Commit | Line | Data |
---|---|---|
5b1158e9 JK |
1 | /* NAT for netfilter; shared with compatibility layer. */ |
2 | ||
3 | /* (C) 1999-2001 Paul `Rusty' Russell | |
4 | * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org> | |
5 | * | |
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. | |
9 | */ | |
10 | ||
11 | #include <linux/module.h> | |
12 | #include <linux/types.h> | |
13 | #include <linux/timer.h> | |
14 | #include <linux/skbuff.h> | |
5a0e3ad6 | 15 | #include <linux/gfp.h> |
5b1158e9 JK |
16 | #include <net/checksum.h> |
17 | #include <net/icmp.h> | |
18 | #include <net/ip.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> | |
23 | ||
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> | |
5d0aa2cc | 34 | #include <net/netfilter/nf_conntrack_zones.h> |
5b1158e9 | 35 | |
02502f62 | 36 | static DEFINE_SPINLOCK(nf_nat_lock); |
5b1158e9 | 37 | |
ce4b1ceb | 38 | static struct nf_conntrack_l3proto *l3proto __read_mostly; |
5b1158e9 | 39 | |
5b1158e9 | 40 | #define MAX_IP_NAT_PROTO 256 |
ce4b1ceb PM |
41 | static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO] |
42 | __read_mostly; | |
5b1158e9 | 43 | |
2b628a08 | 44 | static inline const struct nf_nat_protocol * |
5b1158e9 JK |
45 | __nf_nat_proto_find(u_int8_t protonum) |
46 | { | |
e22a0548 | 47 | return rcu_dereference(nf_nat_protos[protonum]); |
5b1158e9 JK |
48 | } |
49 | ||
2b628a08 | 50 | const struct nf_nat_protocol * |
5b1158e9 JK |
51 | nf_nat_proto_find_get(u_int8_t protonum) |
52 | { | |
2b628a08 | 53 | const struct nf_nat_protocol *p; |
5b1158e9 | 54 | |
e22a0548 | 55 | rcu_read_lock(); |
5b1158e9 JK |
56 | p = __nf_nat_proto_find(protonum); |
57 | if (!try_module_get(p->me)) | |
58 | p = &nf_nat_unknown_protocol; | |
e22a0548 | 59 | rcu_read_unlock(); |
5b1158e9 JK |
60 | |
61 | return p; | |
62 | } | |
63 | EXPORT_SYMBOL_GPL(nf_nat_proto_find_get); | |
64 | ||
65 | void | |
2b628a08 | 66 | nf_nat_proto_put(const struct nf_nat_protocol *p) |
5b1158e9 JK |
67 | { |
68 | module_put(p->me); | |
69 | } | |
70 | EXPORT_SYMBOL_GPL(nf_nat_proto_put); | |
71 | ||
72 | /* We keep an extra hash for each conntrack, for fast searching. */ | |
73 | static inline unsigned int | |
5d0aa2cc PM |
74 | hash_by_src(const struct net *net, u16 zone, |
75 | const struct nf_conntrack_tuple *tuple) | |
5b1158e9 | 76 | { |
34498825 PM |
77 | unsigned int hash; |
78 | ||
5b1158e9 | 79 | /* Original src, to ensure we map it consistently if poss. */ |
34498825 | 80 | hash = jhash_3words((__force u32)tuple->src.u3.ip, |
5d0aa2cc | 81 | (__force u32)tuple->src.u.all ^ zone, |
34498825 | 82 | tuple->dst.protonum, 0); |
d696c7bd | 83 | return ((u64)hash * net->ipv4.nat_htable_size) >> 32; |
5b1158e9 JK |
84 | } |
85 | ||
5b1158e9 JK |
86 | /* Is this tuple already taken? (not by us) */ |
87 | int | |
88 | nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple, | |
89 | const struct nf_conn *ignored_conntrack) | |
90 | { | |
91 | /* Conntrack tracking doesn't keep track of outgoing tuples; only | |
92 | incoming ones. NAT means they don't have a fixed mapping, | |
93 | so we invert the tuple and look for the incoming reply. | |
94 | ||
95 | We could keep a separate hash if this proves too slow. */ | |
96 | struct nf_conntrack_tuple reply; | |
97 | ||
98 | nf_ct_invert_tuplepr(&reply, tuple); | |
99 | return nf_conntrack_tuple_taken(&reply, ignored_conntrack); | |
100 | } | |
101 | EXPORT_SYMBOL(nf_nat_used_tuple); | |
102 | ||
103 | /* If we source map this tuple so reply looks like reply_tuple, will | |
104 | * that meet the constraints of range. */ | |
105 | static int | |
106 | in_range(const struct nf_conntrack_tuple *tuple, | |
107 | const struct nf_nat_range *range) | |
108 | { | |
2b628a08 | 109 | const struct nf_nat_protocol *proto; |
e22a0548 | 110 | int ret = 0; |
