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_SPINLOCK(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
)
82 /* Original src, to ensure we map it consistently if poss. */
83 hash
= jhash_3words((__force u32
)tuple
->src
.u3
.ip
,
84 (__force u32
)tuple
->src
.u
.all
,
85 tuple
->dst
.protonum
, 0);
86 return ((u64
)hash
* nf_nat_htable_size
) >> 32;
89 /* Is this tuple already taken? (not by us) */
91 nf_nat_used_tuple(const struct nf_conntrack_tuple
*tuple
,
92 const struct nf_conn
*ignored_conntrack
)
94 /* Conntrack tracking doesn't keep track of outgoing tuples; only
95 incoming ones. NAT means they don't have a fixed mapping,
96 so we invert the tuple and look for the incoming reply.
98 We could keep a separate hash if this proves too slow. */
99 struct nf_conntrack_tuple reply
;
101 nf_ct_invert_tuplepr(&reply
, tuple
);
102 return nf_conntrack_tuple_taken(&reply
, ignored_conntrack
);
104 EXPORT_SYMBOL(nf_nat_used_tuple
);
106 /* If we source map this tuple so reply looks like reply_tuple, will
107 * that meet the constraints of range. */
109 in_range(const struct nf_conntrack_tuple
*tuple
,
110 const struct nf_nat_range
*range
)
112 const struct nf_nat_protocol
*proto
;
115 /* If we are supposed to map IPs, then we must be in the
116 range specified, otherwise let this drag us onto a new src IP. */
117 if (range
->flags
& IP_NAT_RANGE_MAP_IPS
) {
118 if (ntohl(tuple
->src
.u3
.ip
) < ntohl(range
->min_ip
) ||
119 ntohl(tuple
->src
.u3
.ip
) > ntohl(range
->max_ip
))
124 proto
= __nf_nat_proto_find(tuple
->dst
.protonum
);
125 if (!(range
->flags
& IP_NAT_RANGE_PROTO_SPECIFIED
) ||
126 proto
->in_range(tuple
, IP_NAT_MANIP_SRC
,
127 &range
->min
, &range
->max
))
135 same_src(const struct nf_conn
*ct
,
136 const struct nf_conntrack_tuple
*tuple
)
138 const struct nf_conntrack_tuple
*t
;
140 t
= &ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
;
141 return (t
->dst
.protonum
== tuple
->dst
.protonum
&&
142 t
->src
.u3
.ip
== tuple
->src
.u3
.ip
&&
143 t
->src
.u
.all
== tuple
->src
.u
.all
);
146 /* Only called for SRC manip */
148 find_appropriate_src(const struct nf_conntrack_tuple
*tuple
,
149 struct nf_conntrack_tuple
*result
,
150 const struct nf_nat_range
*range
)
152 unsigned int h
= hash_by_src(tuple
);
153 const struct nf_conn_nat
*nat
;
154 const struct nf_conn
*ct
;
155 const struct hlist_node
*n
;
158 hlist_for_each_entry_rcu(nat
, n
, &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
)) {
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 j
= ((u64
)j
* (maxip
- minip
+ 1)) >> 32;
218 *var_ipp
= htonl(minip
+ j
);
221 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
222 * we change the source to map into the range. For NF_INET_PRE_ROUTING
223 * and NF_INET_LOCAL_OUT, we change the destination to map into the
224 * range. It might not be possible to get a unique tuple, but we try.
