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 /* Noone using conntrack by the time this called. */
91 static void nf_nat_cleanup_conntrack(struct nf_conn *conn)
93 struct nf_conn_nat *nat;
94 if (!(conn->status & IPS_NAT_DONE_MASK))
98 write_lock_bh(&nf_nat_lock);
99 list_del(&nat->info.bysource);
101 write_unlock_bh(&nf_nat_lock);
104 /* Is this tuple already taken? (not by us) */
106 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
107 const struct nf_conn *ignored_conntrack)
109 /* Conntrack tracking doesn't keep track of outgoing tuples; only
110 incoming ones. NAT means they don't have a fixed mapping,
111 so we invert the tuple and look for the incoming reply.
113 We could keep a separate hash if this proves too slow. */
114 struct nf_conntrack_tuple reply;
116 nf_ct_invert_tuplepr(&reply, tuple);
117 return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
119 EXPORT_SYMBOL(nf_nat_used_tuple);
121 /* If we source map this tuple so reply looks like reply_tuple, will
122 * that meet the constraints of range. */
124 in_range(const struct nf_conntrack_tuple *tuple,
125 const struct nf_nat_range *range)
127 struct nf_nat_protocol *proto;
130 /* If we are supposed to map IPs, then we must be in the
131 range specified, otherwise let this drag us onto a new src IP. */
132 if (range->flags & IP_NAT_RANGE_MAP_IPS) {
133 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
134 ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
139 proto = __nf_nat_proto_find(tuple->dst.protonum);
140 if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
141 proto->in_range(tuple, IP_NAT_MANIP_SRC,
142 &range->min, &range->max))
150 same_src(const struct nf_conn *ct,
151 const struct nf_conntrack_tuple *tuple)
153 const struct nf_conntrack_tuple *t;
155 t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
156 return (t->dst.protonum == tuple->dst.protonum &&
157 t->src.u3.ip == tuple->src.u3.ip &&
158 t->src.u.all == tuple->src.u.all);
161 /* Only called for SRC manip */
163 find_appropriate_src(const struct nf_conntrack_tuple *tuple,
164 struct nf_conntrack_tuple *result,
165 const struct nf_nat_range *range)
167 unsigned int h = hash_by_src(tuple);
168 struct nf_conn_nat *nat;
171 read_lock_bh(&nf_nat_lock);
172 list_for_each_entry(nat, &bysource[h], info.bysource) {
174 if (same_src(ct, tuple)) {
175 /* Copy source part from reply tuple. */
176 nf_ct_invert_tuplepr(result,
177 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
178 result->dst = tuple->dst;
180 if (in_range(result, range)) {
181 read_unlock_bh(&nf_nat_lock);
186 read_unlock_bh(&nf_nat_lock);
190 /* For [FUTURE] fragmentation handling, we want the least-used
191 src-ip/dst-ip/proto triple. Fairness doesn't come into it. Thus
192 if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
193 1-65535, we don't do pro-rata allocation based on ports; we choose
194 the ip with the lowest src-ip/dst-ip/proto usage.
197 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
198 const struct nf_nat_range *range,
199 const struct nf_conn *ct,
200 enum nf_nat_manip_type maniptype)
204 u_int32_t minip, maxip, j;
206 /* No IP mapping? Do nothing. */
207 if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
210 if (maniptype == IP_NAT_MANIP_SRC)
211 var_ipp = &tuple->src.u3.ip;
213 var_ipp = &tuple->dst.u3.ip;
215 /* Fast path: only one choice. */
216 if (range->min_ip == range->max_ip) {
217 *var_ipp = range->min_ip;
221 /* Hashing source and destination IPs gives a fairly even
222 * spread in practice (if there are a small number of IPs
223 * involved, there usually aren't that many connections
224 * anyway). The consistency means that servers see the same
225 * client coming from the same IP (some Internet Banking sites
226 * like this), even across reboots. */
227 minip = ntohl(range->min_ip);
228 maxip = ntohl(range->max_ip);
229 j = jhash_2words((__force u32)tuple->src.u3.ip,
230 (__force u32)tuple->dst.u3.ip, 0);
231 *var_ipp = htonl(minip + j % (maxip - minip + 1));
234 /* Manipulate the tuple into the range given. For NF_IP_POST_ROUTING,
235 * we change the source to map into the range. For NF_IP_PRE_ROUTING
236 * and NF_IP_LOCAL_OUT, we change the destination to map into the
237 * range. It might not be possible to get a unique tuple, but we try.
