Commit | Line | Data |
---|---|---|
c7232c99 PM |
1 | /* |
2 | * (C) 1999-2001 Paul `Rusty' Russell | |
5b1158e9 | 3 | * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org> |
c7232c99 | 4 | * (C) 2011 Patrick McHardy <kaber@trash.net> |
5b1158e9 JK |
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> |
c7232c99 | 16 | #include <net/xfrm.h> |
5b1158e9 | 17 | #include <linux/jhash.h> |
c7232c99 | 18 | #include <linux/rtnetlink.h> |
5b1158e9 | 19 | |
5b1158e9 JK |
20 | #include <net/netfilter/nf_conntrack.h> |
21 | #include <net/netfilter/nf_conntrack_core.h> | |
22 | #include <net/netfilter/nf_nat.h> | |
c7232c99 PM |
23 | #include <net/netfilter/nf_nat_l3proto.h> |
24 | #include <net/netfilter/nf_nat_l4proto.h> | |
5b1158e9 JK |
25 | #include <net/netfilter/nf_nat_core.h> |
26 | #include <net/netfilter/nf_nat_helper.h> | |
27 | #include <net/netfilter/nf_conntrack_helper.h> | |
28 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
5d0aa2cc | 29 | #include <net/netfilter/nf_conntrack_zones.h> |
c7232c99 | 30 | #include <linux/netfilter/nf_nat.h> |
5b1158e9 | 31 | |
02502f62 | 32 | static DEFINE_SPINLOCK(nf_nat_lock); |
5b1158e9 | 33 | |
c7232c99 PM |
34 | static DEFINE_MUTEX(nf_nat_proto_mutex); |
35 | static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO] | |
36 | __read_mostly; | |
37 | static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO] | |
ce4b1ceb | 38 | __read_mostly; |
5b1158e9 | 39 | |
c7232c99 PM |
40 | |
41 | inline const struct nf_nat_l3proto * | |
42 | __nf_nat_l3proto_find(u8 family) | |
5b1158e9 | 43 | { |
c7232c99 | 44 | return rcu_dereference(nf_nat_l3protos[family]); |
5b1158e9 JK |
45 | } |
46 | ||
c7232c99 PM |
47 | inline const struct nf_nat_l4proto * |
48 | __nf_nat_l4proto_find(u8 family, u8 protonum) | |
49 | { | |
50 | return rcu_dereference(nf_nat_l4protos[family][protonum]); | |
51 | } | |
52 | EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find); | |
53 | ||
54 | #ifdef CONFIG_XFRM | |
55 | static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl) | |
56 | { | |
57 | const struct nf_nat_l3proto *l3proto; | |
58 | const struct nf_conn *ct; | |
59 | enum ip_conntrack_info ctinfo; | |
60 | enum ip_conntrack_dir dir; | |
61 | unsigned long statusbit; | |
62 | u8 family; | |
63 | ||
64 | ct = nf_ct_get(skb, &ctinfo); | |
65 | if (ct == NULL) | |
66 | return; | |
67 | ||
68 | family = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.l3num; | |
69 | rcu_read_lock(); | |
70 | l3proto = __nf_nat_l3proto_find(family); | |
71 | if (l3proto == NULL) | |
72 | goto out; | |
73 | ||
74 | dir = CTINFO2DIR(ctinfo); | |
75 | if (dir == IP_CT_DIR_ORIGINAL) | |
76 | statusbit = IPS_DST_NAT; | |
77 | else | |
78 | statusbit = IPS_SRC_NAT; | |
79 | ||
80 | l3proto->decode_session(skb, ct, dir, statusbit, fl); | |
81 | out: | |
82 | rcu_read_unlock(); | |
83 | } | |
84 | ||
85 | int nf_xfrm_me_harder(struct sk_buff *skb, unsigned int family) | |
86 | { | |
87 | struct flowi fl; | |
88 | unsigned int hh_len; | |
89 | struct dst_entry *dst; | |
90 | ||
91 | if (xfrm_decode_session(skb, &fl, family) < 0) | |