225 * At worst (or if we race), we will end up with a final duplicate in
226 * __ip_conntrack_confirm and drop the packet. */
228 get_unique_tuple(struct nf_conntrack_tuple
*tuple
,
229 const struct nf_conntrack_tuple
*orig_tuple
,
230 const struct nf_nat_range
*range
,
232 enum nf_nat_manip_type maniptype
)
234 const struct nf_nat_protocol
*proto
;
236 /* 1) If this srcip/proto/src-proto-part is currently mapped,
237 and that same mapping gives a unique tuple within the given
240 This is only required for source (ie. NAT/masq) mappings.
241 So far, we don't do local source mappings, so multiple
242 manips not an issue. */
243 if (maniptype
== IP_NAT_MANIP_SRC
) {
244 if (find_appropriate_src(orig_tuple
, tuple
, range
)) {
245 pr_debug("get_unique_tuple: Found current src map\n");
246 if (!(range
->flags
& IP_NAT_RANGE_PROTO_RANDOM
))
247 if (!nf_nat_used_tuple(tuple
, ct
))
252 /* 2) Select the least-used IP/proto combination in the given
254 *tuple
= *orig_tuple
;
255 find_best_ips_proto(tuple
, range
, ct
, maniptype
);
257 /* 3) The per-protocol part of the manip is made to map into
258 the range to make a unique tuple. */
261 proto
= __nf_nat_proto_find(orig_tuple
->dst
.protonum
);
263 /* Change protocol info to have some randomization */
264 if (range
->flags
& IP_NAT_RANGE_PROTO_RANDOM
) {
265 proto
->unique_tuple(tuple
, range
, maniptype
, ct
);
269 /* Only bother mapping if it's not already in range and unique */
270 if ((!(range
->flags
& IP_NAT_RANGE_PROTO_SPECIFIED
) ||
271 proto
->in_range(tuple
, maniptype
, &range
->min
, &range
->max
)) &&
272 !nf_nat_used_tuple(tuple
, ct
))
275 /* Last change: get protocol to try to obtain unique tuple. */
276 proto
->unique_tuple(tuple
, range
, maniptype
, ct
);
282 nf_nat_setup_info(struct nf_conn
*ct
,
283 const struct nf_nat_range
*range
,
284 enum nf_nat_manip_type maniptype
)
286 struct nf_conntrack_tuple curr_tuple
, new_tuple
;
287 struct nf_conn_nat
*nat
;
288 int have_to_hash
= !(ct
->status
& IPS_NAT_DONE_MASK
);
290 /* nat helper or nfctnetlink also setup binding */
293 nat
= nf_ct_ext_add(ct
, NF_CT_EXT_NAT
, GFP_ATOMIC
);
295 pr_debug("failed to add NAT extension\n");
300 NF_CT_ASSERT(maniptype
== IP_NAT_MANIP_SRC
||
301 maniptype
== IP_NAT_MANIP_DST
);
302 BUG_ON(nf_nat_initialized(ct
, maniptype
));
304 /* What we've got will look like inverse of reply. Normally
305 this is what is in the conntrack, except for prior
306 manipulations (future optimization: if num_manips == 0,
308 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
309 nf_ct_invert_tuplepr(&curr_tuple
,
310 &ct
->tuplehash
[IP_CT_DIR_REPLY
].tuple
);
312 get_unique_tuple(&new_tuple
, &curr_tuple
, range
, ct
, maniptype
);
314 if (!nf_ct_tuple_equal(&new_tuple
, &curr_tuple
)) {
315 struct nf_conntrack_tuple reply
;
317 /* Alter conntrack table so will recognize replies. */
318 nf_ct_invert_tuplepr(&reply
, &new_tuple
);
319 nf_conntrack_alter_reply(ct
