1 /* NAT for netfilter; shared with compatibility layer. */
3 /* (C) 1999-2001 Paul `Rusty' Russell
4 * (C) 2002-2004 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/netfilter_ipv4.h>
16 #include <linux/vmalloc.h>
17 #include <net/checksum.h>
20 #include <net/tcp.h> /* For tcp_prot in getorigdst */
21 #include <linux/icmp.h>
22 #include <linux/udp.h>
23 #include <linux/jhash.h>
25 #include <linux/netfilter_ipv4/ip_conntrack.h>
26 #include <linux/netfilter_ipv4/ip_conntrack_core.h>
27 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
28 #include <linux/netfilter_ipv4/ip_nat.h>
29 #include <linux/netfilter_ipv4/ip_nat_protocol.h>
30 #include <linux/netfilter_ipv4/ip_nat_core.h>
31 #include <linux/netfilter_ipv4/ip_nat_helper.h>
32 #include <linux/netfilter_ipv4/ip_conntrack_helper.h>
37 #define DEBUGP(format, args...)
40 DEFINE_RWLOCK(ip_nat_lock
);
42 /* Calculated at init based on memory size */
43 static unsigned int ip_nat_htable_size
;
45 static struct list_head
*bysource
;
47 #define MAX_IP_NAT_PROTO 256
48 static struct ip_nat_protocol
*ip_nat_protos
[MAX_IP_NAT_PROTO
];
50 static inline struct ip_nat_protocol
*
51 __ip_nat_proto_find(u_int8_t protonum
)
53 return rcu_dereference(ip_nat_protos
[protonum
]);
56 struct ip_nat_protocol
*
57 ip_nat_proto_find_get(u_int8_t protonum
)
59 struct ip_nat_protocol
*p
;
62 p
= __ip_nat_proto_find(protonum
);
63 if (!try_module_get(p
->me
))
64 p
= &ip_nat_unknown_protocol
;
69 EXPORT_SYMBOL_GPL(ip_nat_proto_find_get
);
72 ip_nat_proto_put(struct ip_nat_protocol
*p
)
76 EXPORT_SYMBOL_GPL(ip_nat_proto_put
);
78 /* We keep an extra hash for each conntrack, for fast searching. */
79 static inline unsigned int
80 hash_by_src(const struct ip_conntrack_tuple
*tuple
)
82 /* Original src, to ensure we map it consistently if poss. */
83 return jhash_3words((__force u32
)tuple
->src
.ip
, tuple
->src
.u
.all
,
84 tuple
->dst
.protonum
, 0) % ip_nat_htable_size
;
87 /* Noone using conntrack by the time this called. */
88 static void ip_nat_cleanup_conntrack(struct ip_conntrack
*conn
)
90 if (!(conn
->status
& IPS_NAT_DONE_MASK
))
93 write_lock_bh(&ip_nat_lock
);
94 list_del(&conn
->nat
.info
.bysource
);
95 write_unlock_bh(&ip_nat_lock
);
98 /* Is this tuple already taken? (not by us) */
100 ip_nat_used_tuple(const struct ip_conntrack_tuple
*tuple
,
101 const struct ip_conntrack
*ignored_conntrack
)
103 /* Conntrack tracking doesn't keep track of outgoing tuples; only
104 incoming ones. NAT means they don't have a fixed mapping,
105 so we invert the tuple and look for the incoming reply.