5b1158e9 | 111 | |
5b1158e9 JK |
112 | /* If we are supposed to map IPs, then we must be in the |
113 | range specified, otherwise let this drag us onto a new src IP. */ | |
114 | if (range->flags & IP_NAT_RANGE_MAP_IPS) { | |
115 | if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) || | |
116 | ntohl(tuple->src.u3.ip) > ntohl(range->max_ip)) | |
117 | return 0; | |
118 | } | |
119 | ||
e22a0548 PM |
120 | rcu_read_lock(); |
121 | proto = __nf_nat_proto_find(tuple->dst.protonum); | |
5b1158e9 JK |
122 | if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) || |
123 | proto->in_range(tuple, IP_NAT_MANIP_SRC, | |
124 | &range->min, &range->max)) | |
e22a0548 PM |
125 | ret = 1; |
126 | rcu_read_unlock(); | |
5b1158e9 | 127 | |
e22a0548 | 128 | return ret; |
5b1158e9 JK |
129 | } |
130 | ||
131 | static inline int | |
132 | same_src(const struct nf_conn *ct, | |
133 | const struct nf_conntrack_tuple *tuple) | |
134 | { | |
135 | const struct nf_conntrack_tuple *t; | |
136 | ||
137 | t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple; | |
138 | return (t->dst.protonum == tuple->dst.protonum && | |
139 | t->src.u3.ip == tuple->src.u3.ip && | |
140 | t->src.u.all == tuple->src.u.all); | |
141 | } | |
142 | ||
143 | /* Only called for SRC manip */ | |
144 | static int | |
5d0aa2cc | 145 | find_appropriate_src(struct net *net, u16 zone, |
0c4c9288 | 146 | const struct nf_conntrack_tuple *tuple, |
5b1158e9 JK |
147 | struct nf_conntrack_tuple *result, |
148 | const struct nf_nat_range *range) | |
149 | { | |
5d0aa2cc | 150 | unsigned int h = hash_by_src(net, zone, tuple); |
72b72949 JE |
151 | const struct nf_conn_nat *nat; |
152 | const struct nf_conn *ct; | |
153 | const struct hlist_node *n; | |
5b1158e9 | 154 | |
4d354c57 | 155 | rcu_read_lock(); |
0c4c9288 | 156 | hlist_for_each_entry_rcu(nat, n, &net->ipv4.nat_bysource[h], bysource) { |
b6b84d4a | 157 | ct = nat->ct; |
5d0aa2cc | 158 | if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) { |
5b1158e9 JK |
159 | /* Copy source part from reply tuple. */ |
160 | nf_ct_invert_tuplepr(result, | |
161 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
162 | result->dst = tuple->dst; | |
163 | ||
164 | if (in_range(result, range)) { | |
4d354c57 | 165 | rcu_read_unlock(); |
5b1158e9 JK |
166 | return 1; |
167 | } | |
168 | } | |
169 | } | |
4d354c57 | 170 | rcu_read_unlock(); |
5b1158e9 JK |
171 | return 0; |
172 | } | |
173 | ||
174 | /* For [FUTURE] fragmentation handling, we want the least-used | |
175 | src-ip/dst-ip/proto triple. Fairness doesn't come into it. Thus | |
176 | if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports | |
177 | 1-65535, we don't do pro-rata allocation based on ports; we choose | |
178 | the ip with the lowest src-ip/dst-ip/proto usage. | |
179 | */ | |
180 | static void | |
5d0aa2cc | 181 | find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple, |
5b1158e9 JK |
182 | const struct nf_nat_range *range, |
183 | const struct nf_conn *ct, | |
184 | enum nf_nat_manip_type maniptype) | |
185 | { | |
186 | __be32 *var_ipp; | |
187 | /* Host order */ | |
188 | u_int32_t minip, maxip, j; | |
189 | ||
190 | /* No IP mapping? Do nothing. */ | |
191 | if (!(range->flags & IP_NAT_RANGE_MAP_IPS)) | |
192 | return; | |
193 | ||
194 | if (maniptype == IP_NAT_MANIP_SRC) | |
195 | var_ipp = &tuple->src.u3.ip; | |
196 | else | |
197 | var_ipp = &tuple->dst.u3.ip; | |
198 | ||
199 | /* Fast path: only one choice. */ | |
200 | if (range->min_ip == range->max_ip) { | |
201 | *var_ipp = range->min_ip; | |
202 | return; | |
203 | } | |
204 | ||
205 | /* Hashing source and destination IPs gives a fairly even | |
206 | * spread in practice (if there are a small number of IPs | |
207 | * involved, there usually aren't that many connections | |
