu_read_lock()ed by nf_hook_slow */
468 l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
470 if (!nf_ct_get_tuple(*pskb,
471 ip_hdrlen(*pskb) + sizeof(struct icmphdr),
473 sizeof(struct icmphdr) + inside->ip.ihl * 4),
476 &inner, l3proto, l4proto))
479 /* Change inner back to look like incoming packet. We do the
480 opposite manip on this hook to normal, because it might not
481 pass all hooks (locally-generated ICMP). Consider incoming
482 packet: PREROUTING (DST manip), routing produces ICMP, goes
483 through POSTROUTING (which must correct the DST manip). */
484 if (!manip_pkt(inside->ip.protocol, pskb,
485 ip_hdrlen(*pskb) + sizeof(inside->icmp),
486 &ct->tuplehash[!dir].tuple,
490 if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
491 /* Reloading "inside" here since manip_pkt inner. */
492 inside = (void *)(*pskb)->data + ip_hdrlen(*pskb);
493 inside->icmp.checksum = 0;
494 inside->icmp.checksum =
495 csum_fold(skb_checksum(*pskb, hdrlen,
496 (*pskb)->len - hdrlen, 0));
499 /* Change outer to look the reply to an incoming packet
500 * (proto 0 means don't invert per-proto part). */
501 if (manip == IP_NAT_MANIP_SRC)
502 statusbit = IPS_SRC_NAT;
504 statusbit = IPS_DST_NAT;
506 /* Invert if this is reply dir. */
507 if (dir == IP_CT_DIR_REPLY)
508 statusbit ^= IPS_NAT_MASK;
510 if (ct->status & statusbit) {
511 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
512 if (!manip_pkt(0, pskb, 0, &target, manip))
518 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
520 /* Protocol registration. */
521 int nf_nat_protocol_register(struct nf_nat_protocol *proto)
525 write_lock_bh(&nf_nat_lock);
526 if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
530 rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
532 write_unlock_bh(&nf_nat_lock);
535 EXPORT_SYMBOL(nf_nat_protocol_register);
537 /* Noone stores the protocol anywhere; simply delete it. */
538 void nf_nat_protocol_unregister(struct nf_nat_protocol *proto)
540 write_lock_bh(&nf_nat_lock);
541 rcu_assign_pointer(nf_nat_protos[proto->protonum],
542 &nf_nat_unknown_protocol);
543 write_unlock_bh(&nf_nat_lock);
546 EXPORT_SYMBOL(nf_nat_protocol_unregister);
548 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
550 nf_nat_port_range_to_nfattr(struct sk_buff *skb,
551 const struct nf_nat_range *range)
553 NFA_PUT(skb, CTA_PROTONAT_PORT_MIN, sizeof(__be16),
554 &range->min.tcp.port);
555 NFA_PUT(skb, CTA_PROTONAT_PORT_MAX, sizeof(__be16),
556 &range->max.tcp.port);
563 EXPORT_SYMBOL_GPL(nf_nat_port_nfattr_to_range);
566 nf_nat_port_nfattr_to_range(struct nfattr *tb[], struct nf_nat_range *range)
570 /* we have to return whether we actually parsed something or not */
572 if (tb[CTA_PROTONAT_PORT_MIN-1]) {
574 range->min.tcp.port =
575 *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MIN-1]);
578 if (!tb[CTA_PROTONAT_PORT_MAX-1]) {
580 range->max.tcp.port = range->min.tcp.port;
583 range->max.tcp.port =
584 *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MAX-1]);
589 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nfattr);
592 /* Noone using conntrack by the time this called. */
593 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
595 struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
597 if (nat == NULL || nat->ct == NULL)
600 NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
602 write_lock_bh(&nf_nat_lock);
603 list_del(&nat->bysource);
605 write_unlock_bh(&nf_nat_lock);
608 static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
610 struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
611 struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
612 struct nf_conn *ct = old_nat->ct;
613 unsigned int srchash;
615 if (!(ct->status & IPS_NAT_DONE_MASK))
618 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
620 write_lock_bh(&nf_nat_lock);
621 list_replace(&old_nat->bysource, &new_nat->bysource);
623 write_unlock_bh(&nf_nat_lock);
626 static struct nf_ct_ext_type nat_extend __read_mostly = {
627 .len = sizeof(struct nf_conn_nat),
628 .align = __alignof__(struct nf_conn_nat),
629 .destroy = nf_nat_cleanup_conntrack,
630 .move = nf_nat_move_storage,
632 .flags = NF_CT_EXT_F_PREALLOC,
635 static int __init nf_nat_init(void)
640 ret = nf_ct_extend_register(&nat_extend);
642 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
646 /* Leave them the same for the moment. */
647 nf_nat_htable_size = nf_conntrack_htable_size;
649 /* One vmalloc for both hash tables */
650 bysource = vmalloc(sizeof(struct list_head) * nf_nat_htable_size);
656 /* Sew in builtin protocols. */
657 write_lock_bh(&nf_nat_lock);
658 for (i = 0; i < MAX_IP_NAT_PROTO; i++)
659 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
660 rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
661 rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
662 rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
663 write_unlock_bh(&nf_nat_lock);
665 for (i = 0; i < nf_nat_htable_size; i++) {
666 INIT_LIST_HEAD(&bysource[i]);
669 /* Initialize fake conntrack so that NAT will skip it */
670 nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
672 l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
676 nf_ct_extend_unregister(&nat_extend);
680 /* Clear NAT section of all conntracks, in case we're loaded again. */
681 static int clean_nat(struct nf_conn *i, void *data)
683 struct nf_conn_nat *nat = nfct_nat(i);
687 memset(nat, 0, sizeof(nat));
688 i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
692 static void __exit nf_nat_cleanup(void)
694 nf_ct_iterate_cleanup(&clean_nat, NULL);
697 nf_ct_l3proto_put(l3proto);
698 nf_ct_extend_unregister(&nat_extend);
701 MODULE_LICENSE("GPL");
703 module_init(nf_nat_init);
704 module_exit(nf_nat_cleanup);