92 | return -1; | |
93 | ||
94 | dst = skb_dst(skb); | |
95 | if (dst->xfrm) | |
96 | dst = ((struct xfrm_dst *)dst)->route; | |
97 | dst_hold(dst); | |
98 | ||
99 | dst = xfrm_lookup(dev_net(dst->dev), dst, &fl, skb->sk, 0); | |
100 | if (IS_ERR(dst)) | |
101 | return -1; | |
102 | ||
103 | skb_dst_drop(skb); | |
104 | skb_dst_set(skb, dst); | |
105 | ||
106 | /* Change in oif may mean change in hh_len. */ | |
107 | hh_len = skb_dst(skb)->dev->hard_header_len; | |
108 | if (skb_headroom(skb) < hh_len && | |
109 | pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC)) | |
110 | return -1; | |
111 | return 0; | |
112 | } | |
113 | EXPORT_SYMBOL(nf_xfrm_me_harder); | |
114 | #endif /* CONFIG_XFRM */ | |
115 | ||
5b1158e9 JK |
116 | /* We keep an extra hash for each conntrack, for fast searching. */ |
117 | static inline unsigned int | |
5d0aa2cc PM |
118 | hash_by_src(const struct net *net, u16 zone, |
119 | const struct nf_conntrack_tuple *tuple) | |
5b1158e9 | 120 | { |
34498825 PM |
121 | unsigned int hash; |
122 | ||
5b1158e9 | 123 | /* Original src, to ensure we map it consistently if poss. */ |
c7232c99 PM |
124 | hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32), |
125 | tuple->dst.protonum ^ zone ^ nf_conntrack_hash_rnd); | |
126 | return ((u64)hash * net->ct.nat_htable_size) >> 32; | |
5b1158e9 JK |
127 | } |
128 | ||
5b1158e9 JK |
129 | /* Is this tuple already taken? (not by us) */ |
130 | int | |
131 | nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple, | |
132 | const struct nf_conn *ignored_conntrack) | |
133 | { | |
134 | /* Conntrack tracking doesn't keep track of outgoing tuples; only | |
c7232c99 PM |
135 | * incoming ones. NAT means they don't have a fixed mapping, |
136 | * so we invert the tuple and look for the incoming reply. | |
137 | * | |
138 | * We could keep a separate hash if this proves too slow. | |
139 | */ | |
5b1158e9 JK |
140 | struct nf_conntrack_tuple reply; |
141 | ||
142 | nf_ct_invert_tuplepr(&reply, tuple); | |
143 | return nf_conntrack_tuple_taken(&reply, ignored_conntrack); | |
144 | } | |
145 | EXPORT_SYMBOL(nf_nat_used_tuple); | |
146 | ||
147 | /* If we source map this tuple so reply looks like reply_tuple, will | |
c7232c99 PM |
148 | * that meet the constraints of range. |
149 | */ | |
150 | static int in_range(const struct nf_nat_l3proto *l3proto, | |
151 | const struct nf_nat_l4proto *l4proto, | |
152 | const struct nf_conntrack_tuple *tuple, | |
153 | const struct nf_nat_range *range) | |
5b1158e9 | 154 | { |
5b1158e9 | 155 | /* If we are supposed to map IPs, then we must be in the |
c7232c99 PM |
156 | * range specified, otherwise let this drag us onto a new src IP. |
157 | */ | |
158 | if (range->flags & NF_NAT_RANGE_MAP_IPS && | |
159 | !l3proto->in_range(tuple, range)) | |
160 | return 0; | |
5b1158e9 | 161 | |
cbc9f2f4 | 162 | if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) || |
c7232c99 PM |
163 | l4proto->in_range(tuple, NF_NAT_MANIP_SRC, |
164 | &range->min_proto, &range->max_proto)) | |
165 | return 1; | |
5b1158e9 | 166 | |
c7232c99 | 167 | return 0; |