107 We could keep a separate hash if this proves too slow. */
108 struct ip_conntrack_tuple reply
;
110 invert_tuplepr(&reply
, tuple
);
111 return ip_conntrack_tuple_taken(&reply
, ignored_conntrack
);
113 EXPORT_SYMBOL(ip_nat_used_tuple
);
115 /* If we source map this tuple so reply looks like reply_tuple, will
116 * that meet the constraints of range. */
118 in_range(const struct ip_conntrack_tuple
*tuple
,
119 const struct ip_nat_range
*range
)
121 struct ip_nat_protocol
*proto
;
124 /* If we are supposed to map IPs, then we must be in the
125 range specified, otherwise let this drag us onto a new src IP. */
126 if (range
->flags
& IP_NAT_RANGE_MAP_IPS
) {
127 if (ntohl(tuple
->src
.ip
) < ntohl(range
->min_ip
)
128 || ntohl(tuple
->src
.ip
) > ntohl(range
->max_ip
))
133 proto
= __ip_nat_proto_find(tuple
->dst
.protonum
);
134 if (!(range
->flags
& IP_NAT_RANGE_PROTO_SPECIFIED
)
135 || proto
->in_range(tuple
, IP_NAT_MANIP_SRC
,
136 &range
->min
, &range
->max
))
144 same_src(const struct ip_conntrack
*ct
,
145 const struct ip_conntrack_tuple
*tuple
)
147 return (ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.dst
.protonum
148 == tuple
->dst
.protonum
149 && ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.ip
151 && ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
.src
.u
.all
152 == tuple
->src
.u
.all
);
155 /* Only called for SRC manip */
157 find_appropriate_src(const struct ip_conntrack_tuple
*tuple
,
158 struct ip_conntrack_tuple
*result
,
159 const struct ip_nat_range
*range
)
161 unsigned int h
= hash_by_src(tuple
);
162 struct ip_conntrack
*ct
;
164 read_lock_bh(&ip_nat_lock
);
165 list_for_each_entry(ct
, &bysource
[h
], nat
.info
.bysource
) {
166 if (same_src(ct
, tuple
)) {
167 /* Copy source part from reply tuple. */
168 invert_tuplepr(result
,
169 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
170 result
->dst
= tuple
->dst
;
172 if (in_range(result
, range
)) {
173 read_unlock_bh(&ip_nat_lock
);
178 read_unlock_bh(&ip_nat_lock
);
182 /* For [FUTURE] fragmentation handling, we want the least-used
183 src-ip/dst-ip/proto triple. Fairness doesn't come into it. Thus
184 if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
185 1-65535, we don't do pro-rata allocation based on ports; we choose
186 the ip with the lowest src-ip/dst-ip/proto usage.
189 find_best_ips_proto(struct ip_conntrack_tuple
*tuple
,
190 const struct ip_nat_range
*range
,
191 const struct ip_conntrack
*conntrack
,
192 enum ip_nat_manip_type maniptype
)
196 u_int32_t minip
, maxip
, j
;
198 /* No IP mapping? Do nothing. */
199 if (!(range
->flags
& IP_NAT_RANGE_MAP_IPS
))
202 if (maniptype
== IP_NAT_MANIP_SRC
)
203 var_ipp
= &tuple
->src
.ip
;
205 var_ipp
= &tuple
->dst
.ip
;
207 /* Fast path: only one choice. */
208 if (range
->min_ip
== range
->max_ip
) {
209 *var_ipp
= range
->min_ip
;
213 /* Hashing source and destination IPs gives a fairly even
214 * spread in practice (if there are a small number of IPs
215 * involved, there usually aren't that many connections
216 * anyway). The consistency means that servers see the same
217 * client coming from the same IP (some Internet Banking sites
218 * like this), even across reboots. */
219 minip
= ntohl(range
->min_ip
);
220 maxip
= ntohl(range
->max_ip
);
221 j
= jhash_2words((__force u32
)tuple
->src
.ip
, (__force u32
)tuple
->dst
.ip
, 0);
222 *var_ipp
= htonl(minip
+ j
% (maxip
- minip
+ 1));
225 /* Manipulate the tuple into the range given. For NF_IP_POST_ROUTING,
226 * we change the source to map into the range. For NF_IP_PRE_ROUTING
227 * and NF_IP_LOCAL_OUT, we change the destination to map into the
228 * range. It might not be possible to get a unique tuple, but we try.
