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
);
100 write_unlock_bh(&nf_nat_lock
);
103 /* Is this tuple already taken? (not by us) */
105 nf_nat_used_tuple(const struct nf_conntrack_tuple
*tuple
,
106 const struct nf_conn
*ignored_conntrack
)
108 /* Conntrack tracking doesn't keep track of outgoing tuples; only
109 incoming ones. NAT means they don't have a fixed mapping,
110 so we invert the tuple and look for the incoming reply.
112 We could keep a separate hash if this proves too slow. */
113 struct nf_conntrack_tuple reply
;
115 nf_ct_invert_tuplepr(&reply
, tuple
);
116 return nf_conntrack_tuple_taken(&reply
, ignored_conntrack
);
118 EXPORT_SYMBOL(nf_nat_used_tuple
);
120 /* If we source map this tuple so reply looks like reply_tuple, will
121 * that meet the constraints of range. */
123 in_range(const struct nf_conntrack_tuple
*tuple
,
124 const struct nf_nat_range
*range
)
126 struct nf_nat_protocol
*proto
;
129 /* If we are supposed to map IPs, then we must be in the
130 range specified, otherwise let this drag us onto a new src IP. */
131 if (range
->flags
& IP_NAT_RANGE_MAP_IPS
) {
132 if (ntohl(tuple
->src
.u3
.ip
) < ntohl(range
->min_ip
) ||
133 ntohl(tuple
->src
.u3
.ip
) > ntohl(range
->max_ip
))
138 proto
= __nf_nat_proto_find(tuple
->dst
.protonum
);
139 if (!(range
->flags
& IP_NAT_RANGE_PROTO_SPECIFIED
) ||
140 proto
->in_range(tuple
, IP_NAT_MANIP_SRC
,
141 &range
->min
, &range
->max
))
149 same_src(const struct nf_conn
*ct
,
150 const struct nf_conntrack_tuple
*tuple
)
152 const struct nf_conntrack_tuple
*t
;
154 t
= &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
;
155 return (t
->dst
.protonum
== tuple
->dst
.protonum
&&
156 t
->src
.u3
.ip
== tuple
->src
.u3
.ip
&&
157 t
->src
.u
.all
== tuple
->src
.u
.all
);
160 /* Only called for SRC manip */
162 find_appropriate_src(const struct nf_conntrack_tuple
*tuple
,
163 struct nf_conntrack_tuple
*result
,
164 const struct nf_nat_range
*range
)
166 unsigned int h
= hash_by_src(tuple
);
167 struct nf_conn_nat
*nat
;
170 read_lock_bh(&nf_nat_lock
);
171 list_for_each_entry(nat
, &bysource
[h
], info
.bysource
) {
172 ct
= (struct nf_conn
*)((char *)nat
- offsetof(struct nf_conn
, data
));
173 if (same_src(ct
, tuple
)) {
174 /* Copy source part from reply tuple. */
175 nf_ct_invert_tuplepr(result
,
176 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
177 result
->dst
= tuple
->dst
;
179 if (in_range(result
, range
)) {
180 read_unlock_bh(&nf_nat_lock
);
185 read_unlock_bh(&nf_nat_lock
);
189 /* For [FUTURE] fragmentation handling, we want the least-used
190 src-ip/dst-ip/proto triple. Fairness doesn't come into it. Thus
191 if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
192 1-65535, we don't do pro-rata allocation based on ports; we choose
193 the ip with the lowest src-ip/dst-ip/proto usage.
