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> | |
5b1158e9 JK |
15 | #include <net/checksum.h> |
16 | #include <net/icmp.h> | |
17 | #include <net/ip.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> | |
22 | ||
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> | |
33 | ||
02502f62 | 34 | static DEFINE_SPINLOCK(nf_nat_lock); |
5b1158e9 | 35 | |
ce4b1ceb | 36 | static struct nf_conntrack_l3proto *l3proto __read_mostly; |
5b1158e9 JK |
37 | |
38 | /* Calculated at init based on memory size */ | |
ce4b1ceb | 39 | static unsigned int nf_nat_htable_size __read_mostly; |
5b1158e9 JK |
40 | |
41 | #define MAX_IP_NAT_PROTO 256 | |
ce4b1ceb PM |
42 | static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO] |
43 | __read_mostly; | |
5b1158e9 | 44 | |
2b628a08 | 45 | static inline const struct nf_nat_protocol * |
5b1158e9 JK |
46 | __nf_nat_proto_find(u_int8_t protonum) |
47 | { | |
e22a0548 | 48 | return rcu_dereference(nf_nat_protos[protonum]); |
5b1158e9 JK |
49 | } |
50 | ||
2b628a08 | 51 | const struct nf_nat_protocol * |
5b1158e9 JK |
52 | nf_nat_proto_find_get(u_int8_t protonum) |
53 | { | |
2b628a08 | 54 | const struct nf_nat_protocol *p; |
5b1158e9 | 55 | |
e22a0548 | 56 | rcu_read_lock(); |
5b1158e9 JK |
57 | p = __nf_nat_proto_find(protonum); |
58 | if (!try_module_get(p->me)) | |
59 | p = &nf_nat_unknown_protocol; | |
e22a0548 | 60 | rcu_read_unlock(); |
5b1158e9 JK |
61 | |
62 | return p; | |
63 | } | |
64 | EXPORT_SYMBOL_GPL(nf_nat_proto_find_get); | |
65 | ||
66 | void | |
2b628a08 | 67 | nf_nat_proto_put(const struct nf_nat_protocol *p) |
5b1158e9 JK |
68 | { |
69 | module_put(p->me); | |
70 | } | |
71 | EXPORT_SYMBOL_GPL(nf_nat_proto_put); | |
72 | ||
73 | /* We keep an extra hash for each conntrack, for fast searching. */ | |
74 | static inline unsigned int | |
75 | hash_by_src(const struct nf_conntrack_tuple *tuple) | |
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, |
a34c4589 | 81 | (__force u32)tuple->src.u.all, |
34498825 PM |
82 | tuple->dst.protonum, 0); |
83 | return ((u64)hash * nf_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 | |
0c4c9288 AD |
145 | find_appropriate_src(struct net *net, |
146 | const struct nf_conntrack_tuple *tuple, | |
5b1158e9 JK |
147 | struct nf_conntrack_tuple *result, |
148 | const struct nf_nat_range *range) | |
149 | { | |
150 | unsigned int h = hash_by_src(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; |
5b1158e9 JK |
158 | if (same_src(ct, tuple)) { |
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 | |
181 | find_best_ips_proto(struct nf_conntrack_tuple *tuple, | |
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, | |
214 | (__force u32)tuple->dst.u3.ip, 0); | |
34498825 PM |
215 | j = ((u64)j * (maxip - minip + 1)) >> 32; |
216 | *var_ipp = htonl(minip + j); | |
5b1158e9 JK |
217 | } |
218 | ||
6e23ae2a PM |
219 | /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING, |
220 | * we change the source to map into the range. For NF_INET_PRE_ROUTING | |
221 | * and NF_INET_LOCAL_OUT, we change the destination to map into the | |
5b1158e9 JK |
222 | * range. It might not be possible to get a unique tuple, but we try. |
223 | * At worst (or if we race), we will end up with a final duplicate in | |
224 | * __ip_conntrack_confirm and drop the packet. */ | |
225 | static void | |
226 | get_unique_tuple(struct nf_conntrack_tuple *tuple, | |
227 | const struct nf_conntrack_tuple *orig_tuple, | |
228 | const struct nf_nat_range *range, | |
229 | struct nf_conn *ct, | |
230 | enum nf_nat_manip_type maniptype) | |