229 * At worst (or if we race), we will end up with a final duplicate in
230 * __ip_conntrack_confirm and drop the packet. */
232 get_unique_tuple(struct ip_conntrack_tuple
*tuple
,
233 const struct ip_conntrack_tuple
*orig_tuple
,
234 const struct ip_nat_range
*range
,
235 struct ip_conntrack
*conntrack
,
236 enum ip_nat_manip_type maniptype
)
238 struct ip_nat_protocol
*proto
;
240 /* 1) If this srcip/proto/src-proto-part is currently mapped,
241 and that same mapping gives a unique tuple within the given
244 This is only required for source (ie. NAT/masq) mappings.
245 So far, we don't do local source mappings, so multiple
246 manips not an issue. */
247 if (maniptype
== IP_NAT_MANIP_SRC
) {
248 if (find_appropriate_src(orig_tuple
, tuple
, range
)) {
249 DEBUGP("get_unique_tuple: Found current src map\n");
250 if (!(range
->flags
& IP_NAT_RANGE_PROTO_RANDOM
))
251 if (!ip_nat_used_tuple(tuple
, conntrack
))
256 /* 2) Select the least-used IP/proto combination in the given
258 *tuple
= *orig_tuple
;
259 find_best_ips_proto(tuple
, range
, conntrack
, maniptype
);
261 /* 3) The per-protocol part of the manip is made to map into
262 the range to make a unique tuple. */
265 proto
= __ip_nat_proto_find(orig_tuple
->dst
.protonum
);
267 /* Change protocol info to have some randomization */
268 if (range
->flags
& IP_NAT_RANGE_PROTO_RANDOM
) {
269 proto
->unique_tuple(tuple
, range
, maniptype
, conntrack
);
273 /* Only bother mapping if it's not already in range and unique */
274 if ((!(range
->flags
& IP_NAT_RANGE_PROTO_SPECIFIED
)
275 || proto
->in_range(tuple
, maniptype
, &range
->min
, &range
->max
))
276 && !ip_nat_used_tuple(tuple
, conntrack
))
279 /* Last change: get protocol to try to obtain unique tuple. */
280 proto
->unique_tuple(tuple
, range
, maniptype
, conntrack
);
286 ip_nat_setup_info(struct ip_conntrack
*conntrack
,
287 const struct ip_nat_range
*range
,
288 unsigned int hooknum
)
290 struct ip_conntrack_tuple curr_tuple
, new_tuple
;
291 struct ip_nat_info
*info
= &conntrack
->nat
.info
;
292 int have_to_hash
= !(conntrack
->status
& IPS_NAT_DONE_MASK
);
293 enum ip_nat_manip_type maniptype
= HOOK2MANIP(hooknum
);
295 IP_NF_ASSERT(hooknum
== NF_IP_PRE_ROUTING
296 || hooknum
== NF_IP_POST_ROUTING
297 || hooknum
== NF_IP_LOCAL_IN
298 || hooknum
== NF_IP_LOCAL_OUT
);
299 BUG_ON(ip_nat_initialized(conntrack
, maniptype
));
301 /* What we've got will look like inverse of reply. Normally
302 this is what is in the conntrack, except for prior
303 manipulations (future optimization: if num_manips == 0,
305 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
306 invert_tuplepr(&curr_tuple
,
307 &conntrack
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
309 get_unique_tuple(&new_tuple
, &curr_tuple
, range
, conntrack
, maniptype
);
311 if (!ip_ct_tuple_equal(&new_tuple
, &curr_tuple
)) {
312 struct ip_conntrack_tuple reply
;