208 | * anyway). The consistency means that servers see the same | |
209 | * client coming from the same IP (some Internet Banking sites | |
210 | * like this), even across reboots. */ | |
211 | minip = ntohl(range->min_ip); | |
212 | maxip = ntohl(range->max_ip); | |
213 | j = jhash_2words((__force u32)tuple->src.u3.ip, | |
98d500d6 | 214 | range->flags & IP_NAT_RANGE_PERSISTENT ? |
5d0aa2cc | 215 | 0 : (__force u32)tuple->dst.u3.ip ^ zone, 0); |
34498825 PM |
216 | j = ((u64)j * (maxip - minip + 1)) >> 32; |
217 | *var_ipp = htonl(minip + j); | |
5b1158e9 JK |
218 | } |
219 | ||
6e23ae2a PM |
220 | /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING, |
221 | * we change the source to map into the range. For NF_INET_PRE_ROUTING | |
222 | * and NF_INET_LOCAL_OUT, we change the destination to map into the | |
5b1158e9 JK |
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. */ | |
226 | static void | |
227 | get_unique_tuple(struct nf_conntrack_tuple *tuple, | |
228 | const struct nf_conntrack_tuple *orig_tuple, | |
229 | const struct nf_nat_range *range, | |
230 | struct nf_conn *ct, | |
231 | enum nf_nat_manip_type maniptype) | |
232 | { | |
0c4c9288 | 233 | struct net *net = nf_ct_net(ct); |
2b628a08 | 234 | const struct nf_nat_protocol *proto; |
5d0aa2cc | 235 | u16 zone = nf_ct_zone(ct); |
5b1158e9 JK |
236 | |
237 | /* 1) If this srcip/proto/src-proto-part is currently mapped, | |
238 | and that same mapping gives a unique tuple within the given | |
239 | range, use that. | |
240 | ||
241 | This is only required for source (ie. NAT/masq) mappings. | |
242 | So far, we don't do local source mappings, so multiple | |
243 | manips not an issue. */ | |
0dbff689 CG |
244 | if (maniptype == IP_NAT_MANIP_SRC && |
245 | !(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) { | |
5d0aa2cc | 246 | if (find_appropriate_src(net, zone, orig_tuple, tuple, range)) { |
0d53778e | 247 | pr_debug("get_unique_tuple: Found current src map\n"); |
0dbff689 CG |
248 | if (!nf_nat_used_tuple(tuple, ct)) |
249 | return; | |
5b1158e9 JK |
250 | } |
251 | } | |
252 | ||
253 | /* 2) Select the least-used IP/proto combination in the given | |
254 | range. */ | |
255 | *tuple = *orig_tuple; | |
5d0aa2cc | 256 | find_best_ips_proto(zone, tuple, range, ct, maniptype); |
5b1158e9 JK |
257 | |
258 | /* 3) The per-protocol part of the manip is made to map into | |
259 | the range to make a unique tuple. */ | |
260 | ||
e22a0548 PM |
261 | rcu_read_lock(); |
262 | proto = __nf_nat_proto_find(orig_tuple->dst.protonum); | |
5b1158e9 | 263 | |
41f4689a EL |
264 | /* Change protocol info to have some randomization */ |
265 | if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) { | |
266 | proto->unique_tuple(tuple, range, maniptype, ct); | |
e22a0548 | 267 | goto out; |
41f4689a EL |
268 | } |
269 | ||
5b1158e9 JK |
270 | /* Only bother mapping if it's not already in range and unique */ |
271 | if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) || | |
272 | proto->in_range(tuple, maniptype, &range->min, &range->max)) && | |
e22a0548 PM |
273 | !nf_nat_used_tuple(tuple, ct)) |
274 | goto out; | |
5b1158e9 JK |
275 | |
276 | /* Last change: get protocol to try to obtain unique tuple. */ | |
277 | proto->unique_tuple(tuple, range, maniptype, ct); | |
e22a0548 PM |
278 | out: |
279 | rcu_read_unlock(); | |
5b1158e9 JK |
280 | } |
281 | ||
282 | unsigned int | |
283 | nf_nat_setup_info(struct nf_conn *ct, | |
284 | const struct nf_nat_range *range, | |
cc01dcbd | 285 | enum nf_nat_manip_type maniptype) |
5b1158e9 | 286 | { |
0c4c9288 | 287 | struct net *net = nf_ct_net(ct); |
5b1158e9 | 288 | struct nf_conntrack_tuple curr_tuple, new_tuple; |
2d59e5ca | 289 | struct nf_conn_nat *nat; |
5b1158e9 | 290 | int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK); |