5b1158e9 JK |
168 | } |
169 | ||
170 | static inline int | |
171 | same_src(const struct nf_conn *ct, | |
172 | const struct nf_conntrack_tuple *tuple) | |
173 | { | |
174 | const struct nf_conntrack_tuple *t; | |
175 | ||
176 | t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple; | |
177 | return (t->dst.protonum == tuple->dst.protonum && | |
c7232c99 | 178 | nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) && |
5b1158e9 JK |
179 | t->src.u.all == tuple->src.u.all); |
180 | } | |
181 | ||
182 | /* Only called for SRC manip */ | |
183 | static int | |
5d0aa2cc | 184 | find_appropriate_src(struct net *net, u16 zone, |
c7232c99 PM |
185 | const struct nf_nat_l3proto *l3proto, |
186 | const struct nf_nat_l4proto *l4proto, | |
0c4c9288 | 187 | const struct nf_conntrack_tuple *tuple, |
5b1158e9 | 188 | struct nf_conntrack_tuple *result, |
c7232c99 | 189 | const struct nf_nat_range *range) |
5b1158e9 | 190 | { |
5d0aa2cc | 191 | unsigned int h = hash_by_src(net, zone, tuple); |
72b72949 JE |
192 | const struct nf_conn_nat *nat; |
193 | const struct nf_conn *ct; | |
194 | const struct hlist_node *n; | |
5b1158e9 | 195 | |
c7232c99 | 196 | hlist_for_each_entry_rcu(nat, n, &net->ct.nat_bysource[h], bysource) { |
b6b84d4a | 197 | ct = nat->ct; |
5d0aa2cc | 198 | if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) { |
5b1158e9 JK |
199 | /* Copy source part from reply tuple. */ |
200 | nf_ct_invert_tuplepr(result, | |
201 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
202 | result->dst = tuple->dst; | |
203 | ||
c7232c99 | 204 | if (in_range(l3proto, l4proto, result, range)) { |
4d354c57 | 205 | rcu_read_unlock(); |
5b1158e9 JK |
206 | return 1; |
207 | } | |
208 | } | |
209 | } | |
5b1158e9 JK |
210 | return 0; |
211 | } | |
212 | ||
213 | /* For [FUTURE] fragmentation handling, we want the least-used | |
c7232c99 PM |
214 | * src-ip/dst-ip/proto triple. Fairness doesn't come into it. Thus |
215 | * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports | |
216 | * 1-65535, we don't do pro-rata allocation based on ports; we choose | |
217 | * the ip with the lowest src-ip/dst-ip/proto usage. | |
218 | */ | |
5b1158e9 | 219 | static void |
5d0aa2cc | 220 | find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple, |
c7232c99 | 221 | const struct nf_nat_range *range, |
5b1158e9 JK |
222 | const struct nf_conn *ct, |
223 | enum nf_nat_manip_type maniptype) | |
224 | { | |
c7232c99 PM |
225 | union nf_inet_addr *var_ipp; |
226 | unsigned int i, max; | |
5b1158e9 | 227 | /* Host order */ |
c7232c99 PM |
228 | u32 minip, maxip, j, dist; |
229 | bool full_range; | |
5b1158e9 JK |
230 | |
231 | /* No IP mapping? Do nothing. */ | |
cbc9f2f4 | 232 | if (!(range->flags & NF_NAT_RANGE_MAP_IPS)) |
5b1158e9 JK |
233 | return; |
234 | ||
cbc9f2f4 | 235 | if (maniptype == NF_NAT_MANIP_SRC) |
c7232c99 | 236 | var_ipp = &tuple->src.u3; |
5b1158e9 | 237 | else |
c7232c99 | 238 | var_ipp = &tuple->dst.u3; |
5b1158e9 JK |
239 | |
240 | /* Fast path: only one choice. */ | |
c7232c99 PM |
241 | if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) { |