314 /* Alter conntrack table so will recognize replies. */
315 invert_tuplepr(&reply
, &new_tuple
);
316 ip_conntrack_alter_reply(conntrack
, &reply
);
318 /* Non-atomic: we own this at the moment. */
319 if (maniptype
== IP_NAT_MANIP_SRC
)
320 conntrack
->status
|= IPS_SRC_NAT
;
322 conntrack
->status
|= IPS_DST_NAT
;
325 /* Place in source hash if this is the first time. */
328 = hash_by_src(&conntrack
->tuplehash
[IP_CT_DIR_ORIGINAL
]
330 write_lock_bh(&ip_nat_lock
);
331 list_add(&info
->bysource
, &bysource
[srchash
]);
332 write_unlock_bh(&ip_nat_lock
);
336 if (maniptype
== IP_NAT_MANIP_DST
)
337 set_bit(IPS_DST_NAT_DONE_BIT
, &conntrack
->status
);
339 set_bit(IPS_SRC_NAT_DONE_BIT
, &conntrack
->status
);
343 EXPORT_SYMBOL(ip_nat_setup_info
);
345 /* Returns true if succeeded. */
347 manip_pkt(u_int16_t proto
,
348 struct sk_buff
**pskb
,
349 unsigned int iphdroff
,
350 const struct ip_conntrack_tuple
*target
,
351 enum ip_nat_manip_type maniptype
)
354 struct ip_nat_protocol
*p
;
356 if (!skb_make_writable(pskb
, iphdroff
+ sizeof(*iph
)))
359 iph
= (void *)(*pskb
)->data
+ iphdroff
;
361 /* Manipulate protcol part. */
363 /* rcu_read_lock()ed by nf_hook_slow */
364 p
= __ip_nat_proto_find(proto
);
365 if (!p
->manip_pkt(pskb
, iphdroff
, target
, maniptype
))
368 iph
= (void *)(*pskb
)->data
+ iphdroff
;
370 if (maniptype
== IP_NAT_MANIP_SRC
) {
371 nf_csum_replace4(&iph
->check
, iph
->saddr
, target
->src
.ip
);
372 iph
->saddr
= target
->src
.ip
;
374 nf_csum_replace4(&iph
->check
, iph
->daddr
, target
->dst
.ip
);
375 iph
->daddr
= target
->dst
.ip
;
380 /* Do packet manipulations according to ip_nat_setup_info. */
381 unsigned int ip_nat_packet(struct ip_conntrack
*ct
,
382 enum ip_conntrack_info ctinfo
,
383 unsigned int hooknum
,
384 struct sk_buff
**pskb
)
386 enum ip_conntrack_dir dir
= CTINFO2DIR(ctinfo
);
387 unsigned long statusbit
;
388 enum ip_nat_manip_type mtype
= HOOK2MANIP(hooknum
);
390 if (mtype
== IP_NAT_MANIP_SRC
)
391 statusbit
= IPS_SRC_NAT
;
393 statusbit
= IPS_DST_NAT
;
395 /* Invert if this is reply dir. */
396 if (dir
== IP_CT_DIR_REPLY
)
397 statusbit
^= IPS_NAT_MASK
;
399 /* Non-atomic: these bits don't change. */
400 if (ct
->status
& statusbit
) {
401 struct ip_conntrack_tuple target
;
403 /* We are aiming to look like inverse of other direction. */
404 invert_tuplepr(&target
, &ct
->tuplehash
[!dir
].tuple
);
406 if (!manip_pkt(target
.dst
.protonum
, pskb
, 0, &target
, mtype
))
411 EXPORT_SYMBOL_GPL(ip_nat_packet
);
413 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
414 int ip_nat_icmp_reply_translation(struct ip_conntrack
*ct
,
415 enum ip_conntrack_info ctinfo
,
416 unsigned int hooknum
,
417 struct sk_buff
**pskb
)
423 struct ip_conntrack_protocol
*proto
;
424 struct ip_conntrack_tuple inner
, target
;
425 int hdrlen
= (*pskb
)->nh
.iph
->ihl
* 4;
426 enum ip_conntrack_dir dir
= CTINFO2DIR(ctinfo
);