5b1158e9 | 291 | |
2d59e5ca YK |
292 | /* nat helper or nfctnetlink also setup binding */ |
293 | nat = nfct_nat(ct); | |
294 | if (!nat) { | |
295 | nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC); | |
296 | if (nat == NULL) { | |
0d53778e | 297 | pr_debug("failed to add NAT extension\n"); |
2d59e5ca YK |
298 | return NF_ACCEPT; |
299 | } | |
300 | } | |
301 | ||
cc01dcbd PM |
302 | NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC || |
303 | maniptype == IP_NAT_MANIP_DST); | |
5b1158e9 JK |
304 | BUG_ON(nf_nat_initialized(ct, maniptype)); |
305 | ||
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, | |
309 | orig_tp = | |
310 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */ | |
311 | nf_ct_invert_tuplepr(&curr_tuple, | |
312 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
313 | ||
314 | get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype); | |
315 | ||
316 | if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) { | |
317 | struct nf_conntrack_tuple reply; | |
318 | ||
319 | /* Alter conntrack table so will recognize replies. */ | |
320 | nf_ct_invert_tuplepr(&reply, &new_tuple); | |
321 | nf_conntrack_alter_reply(ct, &reply); | |
322 | ||
323 | /* Non-atomic: we own this at the moment. */ | |
324 | if (maniptype == IP_NAT_MANIP_SRC) | |
325 | ct->status |= IPS_SRC_NAT; | |
326 | else | |
327 | ct->status |= IPS_DST_NAT; | |
328 | } | |
329 | ||
330 | /* Place in source hash if this is the first time. */ | |
331 | if (have_to_hash) { | |
332 | unsigned int srchash; | |
333 | ||
5d0aa2cc PM |
334 | srchash = hash_by_src(net, nf_ct_zone(ct), |
335 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
02502f62 | 336 | spin_lock_bh(&nf_nat_lock); |
2d59e5ca | 337 | /* nf_conntrack_alter_reply might re-allocate exntension aera */ |
b6b84d4a YK |
338 | nat = nfct_nat(ct); |
339 | nat->ct = ct; | |
0c4c9288 AD |
340 | hlist_add_head_rcu(&nat->bysource, |
341 | &net->ipv4.nat_bysource[srchash]); | |
02502f62 | 342 | spin_unlock_bh(&nf_nat_lock); |
5b1158e9 JK |
343 | } |
344 | ||
345 | /* It's done. */ | |
346 | if (maniptype == IP_NAT_MANIP_DST) | |
347 | set_bit(IPS_DST_NAT_DONE_BIT, &ct->status); | |
348 | else | |
349 | set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status); | |
350 | ||
351 | return NF_ACCEPT; | |
352 | } | |
353 | EXPORT_SYMBOL(nf_nat_setup_info); | |
354 | ||
355 | /* Returns true if succeeded. */ | |
f2ea825f | 356 | static bool |
5b1158e9 | 357 | manip_pkt(u_int16_t proto, |
3db05fea | 358 | struct sk_buff *skb, |
5b1158e9 JK |
359 | unsigned int iphdroff, |
360 | const struct nf_conntrack_tuple *target, | |
361 | enum nf_nat_manip_type maniptype) | |
362 | { | |
363 | struct iphdr *iph; | |
2b628a08 | 364 | const struct nf_nat_protocol *p; |
5b1158e9 | 365 | |
3db05fea | 366 | if (!skb_make_writable(skb, iphdroff + sizeof(*iph))) |
f2ea825f | 367 | return false; |
5b1158e9 | 368 | |
3db05fea | 369 | iph = (void *)skb->data + iphdroff; |
5b1158e9 JK |
370 | |
371 | /* Manipulate protcol part. */ | |
e22a0548 PM |
372 | |
373 | /* rcu_read_lock()ed by nf_hook_slow */ | |
374 | p = __nf_nat_proto_find(proto); | |
3db05fea | 375 | if (!p->manip_pkt(skb, iphdroff, target, maniptype)) |
f2ea825f | 376 | return false; |
5b1158e9 | 377 | |
3db05fea | 378 | iph = (void *)skb->data + iphdroff; |
5b1158e9 JK |
379 | |
380 | if (maniptype == IP_NAT_MANIP_SRC) { | |
be0ea7d5 | 381 | csum_replace4(&iph->check, iph->saddr, target->src.u3.ip); |
5b1158e9 JK |
382 | iph->saddr = target->src.u3.ip; |
383 | } else { | |
be0ea7d5 | 384 | csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip); |
5b1158e9 JK |
385 | iph->daddr = target->dst.u3.ip; |
386 | } | |
f2ea825f | 387 | return true; |
5b1158e9 JK |
388 | } |
389 | ||