242 | *var_ipp = range->min_addr; | |
5b1158e9 JK |
243 | return; |
244 | } | |
245 | ||
c7232c99 PM |
246 | if (nf_ct_l3num(ct) == NFPROTO_IPV4) |
247 | max = sizeof(var_ipp->ip) / sizeof(u32) - 1; | |
248 | else | |
249 | max = sizeof(var_ipp->ip6) / sizeof(u32) - 1; | |
250 | ||
5b1158e9 JK |
251 | /* Hashing source and destination IPs gives a fairly even |
252 | * spread in practice (if there are a small number of IPs | |
253 | * involved, there usually aren't that many connections | |
254 | * anyway). The consistency means that servers see the same | |
255 | * client coming from the same IP (some Internet Banking sites | |
c7232c99 PM |
256 | * like this), even across reboots. |
257 | */ | |
258 | j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3), | |
259 | range->flags & NF_NAT_RANGE_PERSISTENT ? | |
260 | 0 : (__force u32)tuple->dst.u3.all[max] ^ zone); | |
261 | ||
262 | full_range = false; | |
263 | for (i = 0; i <= max; i++) { | |
264 | /* If first bytes of the address are at the maximum, use the | |
265 | * distance. Otherwise use the full range. | |
266 | */ | |
267 | if (!full_range) { | |
268 | minip = ntohl((__force __be32)range->min_addr.all[i]); | |
269 | maxip = ntohl((__force __be32)range->max_addr.all[i]); | |
270 | dist = maxip - minip + 1; | |
271 | } else { | |
272 | minip = 0; | |
273 | dist = ~0; | |
274 | } | |
275 | ||
276 | var_ipp->all[i] = (__force __u32) | |
277 | htonl(minip + (((u64)j * dist) >> 32)); | |
278 | if (var_ipp->all[i] != range->max_addr.all[i]) | |
279 | full_range = true; | |
280 | ||
281 | if (!(range->flags & NF_NAT_RANGE_PERSISTENT)) | |
282 | j ^= (__force u32)tuple->dst.u3.all[i]; | |
283 | } | |
5b1158e9 JK |
284 | } |
285 | ||
c7232c99 PM |
286 | /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING, |
287 | * we change the source to map into the range. For NF_INET_PRE_ROUTING | |
6e23ae2a | 288 | * and NF_INET_LOCAL_OUT, we change the destination to map into the |
c7232c99 | 289 | * range. It might not be possible to get a unique tuple, but we try. |
5b1158e9 JK |
290 | * At worst (or if we race), we will end up with a final duplicate in |
291 | * __ip_conntrack_confirm and drop the packet. */ | |
292 | static void | |
293 | get_unique_tuple(struct nf_conntrack_tuple *tuple, | |
294 | const struct nf_conntrack_tuple *orig_tuple, | |
c7232c99 | 295 | const struct nf_nat_range *range, |
5b1158e9 JK |
296 | struct nf_conn *ct, |
297 | enum nf_nat_manip_type maniptype) | |
298 | { | |
c7232c99 PM |
299 | const struct nf_nat_l3proto *l3proto; |
300 | const struct nf_nat_l4proto *l4proto; | |
0c4c9288 | 301 | struct net *net = nf_ct_net(ct); |
5d0aa2cc | 302 | u16 zone = nf_ct_zone(ct); |
5b1158e9 | 303 | |
c7232c99 PM |
304 | rcu_read_lock(); |
305 | l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num); | |
306 | l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num, | |
307 | orig_tuple->dst.protonum); | |
5b1158e9 | 308 | |
c7232c99 PM |
309 | /* 1) If this srcip/proto/src-proto-part is currently mapped, |
310 | * and that same mapping gives a unique tuple within the given | |