196 find_best_ips_proto(struct nf_conntrack_tuple
*tuple
,
197 const struct nf_nat_range
*range
,
198 const struct nf_conn
*ct
,
199 enum nf_nat_manip_type maniptype
)
203 u_int32_t minip
, maxip
, j
;
205 /* No IP mapping? Do nothing. */
206 if (!(range
->flags
& IP_NAT_RANGE_MAP_IPS
))
209 if (maniptype
== IP_NAT_MANIP_SRC
)
210 var_ipp
= &tuple
->src
.u3
.ip
;
212 var_ipp
= &tuple
->dst
.u3
.ip
;
214 /* Fast path: only one choice. */
215 if (range
->min_ip
== range
->max_ip
) {
216 *var_ipp
= range
->min_ip
;
220 /* Hashing source and destination IPs gives a fairly even
221 * spread in practice (if there are a small number of IPs
222 * involved, there usually aren't that many connections
223 * anyway). The consistency means that servers see the same
224 * client coming from the same IP (some Internet Banking sites
225 * like this), even across reboots. */
226 minip
= ntohl(range
->min_ip
);
227 maxip
= ntohl(range
->max_ip
);
228 j
= jhash_2words((__force u32
)tuple
->src
.u3
.ip
,
229 (__force u32
)tuple
->dst
.u3
.ip
, 0);
230 *var_ipp
= htonl(minip
+ j
% (maxip
- minip
+ 1));
233 /* Manipulate the tuple into the range given. For NF_IP_POST_ROUTING,
234 * we change the source to map into the range. For NF_IP_PRE_ROUTING
235 * and NF_IP_LOCAL_OUT, we change the destination to map into the
236 * range. It might not be possible to get a unique tuple, but we try.
237 * At worst (or if we race), we will end up with a final duplicate in
238 * __ip_conntrack_confirm and drop the packet. */
240 get_unique_tuple(struct nf_conntrack_tuple
*tuple
,
241 const struct nf_conntrack_tuple
*orig_tuple
,
242 const struct nf_nat_range
*range
,
244 enum nf_nat_manip_type maniptype
)
246 struct nf_nat_protocol
*proto
;
248 /* 1) If this srcip/proto/src-proto-part is currently mapped,
249 and that same mapping gives a unique tuple within the given
252 This is only required for source (ie. NAT/masq) mappings.
253 So far, we don't do local source mappings, so multiple
254 manips not an issue. */
255 if (maniptype
== IP_NAT_MANIP_SRC
) {
256 if (find_appropriate_src(orig_tuple
, tuple
, range
)) {
257 DEBUGP("get_unique_tuple: Found current src map\n");
258 if (!(range
->flags
& IP_NAT_RANGE_PROTO_RANDOM
))
259 if (!nf_nat_used_tuple(tuple
, ct
))
264 /* 2) Select the least-used IP/proto combination in the given
266 *tuple
= *orig_tuple
;
267 find_best_ips_proto(tuple
, range
, ct
, maniptype
);
269 /* 3) The per-protocol part of the manip is made to map into
270 the range to make a unique tuple. */
273 proto
= __nf_nat_proto_find(orig_tuple
->dst
.protonum
);
275 /* Change protocol info to have some randomization */
276 if (range
->flags
& IP_NAT_RANGE_PROTO_RANDOM
) {
277 proto
->unique_tuple(tuple
, range
, maniptype
, ct
);
281 /* Only bother mapping if it's not already in range and unique */
282 if ((!(range
->flags
& IP_NAT_RANGE_PROTO_SPECIFIED
) ||
283 proto
->in_range(tuple
, maniptype
, &range
->min
, &range
->max
)) &&
284 !nf_nat_used_tuple(tuple
, ct
))
287 /* Last change: get protocol to try to obtain unique tuple. */
288 proto
->unique_tuple(tuple
, range
, maniptype
, ct
);
294 nf_nat_setup_info(struct nf_conn
*ct
,
295 const struct nf_nat_range
*range
,
296 unsigned int hooknum
)
298 struct nf_conntrack_tuple curr_tuple
, new_tuple
;