231 | { | |
0c4c9288 | 232 | struct net *net = nf_ct_net(ct); |
2b628a08 | 233 | const struct nf_nat_protocol *proto; |
5b1158e9 JK |
234 | |
235 | /* 1) If this srcip/proto/src-proto-part is currently mapped, | |
236 | and that same mapping gives a unique tuple within the given | |
237 | range, use that. | |
238 | ||
239 | This is only required for source (ie. NAT/masq) mappings. | |
240 | So far, we don't do local source mappings, so multiple | |
241 | manips not an issue. */ | |
0dbff689 CG |
242 | if (maniptype == IP_NAT_MANIP_SRC && |
243 | !(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) { | |
0c4c9288 | 244 | if (find_appropriate_src(net, orig_tuple, tuple, range)) { |
0d53778e | 245 | pr_debug("get_unique_tuple: Found current src map\n"); |
0dbff689 CG |
246 | if (!nf_nat_used_tuple(tuple, ct)) |
247 | return; | |
5b1158e9 JK |
248 | } |
249 | } | |
250 | ||
251 | /* 2) Select the least-used IP/proto combination in the given | |
252 | range. */ | |
253 | *tuple = *orig_tuple; | |
254 | find_best_ips_proto(tuple, range, ct, maniptype); | |
255 | ||
256 | /* 3) The per-protocol part of the manip is made to map into | |
257 | the range to make a unique tuple. */ | |
258 | ||
e22a0548 PM |
259 | rcu_read_lock(); |
260 | proto = __nf_nat_proto_find(orig_tuple->dst.protonum); | |
5b1158e9 | 261 | |
41f4689a EL |
262 | /* Change protocol info to have some randomization */ |
263 | if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) { | |
264 | proto->unique_tuple(tuple, range, maniptype, ct); | |
e22a0548 | 265 | goto out; |
41f4689a EL |
266 | } |
267 | ||
5b1158e9 JK |
268 | /* Only bother mapping if it's not already in range and unique */ |
269 | if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) || | |
270 | proto->in_range(tuple, maniptype, &range->min, &range->max)) && | |
e22a0548 PM |
271 | !nf_nat_used_tuple(tuple, ct)) |
272 | goto out; | |
5b1158e9 JK |
273 | |
274 | /* Last change: get protocol to try to obtain unique tuple. */ | |
275 | proto->unique_tuple(tuple, range, maniptype, ct); | |
e22a0548 PM |
276 | out: |
277 | rcu_read_unlock(); | |
5b1158e9 JK |
278 | } |
279 | ||
280 | unsigned int | |
281 | nf_nat_setup_info(struct nf_conn *ct, | |
282 | const struct nf_nat_range *range, | |
cc01dcbd | 283 | enum nf_nat_manip_type maniptype) |
5b1158e9 | 284 | { |
0c4c9288 | 285 | struct net *net = nf_ct_net(ct); |
5b1158e9 | 286 | struct nf_conntrack_tuple curr_tuple, new_tuple; |
2d59e5ca | 287 | struct nf_conn_nat *nat; |
5b1158e9 | 288 | int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK); |
5b1158e9 | 289 | |
2d59e5ca YK |
290 | /* nat helper or nfctnetlink also setup binding */ |
291 | nat = nfct_nat(ct); | |
292 | if (!nat) { | |
293 | nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC); | |
294 | if (nat == NULL) { | |
0d53778e | 295 | pr_debug("failed to add NAT extension\n"); |
2d59e5ca YK |
296 | return NF_ACCEPT; |
297 | } | |
298 | } | |
299 | ||
cc01dcbd PM |
300 | NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC || |
301 | maniptype == IP_NAT_MANIP_DST); | |
5b1158e9 JK |
302 | BUG_ON(nf_nat_initialized(ct, maniptype)); |
303 | ||
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, | |
307 | orig_tp = | |
308 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */ | |
309 | nf_ct_invert_tuplepr(&curr_tuple, | |
310 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
311 | ||
312 | get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype); | |
313 | ||
314 | if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) { | |