, &reply
);
321 /* Non-atomic: we own this at the moment. */
322 if (maniptype
== IP_NAT_MANIP_SRC
)
323 ct
->status
|= IPS_SRC_NAT
;
325 ct
->status
|= IPS_DST_NAT
;
328 /* Place in source hash if this is the first time. */
330 unsigned int srchash
;
332 srchash
= hash_by_src(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
333 spin_lock_bh(&nf_nat_lock
);
334 /* nf_conntrack_alter_reply might re-allocate exntension aera */
337 hlist_add_head_rcu(&nat
->bysource
, &bysource
[srchash
]);
338 spin_unlock_bh(&nf_nat_lock
);
342 if (maniptype
== IP_NAT_MANIP_DST
)
343 set_bit(IPS_DST_NAT_DONE_BIT
, &ct
->status
);
345 set_bit(IPS_SRC_NAT_DONE_BIT
, &ct
->status
);
349 EXPORT_SYMBOL(nf_nat_setup_info
);
351 /* Returns true if succeeded. */
353 manip_pkt(u_int16_t proto
,
355 unsigned int iphdroff
,
356 const struct nf_conntrack_tuple
*target
,
357 enum nf_nat_manip_type maniptype
)
360 const struct nf_nat_protocol
*p
;
362 if (!skb_make_writable(skb
, iphdroff
+ sizeof(*iph
)))
365 iph
= (void *)skb
->data
+ iphdroff
;
367 /* Manipulate protcol part. */
369 /* rcu_read_lock()ed by nf_hook_slow */
370 p
= __nf_nat_proto_find(proto
);
371 if (!p
->manip_pkt(skb
, iphdroff
, target
, maniptype
))
374 iph
= (void *)skb
->data
+ iphdroff
;
376 if (maniptype
== IP_NAT_MANIP_SRC
) {
377 csum_replace4(&iph
->check
, iph
->saddr
, target
->src
.u3
.ip
);
378 iph
->saddr
= target
->src
.u3
.ip
;
380 csum_replace4(&iph
->check
, iph
->daddr
, target
->dst
.u3
.ip
);
381 iph
->daddr
= target
->dst
.u3
.ip
;
386 /* Do packet manipulations according to nf_nat_setup_info. */
387 unsigned int nf_nat_packet(struct nf_conn
*ct
,
388 enum ip_conntrack_info ctinfo
,
389 unsigned int hooknum
,
392 enum ip_conntrack_dir dir
= CTINFO2DIR(ctinfo
);
393 unsigned long statusbit
;
394 enum nf_nat_manip_type mtype
= HOOK2MANIP(hooknum
);
396 if (mtype
== IP_NAT_MANIP_SRC
)
397 statusbit
= IPS_SRC_NAT
;
399 statusbit
= IPS_DST_NAT
;
401 /* Invert if this is reply dir. */
402 if (dir
== IP_CT_DIR_REPLY
)
403 statusbit
^= IPS_NAT_MASK
;
405 /* Non-atomic: these bits don't change. */
406 if (ct
->status
& statusbit
) {
407 struct nf_conntrack_tuple target
;
409 /* We are aiming to look like inverse of other direction. */
410 nf_ct_invert_tuplepr(&target
, &ct
->tuplehash
[!dir
].tuple
);
412 if (!manip_pkt(target
.dst
.protonum
, skb
, 0, &target
, mtype
))
417 EXPORT_SYMBOL_GPL(nf_nat_packet
);
419 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
420 int nf_nat_icmp_reply_translation(struct nf_conn
*ct
,
421 enum ip_conntrack_info ctinfo
,
422 unsigned int hooknum
,
429 const struct nf_conntrack_l4proto
*l4proto
;
430 struct nf_conntrack_tuple inner
, target
;
431 int hdrlen
= ip_hdrlen(skb
);
432 enum ip_conntrack_dir dir
= CTINFO2DIR(ctinfo
);
433 unsigned long statusbit
;
434 enum nf_nat_manip_type manip
= HOOK2MANIP(hooknum
);
436 if (!skb_make_writable(skb
, hdrlen
+ sizeof(*inside
)))
439 inside
= (void *)skb
->data