390 | /* Do packet manipulations according to nf_nat_setup_info. */ | |
391 | unsigned int nf_nat_packet(struct nf_conn *ct, | |
392 | enum ip_conntrack_info ctinfo, | |
393 | unsigned int hooknum, | |
3db05fea | 394 | struct sk_buff *skb) |
5b1158e9 JK |
395 | { |
396 | enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo); | |
397 | unsigned long statusbit; | |
398 | enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum); | |
399 | ||
400 | if (mtype == IP_NAT_MANIP_SRC) | |
401 | statusbit = IPS_SRC_NAT; | |
402 | else | |
403 | statusbit = IPS_DST_NAT; | |
404 | ||
405 | /* Invert if this is reply dir. */ | |
406 | if (dir == IP_CT_DIR_REPLY) | |
407 | statusbit ^= IPS_NAT_MASK; | |
408 | ||
409 | /* Non-atomic: these bits don't change. */ | |
410 | if (ct->status & statusbit) { | |
411 | struct nf_conntrack_tuple target; | |
412 | ||
413 | /* We are aiming to look like inverse of other direction. */ | |
414 | nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple); | |
415 | ||
3db05fea | 416 | if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype)) |
5b1158e9 JK |
417 | return NF_DROP; |
418 | } | |
419 | return NF_ACCEPT; | |
420 | } | |
421 | EXPORT_SYMBOL_GPL(nf_nat_packet); | |
422 | ||
423 | /* Dir is direction ICMP is coming from (opposite to packet it contains) */ | |
424 | int nf_nat_icmp_reply_translation(struct nf_conn *ct, | |
425 | enum ip_conntrack_info ctinfo, | |
426 | unsigned int hooknum, | |
3db05fea | 427 | struct sk_buff *skb) |
5b1158e9 JK |
428 | { |
429 | struct { | |
430 | struct icmphdr icmp; | |
431 | struct iphdr ip; | |
432 | } *inside; | |
72b72949 | 433 | const struct nf_conntrack_l4proto *l4proto; |
5b1158e9 | 434 | struct nf_conntrack_tuple inner, target; |
3db05fea | 435 | int hdrlen = ip_hdrlen(skb); |
5b1158e9 JK |
436 | enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo); |
437 | unsigned long statusbit; | |
438 | enum nf_nat_manip_type manip = HOOK2MANIP(hooknum); | |
439 | ||
3db05fea | 440 | if (!skb_make_writable(skb, hdrlen + sizeof(*inside))) |
5b1158e9 JK |
441 | return 0; |
442 | ||
3db05fea | 443 | inside = (void *)skb->data + ip_hdrlen(skb); |
5b1158e9 JK |
444 | |
445 | /* We're actually going to mangle it beyond trivial checksum | |
446 | adjustment, so make sure the current checksum is correct. */ | |
3db05fea | 447 | if (nf_ip_checksum(skb, hooknum, hdrlen, 0)) |
5b1158e9 JK |
448 | return 0; |
449 | ||
450 | /* Must be RELATED */ | |
3db05fea HX |
451 | NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED || |
452 | skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY); | |
5b1158e9 JK |
453 | |
454 | /* Redirects on non-null nats must be dropped, else they'll | |
e905a9ed YH |
455 | start talking to each other without our translation, and be |
456 | confused... --RR */ | |
5b1158e9 JK |
457 | if (inside->icmp.type == ICMP_REDIRECT) { |
458 | /* If NAT isn't finished, assume it and drop. */ | |
459 | if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK) | |
460 | return 0; | |
461 | ||
462 | if (ct->status & IPS_NAT_MASK) | |
463 | return 0; | |
464 | } | |
465 | ||
0d53778e | 466 | pr_debug("icmp_reply_translation: translating error %p manip %u " |
3db05fea | 467 | "dir %s\n", skb, manip, |
0d53778e | 468 | dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY"); |
5b1158e9 | 469 | |
923f4902 PM |
470 | /* rcu_read_lock()ed by nf_hook_slow */ |
471 | l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol); | |
472 | ||
3db05fea HX |
473 | if (!nf_ct_get_tuple(skb, |
474 | ip_hdrlen(skb) + sizeof(struct icmphdr), | |
475 | (ip_hdrlen(skb) + | |
c9bdd4b5 | 476 | sizeof(struct icmphdr) + inside->ip.ihl * 4), |
e905a9ed YH |
477 | (u_int16_t)AF_INET, |
478 | inside->ip.protocol, | |
923f4902 | 479 | &inner, l3proto, l4proto)) |
5b1158e9 JK |