311 | * range, use that. | |
312 | * | |
313 | * This is only required for source (ie. NAT/masq) mappings. | |
314 | * So far, we don't do local source mappings, so multiple | |
315 | * manips not an issue. | |
316 | */ | |
cbc9f2f4 PM |
317 | if (maniptype == NF_NAT_MANIP_SRC && |
318 | !(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) { | |
41a7cab6 | 319 | /* try the original tuple first */ |
c7232c99 | 320 | if (in_range(l3proto, l4proto, orig_tuple, range)) { |
41a7cab6 CG |
321 | if (!nf_nat_used_tuple(orig_tuple, ct)) { |
322 | *tuple = *orig_tuple; | |
c7232c99 | 323 | goto out; |
41a7cab6 | 324 | } |
c7232c99 PM |
325 | } else if (find_appropriate_src(net, zone, l3proto, l4proto, |
326 | orig_tuple, tuple, range)) { | |
0d53778e | 327 | pr_debug("get_unique_tuple: Found current src map\n"); |
0dbff689 | 328 | if (!nf_nat_used_tuple(tuple, ct)) |
c7232c99 | 329 | goto out; |
5b1158e9 JK |
330 | } |
331 | } | |
332 | ||
c7232c99 | 333 | /* 2) Select the least-used IP/proto combination in the given range */ |
5b1158e9 | 334 | *tuple = *orig_tuple; |
5d0aa2cc | 335 | find_best_ips_proto(zone, tuple, range, ct, maniptype); |
5b1158e9 JK |
336 | |
337 | /* 3) The per-protocol part of the manip is made to map into | |
c7232c99 PM |
338 | * the range to make a unique tuple. |
339 | */ | |
5b1158e9 JK |
340 | |
341 | /* Only bother mapping if it's not already in range and unique */ | |
cbc9f2f4 PM |
342 | if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) { |
343 | if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) { | |
c7232c99 PM |
344 | if (l4proto->in_range(tuple, maniptype, |
345 | &range->min_proto, | |
346 | &range->max_proto) && | |
347 | (range->min_proto.all == range->max_proto.all || | |
99ad3c53 CG |
348 | !nf_nat_used_tuple(tuple, ct))) |
349 | goto out; | |
350 | } else if (!nf_nat_used_tuple(tuple, ct)) { | |
351 | goto out; | |
352 | } | |
353 | } | |
5b1158e9 JK |
354 | |
355 | /* Last change: get protocol to try to obtain unique tuple. */ | |
c7232c99 | 356 | l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct); |
e22a0548 PM |
357 | out: |
358 | rcu_read_unlock(); | |
5b1158e9 JK |
359 | } |
360 | ||
361 | unsigned int | |
362 | nf_nat_setup_info(struct nf_conn *ct, | |
c7232c99 | 363 | const struct nf_nat_range *range, |
cc01dcbd | 364 | enum nf_nat_manip_type maniptype) |
5b1158e9 | 365 | { |
0c4c9288 | 366 | struct net *net = nf_ct_net(ct); |
5b1158e9 | 367 | struct nf_conntrack_tuple curr_tuple, new_tuple; |
2d59e5ca | 368 | struct nf_conn_nat *nat; |
5b1158e9 | 369 | |
2d59e5ca YK |
370 | /* nat helper or nfctnetlink also setup binding */ |
371 | nat = nfct_nat(ct); | |
372 | if (!nat) { | |
373 | nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC); | |
374 | if (nat == NULL) { | |
0d53778e | 375 | pr_debug("failed to add NAT extension\n"); |
2d59e5ca YK |
376 | return NF_ACCEPT; |
377 | } | |
378 | } | |
379 | ||
cbc9f2f4 PM |
380 | NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC || |
381 | maniptype == NF_NAT_MANIP_DST); | |
5b1158e9 JK |
382 | BUG_ON(nf_nat_initialized(ct, maniptype)); |