299 struct nf_conn_nat
*nat
= nfct_nat(ct
);
300 struct nf_nat_info
*info
= &nat
->info
;
301 int have_to_hash
= !(ct
->status
& IPS_NAT_DONE_MASK
);
302 enum nf_nat_manip_type maniptype
= HOOK2MANIP(hooknum
);
304 NF_CT_ASSERT(hooknum
== NF_IP_PRE_ROUTING
||
305 hooknum
== NF_IP_POST_ROUTING
||
306 hooknum
== NF_IP_LOCAL_IN
||
307 hooknum
== NF_IP_LOCAL_OUT
);
308 BUG_ON(nf_nat_initialized(ct
, maniptype
));
310 /* What we've got will look like inverse of reply. Normally
311 this is what is in the conntrack, except for prior
312 manipulations (future optimization: if num_manips == 0,
314 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
315 nf_ct_invert_tuplepr(&curr_tuple
,
316 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
318 get_unique_tuple(&new_tuple
, &curr_tuple
, range
, ct
, maniptype
);
320 if (!nf_ct_tuple_equal(&new_tuple
, &curr_tuple
)) {
321 struct nf_conntrack_tuple reply
;
323 /* Alter conntrack table so will recognize replies. */
324 nf_ct_invert_tuplepr(&reply
, &new_tuple
);
325 nf_conntrack_alter_reply(ct
, &reply
);
327 /* Non-atomic: we own this at the moment. */
328 if (maniptype
== IP_NAT_MANIP_SRC
)
329 ct
->status
|= IPS_SRC_NAT
;
331 ct
->status
|= IPS_DST_NAT
;
334 /* Place in source hash if this is the first time. */
336 unsigned int srchash
;
338 srchash
= hash_by_src(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
339 write_lock_bh(&nf_nat_lock
);
340 list_add(&info
->bysource
, &bysource
[srchash
]);
341 write_unlock_bh(&nf_nat_lock
);
345 if (maniptype
== IP_NAT_MANIP_DST
)
346 set_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
);
348 set_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
);
352 EXPORT_SYMBOL(nf_nat_setup_info
);
354 /* Returns true if succeeded. */
356 manip_pkt(u_int16_t proto
,
357 struct sk_buff
**pskb
,
358 unsigned int iphdroff
,
359 const struct nf_conntrack_tuple
*target
,
360 enum nf_nat_manip_type maniptype
)
363 struct nf_nat_protocol
*p
;
365 if (!skb_make_writable(pskb
, iphdroff
+ sizeof(*iph
)))
368 iph
= (void *)(*pskb
)->data
+ iphdroff
;
370 /* Manipulate protcol part. */
372 /* rcu_read_lock()ed by nf_hook_slow */
373 p
= __nf_nat_proto_find(proto
);
374 if (!p
->manip_pkt(pskb
, iphdroff
, target
, maniptype
))
377 iph
= (void *)(*pskb
)->data
+ iphdroff
;
379 if (maniptype
== IP_NAT_MANIP_SRC
) {
380 nf_csum_replace4(&iph
->check
, iph
->saddr
, target
->src
.u3
.ip
);
381 iph
->saddr
= target
->src
.u3
.ip
;
383 nf_csum_replace4(&iph
->check
, iph
->daddr
, target
->dst
.u3
.ip
);
384 iph
->daddr
= target
->dst
.u3
.ip
;
389 /* Do packet manipulations according to nf_nat_setup_info. */
390 unsigned int nf_nat_packet(struct nf_conn
*ct
,
391 enum ip_conntrack_info ctinfo
,
392 unsigned int hooknum
,
393 struct sk_buff
**pskb
)
395 enum ip_conntrack_dir dir
= CTINFO2DIR(ctinfo
);
396 unsigned long statusbit
;
397 enum nf_nat_manip_type mtype
= HOOK2MANIP(hooknum
);
399 if (mtype
== IP_NAT_MANIP_SRC
)
400 statusbit
= IPS_SRC_NAT
;
402 statusbit
= IPS_DST_NAT
;
404 /* Invert if this is reply dir. */
405 if (dir
== IP_CT_DIR_REPLY
)
406 statusbit
^= IPS_NAT_MASK
;
408 /* Non-atomic: these bits don't change. */
409 if (ct
->status
& statusbit
) {
410 struct nf_conntrack_tuple target
;
412 /* We are aiming to look like inverse of other direction. */