315 | struct nf_conntrack_tuple reply; | |
316 | ||
317 | /* Alter conntrack table so will recognize replies. */ | |
318 | nf_ct_invert_tuplepr(&reply, &new_tuple); | |
319 | nf_conntrack_alter_reply(ct, &reply); | |
320 | ||
321 | /* Non-atomic: we own this at the moment. */ | |
322 | if (maniptype == IP_NAT_MANIP_SRC) | |
323 | ct->status |= IPS_SRC_NAT; | |
324 | else | |
325 | ct->status |= IPS_DST_NAT; | |
326 | } | |
327 | ||
328 | /* Place in source hash if this is the first time. */ | |
329 | if (have_to_hash) { | |
330 | unsigned int srchash; | |
331 | ||
332 | srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
02502f62 | 333 | spin_lock_bh(&nf_nat_lock); |
2d59e5ca | 334 | /* nf_conntrack_alter_reply might re-allocate exntension aera */ |
b6b84d4a YK |
335 | nat = nfct_nat(ct); |
336 | nat->ct = ct; | |
0c4c9288 AD |
337 | hlist_add_head_rcu(&nat->bysource, |
338 | &net->ipv4.nat_bysource[srchash]); | |
02502f62 | 339 | spin_unlock_bh(&nf_nat_lock); |
5b1158e9 JK |
340 | } |
341 | ||
342 | /* It's done. */ | |
343 | if (maniptype == IP_NAT_MANIP_DST) | |
344 | set_bit(IPS_DST_NAT_DONE_BIT, &ct->status); | |
345 | else | |
346 | set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status); | |
347 | ||
348 | return NF_ACCEPT; | |
349 | } | |
350 | EXPORT_SYMBOL(nf_nat_setup_info); | |
351 | ||
352 | /* Returns true if succeeded. */ | |
f2ea825f | 353 | static bool |
5b1158e9 | 354 | manip_pkt(u_int16_t proto, |
3db05fea | 355 | struct sk_buff *skb, |
5b1158e9 JK |
356 | unsigned int iphdroff, |
357 | const struct nf_conntrack_tuple *target, | |
358 | enum nf_nat_manip_type maniptype) | |
359 | { | |
360 | struct iphdr *iph; | |
2b628a08 | 361 | const struct nf_nat_protocol *p; |
5b1158e9 | 362 | |
3db05fea | 363 | if (!skb_make_writable(skb, iphdroff + sizeof(*iph))) |
f2ea825f | 364 | return false; |
5b1158e9 | 365 | |
3db05fea | 366 | iph = (void *)skb->data + iphdroff; |
5b1158e9 JK |
367 | |
368 | /* Manipulate protcol part. */ | |
e22a0548 PM |
369 | |
370 | /* rcu_read_lock()ed by nf_hook_slow */ | |
371 | p = __nf_nat_proto_find(proto); | |
3db05fea | 372 | if (!p->manip_pkt(skb, iphdroff, target, maniptype)) |
f2ea825f | 373 | return false; |
5b1158e9 | 374 | |
3db05fea | 375 | iph = (void *)skb->data + iphdroff; |
5b1158e9 JK |
376 | |
377 | if (maniptype == IP_NAT_MANIP_SRC) { | |
be0ea7d5 | 378 | csum_replace4(&iph->check, iph->saddr, target->src.u3.ip); |
5b1158e9 JK |
379 | iph->saddr = target->src.u3.ip; |
380 | } else { | |
be0ea7d5 | 381 | csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip); |
5b1158e9 JK |
382 | iph->daddr = target->dst.u3.ip; |
383 | } | |
f2ea825f | 384 | return true; |
5b1158e9 JK |
385 | } |
386 | ||
387 | /* Do packet manipulations according to nf_nat_setup_info. */ | |
388 | unsigned int nf_nat_packet(struct nf_conn *ct, | |
389 | enum ip_conntrack_info ctinfo, | |
390 | unsigned int hooknum, | |
3db05fea | 391 | struct sk_buff *skb) |
5b1158e9 JK |
392 | { |
393 | enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo); | |
394 | unsigned long statusbit; | |
395 | enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum); | |
396 | ||
397 | if (mtype == IP_NAT_MANIP_SRC) | |
398 | statusbit = IPS_SRC_NAT; | |
399 | else | |
400 | statusbit = IPS_DST_NAT; | |
401 | ||
402 | /* Invert if this is reply dir. */ | |
403 | if (dir == IP_CT_DIR_REPLY) | |
404 | statusbit ^= IPS_NAT_MASK; | |
405 | ||
406 | /* Non-atomic: these bits don't change. */ | |
407 | if (ct->status & statusbit) { | |