238 * At worst (or if we race), we will end up with a final duplicate in
239 * __ip_conntrack_confirm and drop the packet. */
241 get_unique_tuple(struct nf_conntrack_tuple *tuple,
242 const struct nf_conntrack_tuple *orig_tuple,
243 const struct nf_nat_range *range,
245 enum nf_nat_manip_type maniptype)
247 struct nf_nat_protocol *proto;
249 /* 1) If this srcip/proto/src-proto-part is currently mapped,
250 and that same mapping gives a unique tuple within the given
253 This is only required for source (ie. NAT/masq) mappings.
254 So far, we don't do local source mappings, so multiple
255 manips not an issue. */
256 if (maniptype == IP_NAT_MANIP_SRC) {
257 if (find_appropriate_src(orig_tuple, tuple, range)) {
258 DEBUGP("get_unique_tuple: Found current src map\n");
259 if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM))
260 if (!nf_nat_used_tuple(tuple, ct))
265 /* 2) Select the least-used IP/proto combination in the given
267 *tuple = *orig_tuple;
268 find_best_ips_proto(tuple, range, ct, maniptype);
270 /* 3) The per-protocol part of the manip is made to map into
271 the range to make a unique tuple. */
274 proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
276 /* Change protocol info to have some randomization */
277 if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
278 proto->unique_tuple(tuple, range, maniptype, ct);
282 /* Only bother mapping if it's not already in range and unique */
283 if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
284 proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
285 !nf_nat_used_tuple(tuple, ct))
288 /* Last change: get protocol to try to obtain unique tuple. */
289 proto->unique_tuple(tuple, range, maniptype, ct);
295 nf_nat_setup_info(struct nf_conn *ct,
296 const struct nf_nat_range *range,
297 unsigned int hooknum)
299 struct nf_conntrack_tuple curr_tuple, new_tuple;
300 struct nf_conn_nat *nat;
301 struct nf_nat_info *info;
302 int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
303 enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
305 /* nat helper or nfctnetlink also setup binding */
308 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
310 DEBUGP("failed to add NAT extension\n");
315 NF_CT_ASSERT(hooknum == NF_IP_PRE_ROUTING ||
316 hooknum == NF_IP_POST_ROUTING ||
317 hooknum == NF_IP_LOCAL_IN ||
318 hooknum == NF_IP_LOCAL_OUT);
319 BUG_ON(nf_nat_initialized(ct, maniptype));
321 /* What we've got will look like inverse of reply. Normally
322 this is what is in the conntrack, except for prior
323 manipulations (future optimization: if num_manips == 0,
325 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
326 nf_ct_invert_tuplepr(&curr_tuple,
327 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
329 get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
331 if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
332 struct nf_conntrack_tuple reply;
334 /* Alter conntrack table so will recognize replies. */
335 nf_ct_invert_tuplepr(&reply, &new_tuple);
336 nf_conntrack_alter_reply(ct, &reply);
338 /* Non-atomic: we own this at the moment. */
339 if (maniptype == IP_NAT_MANIP_SRC)
340 ct->status |= IPS_SRC_NAT;
342 ct->status |= IPS_DST_NAT;
345 /* Place in source hash if this is the first time. */
347 unsigned int srchash;
349 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
350 write_lock_bh(&nf_nat_lock);
351 /* nf_conntrack_alter_reply might re-allocate exntension aera */
352 info = &nfct_nat(ct)->info;
354 list_add(&info->bysource, &bysource[srchash]);
355 write_unlock_bh(&nf_nat_lock);
359 if (maniptype == IP_NAT_MANIP_DST)
360 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
362 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
366 EXPORT_SYMBOL(nf_nat_setup_info);
368 /* Returns true if succeeded. */
370 manip_pkt(u_int16_t proto,
371 struct sk_buff **pskb,
372 unsigned int iphdroff,
373 const struct nf_conntrack_tuple *target,
374 enum nf_nat_manip_type maniptype)
377 struct nf_nat_protocol *p;
379 if (!skb_make_writable(pskb, iphdroff + sizeof(*iph)))
382 iph = (void *)(*pskb)->data + iphdroff;
384 /* Manipulate protcol part. */