383 | ||
384 | /* What we've got will look like inverse of reply. Normally | |
c7232c99 PM |
385 | * this is what is in the conntrack, except for prior |
386 | * manipulations (future optimization: if num_manips == 0, | |
387 | * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple) | |
388 | */ | |
5b1158e9 JK |
389 | nf_ct_invert_tuplepr(&curr_tuple, |
390 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
391 | ||
392 | get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype); | |
393 | ||
394 | if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) { | |
395 | struct nf_conntrack_tuple reply; | |
396 | ||
397 | /* Alter conntrack table so will recognize replies. */ | |
398 | nf_ct_invert_tuplepr(&reply, &new_tuple); | |
399 | nf_conntrack_alter_reply(ct, &reply); | |
400 | ||
401 | /* Non-atomic: we own this at the moment. */ | |
cbc9f2f4 | 402 | if (maniptype == NF_NAT_MANIP_SRC) |
5b1158e9 JK |
403 | ct->status |= IPS_SRC_NAT; |
404 | else | |
405 | ct->status |= IPS_DST_NAT; | |
406 | } | |
407 | ||
cbc9f2f4 | 408 | if (maniptype == NF_NAT_MANIP_SRC) { |
5b1158e9 JK |
409 | unsigned int srchash; |
410 | ||
5d0aa2cc PM |
411 | srchash = hash_by_src(net, nf_ct_zone(ct), |
412 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
02502f62 | 413 | spin_lock_bh(&nf_nat_lock); |
c7232c99 | 414 | /* nf_conntrack_alter_reply might re-allocate extension aera */ |
b6b84d4a YK |
415 | nat = nfct_nat(ct); |
416 | nat->ct = ct; | |
0c4c9288 | 417 | hlist_add_head_rcu(&nat->bysource, |
c7232c99 | 418 | &net->ct.nat_bysource[srchash]); |
02502f62 | 419 | spin_unlock_bh(&nf_nat_lock); |
5b1158e9 JK |
420 | } |
421 | ||
422 | /* It's done. */ | |
cbc9f2f4 | 423 | if (maniptype == NF_NAT_MANIP_DST) |
a7c2f4d7 | 424 | ct->status |= IPS_DST_NAT_DONE; |
5b1158e9 | 425 | else |
a7c2f4d7 | 426 | ct->status |= IPS_SRC_NAT_DONE; |
5b1158e9 JK |
427 | |
428 | return NF_ACCEPT; | |
429 | } | |
430 | EXPORT_SYMBOL(nf_nat_setup_info); | |
431 | ||
5b1158e9 JK |
432 | /* Do packet manipulations according to nf_nat_setup_info. */ |
433 | unsigned int nf_nat_packet(struct nf_conn *ct, | |
434 | enum ip_conntrack_info ctinfo, | |
435 | unsigned int hooknum, | |
3db05fea | 436 | struct sk_buff *skb) |
5b1158e9 | 437 | { |
c7232c99 PM |
438 | const struct nf_nat_l3proto *l3proto; |
439 | const struct nf_nat_l4proto *l4proto; | |
5b1158e9 JK |
440 | enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo); |
441 | unsigned long statusbit; | |
442 | enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum); | |
443 | ||
cbc9f2f4 | 444 | if (mtype == NF_NAT_MANIP_SRC) |
5b1158e9 JK |
445 | statusbit = IPS_SRC_NAT; |
446 | else | |
447 | statusbit = IPS_DST_NAT; | |
448 | ||
449 | /* Invert if this is reply dir. */ | |
450 | if (dir == IP_CT_DIR_REPLY) | |
451 | statusbit ^= IPS_NAT_MASK; | |
452 | ||
453 | /* Non-atomic: these bits don't change. */ | |
454 | if (ct->status & statusbit) { | |
455 | struct nf_conntrack_tuple target; | |
456 | ||
457 | /* We are aiming to look like inverse of other direction. */ | |
458 | nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple); | |