427 unsigned long statusbit
;
428 enum ip_nat_manip_type manip
= HOOK2MANIP(hooknum
);
430 if (!skb_make_writable(pskb
, hdrlen
+ sizeof(*inside
)))
433 inside
= (void *)(*pskb
)->data
+ (*pskb
)->nh
.iph
->ihl
*4;
435 /* We're actually going to mangle it beyond trivial checksum
436 adjustment, so make sure the current checksum is correct. */
437 if (nf_ip_checksum(*pskb
, hooknum
, hdrlen
, 0))
440 /* Must be RELATED */
441 IP_NF_ASSERT((*pskb
)->nfctinfo
== IP_CT_RELATED
||
442 (*pskb
)->nfctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
);
444 /* Redirects on non-null nats must be dropped, else they'll
445 start talking to each other without our translation, and be
447 if (inside
->icmp
.type
== ICMP_REDIRECT
) {
448 /* If NAT isn't finished, assume it and drop. */
449 if ((ct
->status
& IPS_NAT_DONE_MASK
) != IPS_NAT_DONE_MASK
)
452 if (ct
->status
& IPS_NAT_MASK
)
456 DEBUGP("icmp_reply_translation: translating error %p manp %u dir %s\n",
457 *pskb
, manip
, dir
== IP_CT_DIR_ORIGINAL
? "ORIG" : "REPLY");
459 /* rcu_read_lock()ed by nf_hook_slow */
460 proto
= __ip_conntrack_proto_find(inside
->ip
.protocol
);
461 if (!ip_ct_get_tuple(&inside
->ip
, *pskb
, (*pskb
)->nh
.iph
->ihl
*4 +
462 sizeof(struct icmphdr
) + inside
->ip
.ihl
*4,
466 /* Change inner back to look like incoming packet. We do the
467 opposite manip on this hook to normal, because it might not
468 pass all hooks (locally-generated ICMP). Consider incoming
469 packet: PREROUTING (DST manip), routing produces ICMP, goes
470 through POSTROUTING (which must correct the DST manip). */
471 if (!manip_pkt(inside
->ip
.protocol
, pskb
,
472 (*pskb
)->nh
.iph
->ihl
*4
473 + sizeof(inside
->icmp
),
474 &ct
->tuplehash
[!dir
].tuple
,
478 if ((*pskb
)->ip_summed
!= CHECKSUM_PARTIAL
) {
479 /* Reloading "inside" here since manip_pkt inner. */
480 inside
= (void *)(*pskb
)->data
+ (*pskb
)->nh
.iph
->ihl
*4;
481 inside
->icmp
.checksum
= 0;
482 inside
->icmp
.checksum
= csum_fold(skb_checksum(*pskb
, hdrlen
,
483 (*pskb
)->len
- hdrlen
,
487 /* Change outer to look the reply to an incoming packet
488 * (proto 0 means don't invert per-proto part). */
489 if (manip
== IP_NAT_MANIP_SRC
)
490 statusbit
= IPS_SRC_NAT
;
492 statusbit
= IPS_DST_NAT
;
494 /* Invert if this is reply dir. */
495 if (dir
== IP_CT_DIR_REPLY
)
496 statusbit
^= IPS_NAT_MASK
;
498 if (ct
->status
& statusbit
) {
499 invert_tuplepr(&target
, &ct
->tuplehash
[!dir
].tuple
);
500 if (!manip_pkt(0, pskb
, 0, &target
, manip
))
506 EXPORT_SYMBOL_GPL(ip_nat_icmp_reply_translation
);
508 /* Protocol registration. */
509 int ip_nat_protocol_register(struct ip_nat_protocol
*proto
)
513 write_lock_bh(&ip_nat_lock
);
514 if (ip_nat_protos
[proto
->protonum
] != &ip_nat_unknown_protocol
) {
518 rcu_assign_pointer(ip_nat_protos
[proto
->protonum
], proto
);
520 write_unlock_bh(&ip_nat_lock
);
523 EXPORT_SYMBOL(ip_nat_protocol_register