386 /* rcu_read_lock()ed by nf_hook_slow */
387 p = __nf_nat_proto_find(proto);
388 if (!p->manip_pkt(pskb, iphdroff, target, maniptype))
391 iph = (void *)(*pskb)->data + iphdroff;
393 if (maniptype == IP_NAT_MANIP_SRC) {
394 nf_csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
395 iph->saddr = target->src.u3.ip;
397 nf_csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
398 iph->daddr = target->dst.u3.ip;
403 /* Do packet manipulations according to nf_nat_setup_info. */
404 unsigned int nf_nat_packet(struct nf_conn *ct,
405 enum ip_conntrack_info ctinfo,
406 unsigned int hooknum,
407 struct sk_buff **pskb)
409 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
410 unsigned long statusbit;
411 enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
413 if (mtype == IP_NAT_MANIP_SRC)
414 statusbit = IPS_SRC_NAT;
416 statusbit = IPS_DST_NAT;
418 /* Invert if this is reply dir. */
419 if (dir == IP_CT_DIR_REPLY)
420 statusbit ^= IPS_NAT_MASK;
422 /* Non-atomic: these bits don't change. */
423 if (ct->status & statusbit) {
424 struct nf_conntrack_tuple target;
426 /* We are aiming to look like inverse of other direction. */
427 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
429 if (!manip_pkt(target.dst.protonum, pskb, 0, &target, mtype))
434 EXPORT_SYMBOL_GPL(nf_nat_packet);
436 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
437 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
438 enum ip_conntrack_info ctinfo,
439 unsigned int hooknum,
440 struct sk_buff **pskb)
446 struct nf_conntrack_l4proto *l4proto;
447 struct nf_conntrack_tuple inner, target;
448 int hdrlen = ip_hdrlen(*pskb);
449 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
450 unsigned long statusbit;
451 enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
453 if (!skb_make_writable(pskb, hdrlen + sizeof(*inside)))
456 inside = (void *)(*pskb)->data + ip_hdrlen(*pskb);
458 /* We're actually going to mangle it beyond trivial checksum
459 adjustment, so make sure the current checksum is correct. */
460 if (nf_ip_checksum(*pskb, hooknum, hdrlen, 0))
463 /* Must be RELATED */
464 NF_CT_ASSERT((*pskb)->nfctinfo == IP_CT_RELATED ||
465 (*pskb)->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
467 /* Redirects on non-null nats must be dropped, else they'll
468 start talking to each other without our translation, and be
470 if (inside->icmp.type == ICMP_REDIRECT) {
471 /* If NAT isn't finished, assume it and drop. */
472 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
475 if (ct->status & IPS_NAT_MASK)
479 DEBUGP("icmp_reply_translation: translating error %p manp %u dir %s\n",
480 *pskb, manip, dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
482 /* rcu_read_lock()ed by nf_hook_slow */
483 l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
485 if (!nf_ct_get_tuple(*pskb,
486 ip_hdrlen(*pskb) + sizeof(struct icmphdr),
488 sizeof(struct icmphdr) + inside->ip.ihl * 4),
491 &inner, l3proto, l4proto))
494 /* Change inner back to look like incoming packet. We do the
495 opposite manip on this hook to normal, because it might not
496 pass all hooks (locally-generated ICMP). Consider incoming
497 packet: PREROUTING (DST manip), routing produces ICMP, goes
498 through POSTROUTING (which must correct the DST manip). */
499 if (!manip_pkt(inside->ip.protocol, pskb,
500 ip_hdrlen(*pskb) + sizeof(inside->icmp),
501 &ct->tuplehash[!dir].tuple,
505 if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
506 /* Reloading "inside" here since manip_pkt inner. */
507 inside = (void *)(*pskb)->data + ip_hdrlen(*pskb);
508 inside->icmp.checksum = 0;
509 inside->icmp.checksum =
510 csum_fold(skb_checksum(*pskb, hdrlen,
511 (*pskb)->len - hdrlen, 0));
514 /* Change outer to look the reply to an incoming packet
515 * (proto 0 means don't invert per-proto part). */
516 if (manip == IP_NAT_MANIP_SRC)
517 statusbit = IPS_SRC_NAT;
519 statusbit = IPS_DST_NAT;
521 /* Invert if this is reply dir. */
522 if (dir == IP_CT_DIR_REPLY)
523 statusbit ^= IPS_NAT_MASK;
525 if (ct->status & statusbit) {
526 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
527 if (!manip_pkt(0, pskb, 0, &target, manip))