480 | return 0; |
481 | ||
482 | /* Change inner back to look like incoming packet. We do the | |
483 | opposite manip on this hook to normal, because it might not | |
484 | pass all hooks (locally-generated ICMP). Consider incoming | |
485 | packet: PREROUTING (DST manip), routing produces ICMP, goes | |
486 | through POSTROUTING (which must correct the DST manip). */ | |
3db05fea HX |
487 | if (!manip_pkt(inside->ip.protocol, skb, |
488 | ip_hdrlen(skb) + sizeof(inside->icmp), | |
5b1158e9 JK |
489 | &ct->tuplehash[!dir].tuple, |
490 | !manip)) | |
491 | return 0; | |
492 | ||
3db05fea | 493 | if (skb->ip_summed != CHECKSUM_PARTIAL) { |
5b1158e9 | 494 | /* Reloading "inside" here since manip_pkt inner. */ |
3db05fea | 495 | inside = (void *)skb->data + ip_hdrlen(skb); |
5b1158e9 JK |
496 | inside->icmp.checksum = 0; |
497 | inside->icmp.checksum = | |
3db05fea HX |
498 | csum_fold(skb_checksum(skb, hdrlen, |
499 | skb->len - hdrlen, 0)); | |
5b1158e9 JK |
500 | } |
501 | ||
502 | /* Change outer to look the reply to an incoming packet | |
503 | * (proto 0 means don't invert per-proto part). */ | |
504 | if (manip == IP_NAT_MANIP_SRC) | |
505 | statusbit = IPS_SRC_NAT; | |
506 | else | |
507 | statusbit = IPS_DST_NAT; | |
508 | ||
509 | /* Invert if this is reply dir. */ | |
510 | if (dir == IP_CT_DIR_REPLY) | |
511 | statusbit ^= IPS_NAT_MASK; | |
512 | ||
513 | if (ct->status & statusbit) { | |
514 | nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple); | |
3db05fea | 515 | if (!manip_pkt(0, skb, 0, &target, manip)) |
5b1158e9 JK |
516 | return 0; |
517 | } | |
518 | ||
519 | return 1; | |
520 | } | |
521 | EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation); | |
522 | ||
523 | /* Protocol registration. */ | |
2b628a08 | 524 | int nf_nat_protocol_register(const struct nf_nat_protocol *proto) |
5b1158e9 JK |
525 | { |
526 | int ret = 0; | |
527 | ||
02502f62 | 528 | spin_lock_bh(&nf_nat_lock); |
5b1158e9 JK |
529 | if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) { |
530 | ret = -EBUSY; | |
531 | goto out; | |
532 | } | |
e22a0548 | 533 | rcu_assign_pointer(nf_nat_protos[proto->protonum], proto); |
5b1158e9 | 534 | out: |
02502f62 | 535 | spin_unlock_bh(&nf_nat_lock); |
5b1158e9 JK |
536 | return ret; |
537 | } | |
538 | EXPORT_SYMBOL(nf_nat_protocol_register); | |
539 | ||
540 | /* Noone stores the protocol anywhere; simply delete it. */ | |
2b628a08 | 541 | void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto) |
5b1158e9 | 542 | { |
02502f62 | 543 | spin_lock_bh(&nf_nat_lock); |
e22a0548 PM |
544 | rcu_assign_pointer(nf_nat_protos[proto->protonum], |
545 | &nf_nat_unknown_protocol); | |
02502f62 | 546 | spin_unlock_bh(&nf_nat_lock); |
e22a0548 | 547 | synchronize_rcu(); |
5b1158e9 JK |
548 | } |
549 | EXPORT_SYMBOL(nf_nat_protocol_unregister); | |
550 | ||
d8a0509a YK |
551 | /* Noone using conntrack by the time this called. */ |
552 | static void nf_nat_cleanup_conntrack(struct nf_conn *ct) | |
553 | { | |
554 | struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT); | |
555 | ||
b6b84d4a | 556 | if (nat == NULL || nat->ct == NULL) |
d8a0509a YK |
557 | return; |
558 | ||
b6b84d4a | 559 | NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK); |
d8a0509a | 560 | |
02502f62 | 561 | spin_lock_bh(&nf_nat_lock); |
4d354c57 | 562 | hlist_del_rcu(&nat->bysource); |
02502f62 | 563 | spin_unlock_bh(&nf_nat_lock); |
d8a0509a YK |
564 | } |
565 | ||
86577c66 | 566 | static void nf_nat_move_storage(void *new, void *old) |
2d59e5ca | 567 | { |
86577c66 PM |
568 | struct nf_conn_nat *new_nat = new; |
569 | struct nf_conn_nat *old_nat = old; | |
b6b84d4a | 570 | struct nf_conn *ct = old_nat->ct; |
2d59e5ca | 571 | |
1f305323 | 572 | if (!ct || !(ct->status & IPS_NAT_DONE_MASK)) |
2d59e5ca YK |