408 | struct nf_conntrack_tuple target; | |
409 | ||
410 | /* We are aiming to look like inverse of other direction. */ | |
411 | nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple); | |
412 | ||
3db05fea | 413 | if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype)) |
5b1158e9 JK |
414 | return NF_DROP; |
415 | } | |
416 | return NF_ACCEPT; | |
417 | } | |
418 | EXPORT_SYMBOL_GPL(nf_nat_packet); | |
419 | ||
420 | /* Dir is direction ICMP is coming from (opposite to packet it contains) */ | |
421 | int nf_nat_icmp_reply_translation(struct nf_conn *ct, | |
422 | enum ip_conntrack_info ctinfo, | |
423 | unsigned int hooknum, | |
3db05fea | 424 | struct sk_buff *skb) |
5b1158e9 JK |
425 | { |
426 | struct { | |
427 | struct icmphdr icmp; | |
428 | struct iphdr ip; | |
429 | } *inside; | |
72b72949 | 430 | const struct nf_conntrack_l4proto *l4proto; |
5b1158e9 | 431 | struct nf_conntrack_tuple inner, target; |
3db05fea | 432 | int hdrlen = ip_hdrlen(skb); |
5b1158e9 JK |
433 | enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo); |
434 | unsigned long statusbit; | |
435 | enum nf_nat_manip_type manip = HOOK2MANIP(hooknum); | |
436 | ||
3db05fea | 437 | if (!skb_make_writable(skb, hdrlen + sizeof(*inside))) |
5b1158e9 JK |
438 | return 0; |
439 | ||
3db05fea | 440 | inside = (void *)skb->data + ip_hdrlen(skb); |
5b1158e9 JK |
441 | |
442 | /* We're actually going to mangle it beyond trivial checksum | |
443 | adjustment, so make sure the current checksum is correct. */ | |
3db05fea | 444 | if (nf_ip_checksum(skb, hooknum, hdrlen, 0)) |
5b1158e9 JK |
445 | return 0; |
446 | ||
447 | /* Must be RELATED */ | |
3db05fea HX |
448 | NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED || |
449 | skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY); | |
5b1158e9 JK |
450 | |
451 | /* Redirects on non-null nats must be dropped, else they'll | |
e905a9ed YH |
452 | start talking to each other without our translation, and be |
453 | confused... --RR */ | |
5b1158e9 JK |
454 | if (inside->icmp.type == ICMP_REDIRECT) { |
455 | /* If NAT isn't finished, assume it and drop. */ | |
456 | if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK) | |
457 | return 0; | |
458 | ||
459 | if (ct->status & IPS_NAT_MASK) | |
460 | return 0; | |
461 | } | |
462 | ||
0d53778e | 463 | pr_debug("icmp_reply_translation: translating error %p manip %u " |
3db05fea | 464 | "dir %s\n", skb, manip, |
0d53778e | 465 | dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY"); |
5b1158e9 | 466 | |
923f4902 PM |
467 | /* rcu_read_lock()ed by nf_hook_slow */ |
468 | l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol); | |
469 | ||
3db05fea HX |
470 | if (!nf_ct_get_tuple(skb, |
471 | ip_hdrlen(skb) + sizeof(struct icmphdr), | |
472 | (ip_hdrlen(skb) + | |
c9bdd4b5 | 473 | sizeof(struct icmphdr) + inside->ip.ihl * 4), |
e905a9ed YH |
474 | (u_int16_t)AF_INET, |
475 | inside->ip.protocol, | |
923f4902 | 476 | &inner, l3proto, l4proto)) |
5b1158e9 JK |
477 | return 0; |
478 | ||
479 | /* Change inner back to look like incoming packet. We do the | |
480 | opposite manip on this hook to normal, because it might not | |
481 | pass all hooks (locally-generated ICMP). Consider incoming | |
482 | packet: PREROUTING (DST manip), routing produces ICMP, goes | |
483 | through POSTROUTING (which must correct the DST manip). */ | |
3db05fea HX |
484 | if (!manip_pkt(inside->ip.protocol, skb, |
485 | ip_hdrlen(skb) + sizeof(inside->icmp), | |
5b1158e9 JK |
486 | &ct->tuplehash[!dir].tuple, |
487 | !manip)) | |
488 | return 0; | |