533 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
535 /* Protocol registration. */
536 int nf_nat_protocol_register(struct nf_nat_protocol *proto)
540 write_lock_bh(&nf_nat_lock);
541 if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
545 rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
547 write_unlock_bh(&nf_nat_lock);
550 EXPORT_SYMBOL(nf_nat_protocol_register);
552 /* Noone stores the protocol anywhere; simply delete it. */
553 void nf_nat_protocol_unregister(struct nf_nat_protocol *proto)
555 write_lock_bh(&nf_nat_lock);
556 rcu_assign_pointer(nf_nat_protos[proto->protonum],
557 &nf_nat_unknown_protocol);
558 write_unlock_bh(&nf_nat_lock);
561 EXPORT_SYMBOL(nf_nat_protocol_unregister);
563 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
565 nf_nat_port_range_to_nfattr(struct sk_buff *skb,
566 const struct nf_nat_range *range)
568 NFA_PUT(skb, CTA_PROTONAT_PORT_MIN, sizeof(__be16),
569 &range->min.tcp.port);
570 NFA_PUT(skb, CTA_PROTONAT_PORT_MAX, sizeof(__be16),
571 &range->max.tcp.port);
578 EXPORT_SYMBOL_GPL(nf_nat_port_nfattr_to_range);
581 nf_nat_port_nfattr_to_range(struct nfattr *tb[], struct nf_nat_range *range)
585 /* we have to return whether we actually parsed something or not */
587 if (tb[CTA_PROTONAT_PORT_MIN-1]) {
589 range->min.tcp.port =
590 *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MIN-1]);
593 if (!tb[CTA_PROTONAT_PORT_MAX-1]) {
595 range->max.tcp.port = range->min.tcp.port;
598 range->max.tcp.port =
599 *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MAX-1]);
604 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nfattr);
607 static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
609 struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
610 struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
611 struct nf_conn *ct = old_nat->info.ct;
612 unsigned int srchash;
614 if (!(ct->status & IPS_NAT_DONE_MASK))
617 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
619 write_lock_bh(&nf_nat_lock);
620 list_replace(&old_nat->info.bysource, &new_nat->info.bysource);
621 new_nat->info.ct = ct;
622 write_unlock_bh(&nf_nat_lock);
625 struct nf_ct_ext_type nat_extend = {
626 .len = sizeof(struct nf_conn_nat),
627 .align = __alignof__(struct nf_conn_nat),
628 .move = nf_nat_move_storage,
630 .flags = NF_CT_EXT_F_PREALLOC,
633 static int __init nf_nat_init(void)
638 ret = nf_ct_extend_register(&nat_extend);
640 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
644 /* Leave them the same for the moment. */
645 nf_nat_htable_size = nf_conntrack_htable_size;
647 /* One vmalloc for both hash tables */
648 bysource = vmalloc(sizeof(struct list_head) * nf_nat_htable_size);
654 /* Sew in builtin protocols. */
655 write_lock_bh(&nf_nat_lock);
656 for (i = 0; i < MAX_IP_NAT_PROTO; i++)
657 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
658 rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
659 rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
660 rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
661 write_unlock_bh(&nf_nat_lock);
663 for (i = 0; i < nf_nat_htable_size; i++) {
664 INIT_LIST_HEAD(&bysource[i]);
667 /* FIXME: Man, this is a hack. <SIGH> */
668 NF_CT_ASSERT(rcu_dereference(nf_conntrack_destroyed) == NULL);
669 rcu_assign_pointer(nf_conntrack_destroyed, nf_nat_cleanup_conntrack);
671 /* Initialize fake conntrack so that NAT will skip it */
672 nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
674 l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
678 nf_ct_extend_unregister(&nat_extend);
682 /* Clear NAT section of all conntracks, in case we're loaded again. */
683 static int clean_nat(struct nf_conn *i, void *data)
685 struct nf_conn_nat *nat = nfct_nat(i);
689 memset(nat, 0, sizeof(nat));
690 i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
694 static void __exit nf_nat_cleanup(void)
696 nf_ct_iterate_cleanup(&clean_nat, NULL);
697 rcu_assign_pointer(nf_conntrack_destroyed, NULL);
700 nf_ct_l3proto_put(l3proto);
701 nf_ct_extend_unregister(&nat_extend);
704 MODULE_LICENSE("GPL");
706 module_init(nf_nat_init);
707 module_exit(nf_nat_cleanup);