573 | return; |
574 | ||
02502f62 | 575 | spin_lock_bh(&nf_nat_lock); |
b6b84d4a | 576 | new_nat->ct = ct; |
68b80f11 | 577 | hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource); |
02502f62 | 578 | spin_unlock_bh(&nf_nat_lock); |
2d59e5ca YK |
579 | } |
580 | ||
61eb3107 | 581 | static struct nf_ct_ext_type nat_extend __read_mostly = { |
d8a0509a YK |
582 | .len = sizeof(struct nf_conn_nat), |
583 | .align = __alignof__(struct nf_conn_nat), | |
584 | .destroy = nf_nat_cleanup_conntrack, | |
585 | .move = nf_nat_move_storage, | |
586 | .id = NF_CT_EXT_NAT, | |
587 | .flags = NF_CT_EXT_F_PREALLOC, | |
2d59e5ca YK |
588 | }; |
589 | ||
e6a7d3c0 PNA |
590 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
591 | ||
592 | #include <linux/netfilter/nfnetlink.h> | |
593 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
594 | ||
595 | static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = { | |
596 | [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 }, | |
597 | [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 }, | |
598 | }; | |
599 | ||
600 | static int nfnetlink_parse_nat_proto(struct nlattr *attr, | |
601 | const struct nf_conn *ct, | |
602 | struct nf_nat_range *range) | |
603 | { | |
604 | struct nlattr *tb[CTA_PROTONAT_MAX+1]; | |
605 | const struct nf_nat_protocol *npt; | |
606 | int err; | |
607 | ||
608 | err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy); | |
609 | if (err < 0) | |
610 | return err; | |
611 | ||
612 | npt = nf_nat_proto_find_get(nf_ct_protonum(ct)); | |
613 | if (npt->nlattr_to_range) | |
614 | err = npt->nlattr_to_range(tb, range); | |
615 | nf_nat_proto_put(npt); | |
616 | return err; | |
617 | } | |
618 | ||
619 | static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = { | |
620 | [CTA_NAT_MINIP] = { .type = NLA_U32 }, | |
621 | [CTA_NAT_MAXIP] = { .type = NLA_U32 }, | |
622 | }; | |
623 | ||
624 | static int | |
39938324 | 625 | nfnetlink_parse_nat(const struct nlattr *nat, |
e6a7d3c0 PNA |
626 | const struct nf_conn *ct, struct nf_nat_range *range) |
627 | { | |
628 | struct nlattr *tb[CTA_NAT_MAX+1]; | |
629 | int err; | |
630 | ||
631 | memset(range, 0, sizeof(*range)); | |
632 | ||
633 | err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy); | |
634 | if (err < 0) | |
635 | return err; | |
636 | ||
637 | if (tb[CTA_NAT_MINIP]) | |
638 | range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]); | |
639 | ||
640 | if (!tb[CTA_NAT_MAXIP]) | |
641 | range->max_ip = range->min_ip; | |
642 | else | |
643 | range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]); | |
644 | ||
645 | if (range->min_ip) | |
646 | range->flags |= IP_NAT_RANGE_MAP_IPS; | |
647 | ||
648 | if (!tb[CTA_NAT_PROTO]) | |
649 | return 0; | |
650 | ||
651 | err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range); | |
652 | if (err < 0) | |
653 | return err; | |
654 | ||
655 | return 0; | |
656 | } | |
657 | ||
658 | static int | |
659 | nfnetlink_parse_nat_setup(struct nf_conn *ct, | |
660 | enum nf_nat_manip_type manip, | |
39938324 | 661 | const struct nlattr *attr) |
e6a7d3c0 PNA |
662 | { |
663 | struct nf_nat_range range; | |
664 | ||
665 | if (nfnetlink_parse_nat(attr, ct, &range) < 0) | |
666 | return -EINVAL; | |
667 | if (nf_nat_initialized(ct, manip)) | |
668 | return -EEXIST; | |
669 | ||
670 | return nf_nat_setup_info(ct, &range, manip); | |
671 | } | |
672 | #else | |
673 | static int | |
674 | nfnetlink_parse_nat_setup(struct nf_conn *ct, | |
675 | enum nf_nat_manip_type manip, | |
39938324 | 676 | const struct nlattr *attr) |
e6a7d3c0 PNA |
677 | { |
678 | return -EOPNOTSUPP; | |
679 | } | |
680 | #endif | |
681 | ||
0c4c9288 AD |
682 | static int __net_init nf_nat_net_init(struct net *net) |
683 | { | |
d696c7bd PM |
684 | /* Leave them the same for the moment. */ |