+ ip_hdrlen(skb
);
441 /* We're actually going to mangle it beyond trivial checksum
442 adjustment, so make sure the current checksum is correct. */
443 if (nf_ip_checksum(skb
, hooknum
, hdrlen
, 0))
446 /* Must be RELATED */
447 NF_CT_ASSERT(skb
->nfctinfo
== IP_CT_RELATED
||
448 skb
->nfctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
);
450 /* Redirects on non-null nats must be dropped, else they'll
451 start talking to each other without our translation, and be
453 if (inside
->icmp
.type
== ICMP_REDIRECT
) {
454 /* If NAT isn't finished, assume it and drop. */
455 if ((ct
->status
& IPS_NAT_DONE_MASK
) != IPS_NAT_DONE_MASK
)
458 if (ct
->status
& IPS_NAT_MASK
)
462 pr_debug("icmp_reply_translation: translating error %p manip %u "
463 "dir %s\n", skb
, manip
,
464 dir
== IP_CT_DIR_ORIGINAL
? "ORIG" : "REPLY");
466 /* rcu_read_lock()ed by nf_hook_slow */
467 l4proto
= __nf_ct_l4proto_find(PF_INET
, inside
->ip
.protocol
);
469 if (!nf_ct_get_tuple(skb
,
470 ip_hdrlen(skb
) + sizeof(struct icmphdr
),
472 sizeof(struct icmphdr
) + inside
->ip
.ihl
* 4),
475 &inner
, l3proto
, l4proto
))
478 /* Change inner back to look like incoming packet. We do the
479 opposite manip on this hook to normal, because it might not
480 pass all hooks (locally-generated ICMP). Consider incoming
481 packet: PREROUTING (DST manip), routing produces ICMP, goes
482 through POSTROUTING (which must correct the DST manip). */
483 if (!manip_pkt(inside
->ip
.protocol
, skb
,
484 ip_hdrlen(skb
) + sizeof(inside
->icmp
),
485 &ct
->tuplehash
[!dir
].tuple
,
489 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
490 /* Reloading "inside" here since manip_pkt inner. */
491 inside
= (void *)skb
->data
+ ip_hdrlen(skb
);
492 inside
->icmp
.checksum
= 0;
493 inside
->icmp
.checksum
=
494 csum_fold(skb_checksum(skb
, hdrlen
,
495 skb
->len
- hdrlen
, 0));
498 /* Change outer to look the reply to an incoming packet
499 * (proto 0 means don't invert per-proto part). */
500 if (manip
== IP_NAT_MANIP_SRC
)
501 statusbit
= IPS_SRC_NAT
;
503 statusbit
= IPS_DST_NAT
;
505 /* Invert if this is reply dir. */
506 if (dir
== IP_CT_DIR_REPLY
)
507 statusbit
^= IPS_NAT_MASK
;
509 if (ct
->status
& statusbit
) {
510 nf_ct_invert_tuplepr(&target
, &ct
->tuplehash
[!dir
].tuple
);
511 if (!manip_pkt(0, skb
, 0, &target
, manip
))
517 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation
);
519 /* Protocol registration. */
520 int nf_nat_protocol_register(const struct nf_nat_protocol
*proto
)
524 spin_lock_bh(&nf_nat_lock
);
525 if (nf_nat_protos
[proto
->protonum
] != &nf_nat_unknown_protocol
) {
529 rcu_assign_pointer(nf_nat_protos
[proto
->protonum
], proto
);
531 spin_unlock_bh(&nf_nat_lock
);
534 EXPORT_SYMBOL(nf_nat_protocol_register
);
536 /* Noone stores the protocol anywhere; simply delete it. */
537 void nf_nat_protocol_unregister(const struct nf_nat_protocol
*proto
)
539 spin_lock_bh(&nf_nat_lock
);
540 rcu_assign_pointer(nf_nat_protos