489 | ||
3db05fea | 490 | if (skb->ip_summed != CHECKSUM_PARTIAL) { |
5b1158e9 | 491 | /* Reloading "inside" here since manip_pkt inner. */ |
3db05fea | 492 | inside = (void *)skb->data + ip_hdrlen(skb); |
5b1158e9 JK |
493 | inside->icmp.checksum = 0; |
494 | inside->icmp.checksum = | |
3db05fea HX |
495 | csum_fold(skb_checksum(skb, hdrlen, |
496 | skb->len - hdrlen, 0)); | |
5b1158e9 JK |
497 | } |
498 | ||
499 | /* Change outer to look the reply to an incoming packet | |
500 | * (proto 0 means don't invert per-proto part). */ | |
501 | if (manip == IP_NAT_MANIP_SRC) | |
502 | statusbit = IPS_SRC_NAT; | |
503 | else | |
504 | statusbit = IPS_DST_NAT; | |
505 | ||
506 | /* Invert if this is reply dir. */ | |
507 | if (dir == IP_CT_DIR_REPLY) | |
508 | statusbit ^= IPS_NAT_MASK; | |
509 | ||
510 | if (ct->status & statusbit) { | |
511 | nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple); | |
3db05fea | 512 | if (!manip_pkt(0, skb, 0, &target, manip)) |
5b1158e9 JK |
513 | return 0; |
514 | } | |
515 | ||
516 | return 1; | |
517 | } | |
518 | EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation); | |
519 | ||
520 | /* Protocol registration. */ | |
2b628a08 | 521 | int nf_nat_protocol_register(const struct nf_nat_protocol *proto) |
5b1158e9 JK |
522 | { |
523 | int ret = 0; | |
524 | ||
02502f62 | 525 | spin_lock_bh(&nf_nat_lock); |
5b1158e9 JK |
526 | if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) { |
527 | ret = -EBUSY; | |
528 | goto out; | |
529 | } | |
e22a0548 | 530 | rcu_assign_pointer(nf_nat_protos[proto->protonum], proto); |
5b1158e9 | 531 | out: |
02502f62 | 532 | spin_unlock_bh(&nf_nat_lock); |
5b1158e9 JK |
533 | return ret; |
534 | } | |
535 | EXPORT_SYMBOL(nf_nat_protocol_register); | |
536 | ||
537 | /* Noone stores the protocol anywhere; simply delete it. */ | |
2b628a08 | 538 | void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto) |
5b1158e9 | 539 | { |
02502f62 | 540 | spin_lock_bh(&nf_nat_lock); |
e22a0548 PM |
541 | rcu_assign_pointer(nf_nat_protos[proto->protonum], |
542 | &nf_nat_unknown_protocol); | |
02502f62 | 543 | spin_unlock_bh(&nf_nat_lock); |
e22a0548 | 544 | synchronize_rcu(); |
5b1158e9 JK |
545 | } |
546 | EXPORT_SYMBOL(nf_nat_protocol_unregister); | |
547 | ||
d8a0509a YK |
548 | /* Noone using conntrack by the time this called. */ |
549 | static void nf_nat_cleanup_conntrack(struct nf_conn *ct) | |
550 | { | |
551 | struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT); | |
552 | ||
b6b84d4a | 553 | if (nat == NULL || nat->ct == NULL) |
d8a0509a YK |
554 | return; |
555 | ||
b6b84d4a | 556 | NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK); |
d8a0509a | 557 | |
02502f62 | 558 | spin_lock_bh(&nf_nat_lock); |
4d354c57 | 559 | hlist_del_rcu(&nat->bysource); |
02502f62 | 560 | spin_unlock_bh(&nf_nat_lock); |
d8a0509a YK |
561 | } |
562 | ||
86577c66 | 563 | static void nf_nat_move_storage(void *new, void *old) |
2d59e5ca | 564 | { |
86577c66 PM |
565 | struct nf_conn_nat *new_nat = new; |
566 | struct nf_conn_nat *old_nat = old; | |
b6b84d4a | 567 | struct nf_conn *ct = old_nat->ct; |
2d59e5ca | 568 | |
1f305323 | 569 | if (!ct || !(ct->status & IPS_NAT_DONE_MASK)) |
2d59e5ca YK |
570 | return; |
571 | ||
02502f62 | 572 | spin_lock_bh(&nf_nat_lock); |
b6b84d4a | 573 | new_nat->ct = ct; |
68b80f11 | 574 | hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource); |
02502f62 | 575 | spin_unlock_bh(&nf_nat_lock); |
2d59e5ca YK |
576 | } |
577 | ||
61eb3107 | 578 | static struct nf_ct_ext_type nat_extend __read_mostly = { |