413 nf_ct_invert_tuplepr(&target
, &ct
->tuplehash
[!dir
].tuple
);
415 if (!manip_pkt(target
.dst
.protonum
, pskb
, 0, &target
, mtype
))
420 EXPORT_SYMBOL_GPL(nf_nat_packet
);
422 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
423 int nf_nat_icmp_reply_translation(struct nf_conn
*ct
,
424 enum ip_conntrack_info ctinfo
,
425 unsigned int hooknum
,
426 struct sk_buff
**pskb
)
432 struct nf_conntrack_l4proto
*l4proto
;
433 struct nf_conntrack_tuple inner
, target
;
434 int hdrlen
= (*pskb
)->nh
.iph
->ihl
* 4;
435 enum ip_conntrack_dir dir
= CTINFO2DIR(ctinfo
);
436 unsigned long statusbit
;
437 enum nf_nat_manip_type manip
= HOOK2MANIP(hooknum
);
439 if (!skb_make_writable(pskb
, hdrlen
+ sizeof(*inside
)))
442 inside
= (void *)(*pskb
)->data
+ (*pskb
)->nh
.iph
->ihl
*4;
444 /* We're actually going to mangle it beyond trivial checksum
445 adjustment, so make sure the current checksum is correct. */
446 if (nf_ip_checksum(*pskb
, hooknum
, hdrlen
, 0))
449 /* Must be RELATED */
450 NF_CT_ASSERT((*pskb
)->nfctinfo
== IP_CT_RELATED
||
451 (*pskb
)->nfctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
);
453 /* Redirects on non-null nats must be dropped, else they'll
454 start talking to each other without our translation, and be
456 if (inside
->icmp
.type
== ICMP_REDIRECT
) {
457 /* If NAT isn't finished, assume it and drop. */
458 if ((ct
->status
& IPS_NAT_DONE_MASK
) != IPS_NAT_DONE_MASK
)
461 if (ct
->status
& IPS_NAT_MASK
)
465 DEBUGP("icmp_reply_translation: translating error %p manp %u dir %s\n",
466 *pskb
, manip
, dir
== IP_CT_DIR_ORIGINAL
? "ORIG" : "REPLY");
468 /* rcu_read_lock()ed by nf_hook_slow */
469 l4proto
= __nf_ct_l4proto_find(PF_INET
, inside
->ip
.protocol
);
471 if (!nf_ct_get_tuple(*pskb
,
472 (*pskb
)->nh
.iph
->ihl
*4 + sizeof(struct icmphdr
),
473 (*pskb
)->nh
.iph
->ihl
*4 +
474 sizeof(struct icmphdr
) + inside
->ip
.ihl
*4,
477 &inner
, l3proto
, l4proto
))
480 /* Change inner back to look like incoming packet. We do the
481 opposite manip on this hook to normal, because it might not
482 pass all hooks (locally-generated ICMP). Consider incoming
483 packet: PREROUTING (DST manip), routing produces ICMP, goes
484 through POSTROUTING (which must correct the DST manip). */
485 if (!manip_pkt(inside
->ip
.protocol
, pskb
,
486 (*pskb
)->nh
.iph
->ihl
*4 + sizeof(inside
->icmp
),
487 &ct
->tuplehash
[!dir
].tuple
,
491 if ((*pskb
)->ip_summed
!= CHECKSUM_PARTIAL
) {
492 /* Reloading "inside" here since manip_pkt inner. */
493 inside
= (void *)(*pskb
)->data
+ (*pskb
)->nh
.iph
->ihl
*4;
494 inside
->icmp
.checksum
= 0;
495 inside
->icmp
.checksum
=
496 csum_fold(skb_checksum(*pskb
, hdrlen
,
497 (*pskb
)->len
- hdrlen
, 0));
500 /* Change outer to look the reply to an incoming packet
501 * (proto 0 means don't invert per-proto part). */
502 if (manip
== IP_NAT_MANIP_SRC
)
503 statusbit
= IPS_SRC_NAT
;
505 statusbit
= IPS_DST_NAT
;
507 /* Invert if this is reply dir. */
508 if (dir
== IP_CT_DIR_REPLY
)
509 statusbit
^= IPS_NAT_MASK
;
511 if (ct
->status
& statusbit
) {
512 nf_ct_invert_tuplepr(&target
, &ct
->tuplehash
[!dir
].tuple
);
513 if (!manip_pkt(0, pskb
, 0, &target
, manip
))
519 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation
);
521 /* Protocol registration. */