[proto
->protonum
],
541 &nf_nat_unknown_protocol
);
542 spin_unlock_bh(&nf_nat_lock
);
545 EXPORT_SYMBOL(nf_nat_protocol_unregister
);
547 /* Noone using conntrack by the time this called. */
548 static void nf_nat_cleanup_conntrack(struct nf_conn
*ct
)
550 struct nf_conn_nat
*nat
= nf_ct_ext_find(ct
, NF_CT_EXT_NAT
);
552 if (nat
== NULL
|| nat
->ct
== NULL
)
555 NF_CT_ASSERT(nat
->ct
->status
& IPS_NAT_DONE_MASK
);
557 spin_lock_bh(&nf_nat_lock
);
558 hlist_del_rcu(&nat
->bysource
);
560 spin_unlock_bh(&nf_nat_lock
);
563 static void nf_nat_move_storage(void *new, void *old
)
565 struct nf_conn_nat
*new_nat
= new;
566 struct nf_conn_nat
*old_nat
= old
;
567 struct nf_conn
*ct
= old_nat
->ct
;
569 if (!ct
|| !(ct
->status
& IPS_NAT_DONE_MASK
))
572 spin_lock_bh(&nf_nat_lock
);
573 hlist_replace_rcu(&old_nat
->bysource
, &new_nat
->bysource
);
575 spin_unlock_bh(&nf_nat_lock
);
578 static struct nf_ct_ext_type nat_extend __read_mostly
= {
579 .len
= sizeof(struct nf_conn_nat
),
580 .align
= __alignof__(struct nf_conn_nat
),
581 .destroy
= nf_nat_cleanup_conntrack
,
582 .move
= nf_nat_move_storage
,
584 .flags
= NF_CT_EXT_F_PREALLOC
,
587 static int __init
nf_nat_init(void)
592 ret
= nf_ct_extend_register(&nat_extend
);
594 printk(KERN_ERR
"nf_nat_core: Unable to register extension\n");
598 /* Leave them the same for the moment. */
599 nf_nat_htable_size
= nf_conntrack_htable_size
;
601 bysource
= nf_ct_alloc_hashtable(&nf_nat_htable_size
,
608 /* Sew in builtin protocols. */
609 spin_lock_bh(&nf_nat_lock
);
610 for (i
= 0; i
< MAX_IP_NAT_PROTO
; i
++)
611 rcu_assign_pointer(nf_nat_protos
[i
], &nf_nat_unknown_protocol
);
612 rcu_assign_pointer(nf_nat_protos
[IPPROTO_TCP
], &nf_nat_protocol_tcp
);
613 rcu_assign_pointer(nf_nat_protos
[IPPROTO_UDP
], &nf_nat_protocol_udp
);
614 rcu_assign_pointer(nf_nat_protos
[IPPROTO_ICMP
], &nf_nat_protocol_icmp
);
615 spin_unlock_bh(&nf_nat_lock
);
617 /* Initialize fake conntrack so that NAT will skip it */
618 nf_conntrack_untracked
.status
|= IPS_NAT_DONE_MASK
;
620 l3proto
= nf_ct_l3proto_find_get((u_int16_t
)AF_INET
);
622 BUG_ON(nf_nat_seq_adjust_hook
!= NULL
);
623 rcu_assign_pointer(nf_nat_seq_adjust_hook
, nf_nat_seq_adjust
);
627 nf_ct_extend_unregister(&nat_extend
);
631 /* Clear NAT section of all conntracks, in case we're loaded again. */
632 static int clean_nat(struct nf_conn
*i
, void *data
)
634 struct nf_conn_nat
*nat
= nfct_nat(i
);
638 memset(nat
, 0, sizeof(*nat
));
639 i
->status
&= ~(IPS_NAT_MASK
| IPS_NAT_DONE_MASK
| IPS_SEQ_ADJUST
);
643 static void __exit
nf_nat_cleanup(void)
645 nf_ct_iterate_cleanup(&clean_nat
, NULL
);
647 nf_ct_free_hashtable(bysource
, nf_nat_vmalloced
, nf_nat_htable_size
);
648 nf_ct_l3proto_put(l3proto
);
649 nf_ct_extend_unregister(&nat_extend
);
650 rcu_assign_pointer(nf_nat_seq_adjust_hook
, NULL
);
654 MODULE_LICENSE("GPL");
656 module_init(nf_nat_init
);
657 module_exit(nf_nat_cleanup
);