685 | net->ipv4.nat_htable_size = net->ct.htable_size; | |
686 | net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.nat_htable_size, | |
687 | &net->ipv4.nat_vmalloced, 0); | |
0c4c9288 AD |
688 | if (!net->ipv4.nat_bysource) |
689 | return -ENOMEM; | |
690 | return 0; | |
691 | } | |
692 | ||
693 | /* Clear NAT section of all conntracks, in case we're loaded again. */ | |
694 | static int clean_nat(struct nf_conn *i, void *data) | |
695 | { | |
696 | struct nf_conn_nat *nat = nfct_nat(i); | |
697 | ||
698 | if (!nat) | |
699 | return 0; | |
700 | memset(nat, 0, sizeof(*nat)); | |
701 | i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST); | |
702 | return 0; | |
703 | } | |
704 | ||
705 | static void __net_exit nf_nat_net_exit(struct net *net) | |
706 | { | |
707 | nf_ct_iterate_cleanup(net, &clean_nat, NULL); | |
708 | synchronize_rcu(); | |
709 | nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_vmalloced, | |
d696c7bd | 710 | net->ipv4.nat_htable_size); |
0c4c9288 AD |
711 | } |
712 | ||
713 | static struct pernet_operations nf_nat_net_ops = { | |
714 | .init = nf_nat_net_init, | |
715 | .exit = nf_nat_net_exit, | |
716 | }; | |
717 | ||
5b1158e9 JK |
718 | static int __init nf_nat_init(void) |
719 | { | |
720 | size_t i; | |
2d59e5ca YK |
721 | int ret; |
722 | ||
475959d4 JE |
723 | need_ipv4_conntrack(); |
724 | ||
2d59e5ca YK |
725 | ret = nf_ct_extend_register(&nat_extend); |
726 | if (ret < 0) { | |
727 | printk(KERN_ERR "nf_nat_core: Unable to register extension\n"); | |
728 | return ret; | |
729 | } | |
5b1158e9 | 730 | |
0c4c9288 AD |
731 | ret = register_pernet_subsys(&nf_nat_net_ops); |
732 | if (ret < 0) | |
2d59e5ca | 733 | goto cleanup_extend; |
5b1158e9 JK |
734 | |
735 | /* Sew in builtin protocols. */ | |
02502f62 | 736 | spin_lock_bh(&nf_nat_lock); |
5b1158e9 | 737 | for (i = 0; i < MAX_IP_NAT_PROTO; i++) |
e22a0548 PM |
738 | rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol); |
739 | rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp); | |
740 | rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp); | |
741 | rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp); | |
02502f62 | 742 | spin_unlock_bh(&nf_nat_lock); |
5b1158e9 | 743 | |
5b1158e9 JK |
744 | /* Initialize fake conntrack so that NAT will skip it */ |
745 | nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK; | |
746 | ||
747 | l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET); | |
dd13b010 PM |
748 | |
749 | BUG_ON(nf_nat_seq_adjust_hook != NULL); | |
750 | rcu_assign_pointer(nf_nat_seq_adjust_hook, nf_nat_seq_adjust); | |
e6a7d3c0 PNA |
751 | BUG_ON(nfnetlink_parse_nat_setup_hook != NULL); |
752 | rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, | |
753 | nfnetlink_parse_nat_setup); | |
f9dd09c7 JK |
754 | BUG_ON(nf_ct_nat_offset != NULL); |
755 | rcu_assign_pointer(nf_ct_nat_offset, nf_nat_get_offset); | |
5b1158e9 | 756 | return 0; |
2d59e5ca YK |
757 | |
758 | cleanup_extend: | |
759 | nf_ct_extend_unregister(&nat_extend); | |
760 | return ret; | |
5b1158e9 JK |
761 | } |
762 | ||
5b1158e9 JK |
763 | static void __exit nf_nat_cleanup(void) |
764 | { | |
0c4c9288 | 765 | unregister_pernet_subsys(&nf_nat_net_ops); |
5b1158e9 | 766 | nf_ct_l3proto_put(l3proto); |
2d59e5ca | 767 | nf_ct_extend_unregister(&nat_extend); |
dd13b010 | 768 | rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL); |
e6a7d3c0 | 769 | rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL); |
f9dd09c7 | 770 | rcu_assign_pointer(nf_ct_nat_offset, NULL); |
dd13b010 | 771 | synchronize_net(); |
5b1158e9 JK |
772 | } |
773 | ||
774 | MODULE_LICENSE("GPL"); | |
e6a7d3c0 | 775 | MODULE_ALIAS("nf-nat-ipv4"); |
5b1158e9 JK |
776 | |
777 | module_init(nf_nat_init); | |
778 | module_exit(nf_nat_cleanup); |