522 int nf_nat_protocol_register(struct nf_nat_protocol
*proto
)
526 write_lock_bh(&nf_nat_lock
);
527 if (nf_nat_protos
[proto
->protonum
] != &nf_nat_unknown_protocol
) {
531 rcu_assign_pointer(nf_nat_protos
[proto
->protonum
], proto
);
533 write_unlock_bh(&nf_nat_lock
);
536 EXPORT_SYMBOL(nf_nat_protocol_register
);
538 /* Noone stores the protocol anywhere; simply delete it. */
539 void nf_nat_protocol_unregister(struct nf_nat_protocol
*proto
)
541 write_lock_bh(&nf_nat_lock
);
542 rcu_assign_pointer(nf_nat_protos
[proto
->protonum
],
543 &nf_nat_unknown_protocol
);
544 write_unlock_bh(&nf_nat_lock
);
547 EXPORT_SYMBOL(nf_nat_protocol_unregister
);
549 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
551 nf_nat_port_range_to_nfattr(struct sk_buff
*skb
,
552 const struct nf_nat_range
*range
)
554 NFA_PUT(skb
, CTA_PROTONAT_PORT_MIN
, sizeof(__be16
),
555 &range
->min
.tcp
.port
);
556 NFA_PUT(skb
, CTA_PROTONAT_PORT_MAX
, sizeof(__be16
),
557 &range
->max
.tcp
.port
);
564 EXPORT_SYMBOL_GPL(nf_nat_port_nfattr_to_range
);
567 nf_nat_port_nfattr_to_range(struct nfattr
*tb
[], struct nf_nat_range
*range
)
571 /* we have to return whether we actually parsed something or not */
573 if (tb
[CTA_PROTONAT_PORT_MIN
-1]) {
575 range
->min
.tcp
.port
=
576 *(__be16
*)NFA_DATA(tb
[CTA_PROTONAT_PORT_MIN
-1]);
579 if (!tb
[CTA_PROTONAT_PORT_MAX
-1]) {
581 range
->max
.tcp
.port
= range
->min
.tcp
.port
;
584 range
->max
.tcp
.port
=
585 *(__be16
*)NFA_DATA(tb
[CTA_PROTONAT_PORT_MAX
-1]);
590 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nfattr
);
593 static int __init
nf_nat_init(void)
597 /* Leave them the same for the moment. */
598 nf_nat_htable_size
= nf_conntrack_htable_size
;
600 /* One vmalloc for both hash tables */
601 bysource
= vmalloc(sizeof(struct list_head
) * nf_nat_htable_size
);
605 /* Sew in builtin protocols. */
606 write_lock_bh(&nf_nat_lock
);
607 for (i
= 0; i
< MAX_IP_NAT_PROTO
; i
++)
608 rcu_assign_pointer(nf_nat_protos
[i
], &nf_nat_unknown_protocol
);
609 rcu_assign_pointer(nf_nat_protos
[IPPROTO_TCP
], &nf_nat_protocol_tcp
);
610 rcu_assign_pointer(nf_nat_protos
[IPPROTO_UDP
], &nf_nat_protocol_udp
);
611 rcu_assign_pointer(nf_nat_protos
[IPPROTO_ICMP
], &nf_nat_protocol_icmp
);
612 write_unlock_bh(&nf_nat_lock
);
614 for (i
= 0; i
< nf_nat_htable_size
; i
++) {
615 INIT_LIST_HEAD(&bysource
[i
]);
618 /* FIXME: Man, this is a hack. <SIGH> */
619 NF_CT_ASSERT(rcu_dereference(nf_conntrack_destroyed
) == NULL
);
620 rcu_assign_pointer(nf_conntrack_destroyed
, nf_nat_cleanup_conntrack
);
622 /* Initialize fake conntrack so that NAT will skip it */
623 nf_conntrack_untracked
.status
|= IPS_NAT_DONE_MASK
;
625 l3proto
= nf_ct_l3proto_find_get((u_int16_t
)AF_INET
);
629 /* Clear NAT section of all conntracks, in case we're loaded again. */
630 static int clean_nat(struct nf_conn
*i
, void *data
)
632 struct nf_conn_nat
*nat
= nfct_nat(i
);
636 memset(nat
, 0, sizeof(nat
));
637 i
->status
&= ~(IPS_NAT_MASK
| IPS_NAT_DONE_MASK
| IPS_SEQ_ADJUST
);
641 static void __exit
nf_nat_cleanup(void)
643 nf_ct_iterate_cleanup(&clean_nat
, NULL
);
644 rcu_assign_pointer(nf_conntrack_destroyed
, NULL
);
647 nf_ct_l3proto_put(l3proto
);
650 MODULE_LICENSE("GPL");
652 module_init(nf_nat_init
);
653 module_exit(nf_nat_cleanup
);