d8a0509a YK |
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, | |
583 | .id = NF_CT_EXT_NAT, | |
584 | .flags = NF_CT_EXT_F_PREALLOC, | |
2d59e5ca YK |
585 | }; |
586 | ||
0c4c9288 AD |
587 | static int __net_init nf_nat_net_init(struct net *net) |
588 | { | |
589 | net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, | |
590 | &net->ipv4.nat_vmalloced); | |
591 | if (!net->ipv4.nat_bysource) | |
592 | return -ENOMEM; | |
593 | return 0; | |
594 | } | |
595 | ||
596 | /* Clear NAT section of all conntracks, in case we're loaded again. */ | |
597 | static int clean_nat(struct nf_conn *i, void *data) | |
598 | { | |
599 | struct nf_conn_nat *nat = nfct_nat(i); | |
600 | ||
601 | if (!nat) | |
602 | return 0; | |
603 | memset(nat, 0, sizeof(*nat)); | |
604 | i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST); | |
605 | return 0; | |
606 | } | |
607 | ||
608 | static void __net_exit nf_nat_net_exit(struct net *net) | |
609 | { | |
610 | nf_ct_iterate_cleanup(net, &clean_nat, NULL); | |
611 | synchronize_rcu(); | |
612 | nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_vmalloced, | |
613 | nf_nat_htable_size); | |
614 | } | |
615 | ||
616 | static struct pernet_operations nf_nat_net_ops = { | |
617 | .init = nf_nat_net_init, | |
618 | .exit = nf_nat_net_exit, | |
619 | }; | |
620 | ||
5b1158e9 JK |
621 | static int __init nf_nat_init(void) |
622 | { | |
623 | size_t i; | |
2d59e5ca YK |
624 | int ret; |
625 | ||
475959d4 JE |
626 | need_ipv4_conntrack(); |
627 | ||
2d59e5ca YK |
628 | ret = nf_ct_extend_register(&nat_extend); |
629 | if (ret < 0) { | |
630 | printk(KERN_ERR "nf_nat_core: Unable to register extension\n"); | |
631 | return ret; | |
632 | } | |
5b1158e9 JK |
633 | |
634 | /* Leave them the same for the moment. */ | |
635 | nf_nat_htable_size = nf_conntrack_htable_size; | |
636 | ||
0c4c9288 AD |
637 | ret = register_pernet_subsys(&nf_nat_net_ops); |
638 | if (ret < 0) | |
2d59e5ca | 639 | goto cleanup_extend; |
5b1158e9 JK |
640 | |
641 | /* Sew in builtin protocols. */ | |
02502f62 | 642 | spin_lock_bh(&nf_nat_lock); |
5b1158e9 | 643 | for (i = 0; i < MAX_IP_NAT_PROTO; i++) |
e22a0548 PM |
644 | rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol); |
645 | rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp); | |
646 | rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp); | |
647 | rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp); | |
02502f62 | 648 | spin_unlock_bh(&nf_nat_lock); |
5b1158e9 | 649 | |
5b1158e9 JK |
650 | /* Initialize fake conntrack so that NAT will skip it */ |
651 | nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK; | |
652 | ||
653 | l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET); | |
dd13b010 PM |
654 | |
655 | BUG_ON(nf_nat_seq_adjust_hook != NULL); | |
656 | rcu_assign_pointer(nf_nat_seq_adjust_hook, nf_nat_seq_adjust); | |
5b1158e9 | 657 | return 0; |
2d59e5ca YK |
658 | |
659 | cleanup_extend: | |
660 | nf_ct_extend_unregister(&nat_extend); | |
661 | return ret; | |
5b1158e9 JK |
662 | } |
663 | ||
5b1158e9 JK |
664 | static void __exit nf_nat_cleanup(void) |
665 | { | |
0c4c9288 | 666 | unregister_pernet_subsys(&nf_nat_net_ops); |
5b1158e9 | 667 | nf_ct_l3proto_put(l3proto); |
2d59e5ca | 668 | nf_ct_extend_unregister(&nat_extend); |
dd13b010 PM |
669 | rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL); |
670 | synchronize_net(); | |
5b1158e9 JK |
671 | } |
672 | ||
673 | MODULE_LICENSE("GPL"); | |
674 | ||
675 | module_init(nf_nat_init); | |
676 | module_exit(nf_nat_cleanup); |