);
525 /* Noone stores the protocol anywhere; simply delete it. */
526 void ip_nat_protocol_unregister(struct ip_nat_protocol
*proto
)
528 write_lock_bh(&ip_nat_lock
);
529 rcu_assign_pointer(ip_nat_protos
[proto
->protonum
],
530 &ip_nat_unknown_protocol
);
531 write_unlock_bh(&ip_nat_lock
);
534 EXPORT_SYMBOL(ip_nat_protocol_unregister
);
536 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
537 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
539 ip_nat_port_range_to_nfattr(struct sk_buff
*skb
,
540 const struct ip_nat_range
*range
)
542 NFA_PUT(skb
, CTA_PROTONAT_PORT_MIN
, sizeof(__be16
),
543 &range
->min
.tcp
.port
);
544 NFA_PUT(skb
, CTA_PROTONAT_PORT_MAX
, sizeof(__be16
),
545 &range
->max
.tcp
.port
);
554 ip_nat_port_nfattr_to_range(struct nfattr
*tb
[], struct ip_nat_range
*range
)
558 /* we have to return whether we actually parsed something or not */
560 if (tb
[CTA_PROTONAT_PORT_MIN
-1]) {
562 range
->min
.tcp
.port
=
563 *(__be16
*)NFA_DATA(tb
[CTA_PROTONAT_PORT_MIN
-1]);
566 if (!tb
[CTA_PROTONAT_PORT_MAX
-1]) {
568 range
->max
.tcp
.port
= range
->min
.tcp
.port
;
571 range
->max
.tcp
.port
=
572 *(__be16
*)NFA_DATA(tb
[CTA_PROTONAT_PORT_MAX
-1]);
577 EXPORT_SYMBOL_GPL(ip_nat_port_nfattr_to_range
);
578 EXPORT_SYMBOL_GPL(ip_nat_port_range_to_nfattr
);
581 static int __init
ip_nat_init(void)
585 /* Leave them the same for the moment. */
586 ip_nat_htable_size
= ip_conntrack_htable_size
;
588 /* One vmalloc for both hash tables */
589 bysource
= vmalloc(sizeof(struct list_head
) * ip_nat_htable_size
);
593 /* Sew in builtin protocols. */
594 write_lock_bh(&ip_nat_lock
);
595 for (i
= 0; i
< MAX_IP_NAT_PROTO
; i
++)
596 rcu_assign_pointer(ip_nat_protos
[i
], &ip_nat_unknown_protocol
);
597 rcu_assign_pointer(ip_nat_protos
[IPPROTO_TCP
], &ip_nat_protocol_tcp
);
598 rcu_assign_pointer(ip_nat_protos
[IPPROTO_UDP
], &ip_nat_protocol_udp
);
599 rcu_assign_pointer(ip_nat_protos
[IPPROTO_ICMP
], &ip_nat_protocol_icmp
);
600 write_unlock_bh(&ip_nat_lock
);
602 for (i
= 0; i
< ip_nat_htable_size
; i
++) {
603 INIT_LIST_HEAD(&bysource
[i
]);
606 /* FIXME: Man, this is a hack. <SIGH> */
607 IP_NF_ASSERT(rcu_dereference(ip_conntrack_destroyed
) == NULL
);
608 rcu_assign_pointer(ip_conntrack_destroyed
, ip_nat_cleanup_conntrack
);
610 /* Initialize fake conntrack so that NAT will skip it */
611 ip_conntrack_untracked
.status
|= IPS_NAT_DONE_MASK
;
615 /* Clear NAT section of all conntracks, in case we're loaded again. */
616 static int clean_nat(struct ip_conntrack
*i
, void *data
)
618 memset(&i
->nat
, 0, sizeof(i
->nat
));
619 i
->status
&= ~(IPS_NAT_MASK
| IPS_NAT_DONE_MASK
| IPS_SEQ_ADJUST
);
623 static void __exit
ip_nat_cleanup(void)
625 ip_ct_iterate_cleanup(&clean_nat
, NULL
);
626 rcu_assign_pointer(ip_conntrack_destroyed
, NULL
);
631 MODULE_LICENSE("GPL");
633 module_init(ip_nat_init
);
634 module_exit(ip_nat_cleanup
);