459 | ||
c7232c99 PM |
460 | l3proto = __nf_nat_l3proto_find(target.src.l3num); |
461 | l4proto = __nf_nat_l4proto_find(target.src.l3num, | |
462 | target.dst.protonum); | |
463 | if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype)) | |
5b1158e9 JK |
464 | return NF_DROP; |
465 | } | |
466 | return NF_ACCEPT; | |
467 | } | |
468 | EXPORT_SYMBOL_GPL(nf_nat_packet); | |
469 | ||
c7232c99 PM |
470 | struct nf_nat_proto_clean { |
471 | u8 l3proto; | |
472 | u8 l4proto; | |
473 | bool hash; | |
474 | }; | |
475 | ||
476 | /* Clear NAT section of all conntracks, in case we're loaded again. */ | |
477 | static int nf_nat_proto_clean(struct nf_conn *i, void *data) | |
5b1158e9 | 478 | { |
c7232c99 PM |
479 | const struct nf_nat_proto_clean *clean = data; |
480 | struct nf_conn_nat *nat = nfct_nat(i); | |
5b1158e9 | 481 | |
c7232c99 | 482 | if (!nat) |
5b1158e9 | 483 | return 0; |
c7232c99 PM |
484 | if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) || |
485 | (clean->l4proto && nf_ct_protonum(i) != clean->l4proto)) | |
5b1158e9 JK |
486 | return 0; |
487 | ||
c7232c99 PM |
488 | if (clean->hash) { |
489 | spin_lock_bh(&nf_nat_lock); | |
490 | hlist_del_rcu(&nat->bysource); | |
491 | spin_unlock_bh(&nf_nat_lock); | |
492 | } else { | |
493 | memset(nat, 0, sizeof(*nat)); | |
494 | i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | | |
495 | IPS_SEQ_ADJUST); | |
5b1158e9 | 496 | } |
c7232c99 PM |
497 | return 0; |
498 | } | |
5b1158e9 | 499 | |
c7232c99 PM |
500 | static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto) |
501 | { | |
502 | struct nf_nat_proto_clean clean = { | |
503 | .l3proto = l3proto, | |
504 | .l4proto = l4proto, | |
505 | }; | |
506 | struct net *net; | |
507 | ||
508 | rtnl_lock(); | |
509 | /* Step 1 - remove from bysource hash */ | |
510 | clean.hash = true; | |
511 | for_each_net(net) | |
512 | nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean); | |
513 | synchronize_rcu(); | |
5b1158e9 | 514 | |
c7232c99 PM |
515 | /* Step 2 - clean NAT section */ |
516 | clean.hash = false; | |
517 | for_each_net(net) | |
518 | nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean); | |
519 | rtnl_unlock(); | |
520 | } | |
5b1158e9 | 521 | |
c7232c99 PM |
522 | static void nf_nat_l3proto_clean(u8 l3proto) |
523 | { | |
524 | struct nf_nat_proto_clean clean = { | |
525 | .l3proto = l3proto, | |
526 | }; | |
527 | struct net *net; | |
528 | ||
529 | rtnl_lock(); | |
530 | /* Step 1 - remove from bysource hash */ | |
531 | clean.hash = true; | |
532 | for_each_net(net) | |
533 | nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean); | |
534 | synchronize_rcu(); | |
5b1158e9 | 535 | |
c7232c99 PM |
536 | /* Step 2 - clean NAT section */ |
537 | clean.hash = false; | |
538 | for_each_net(net) | |
539 | nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean); | |
540 | rtnl_unlock(); | |
5b1158e9 | 541 | } |
5b1158e9 JK |
542 | |
543 | /* Protocol registration. */ | |
c7232c99 | 544 | int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto) |
5b1158e9 | 545 | { |
c7232c99 PM |
546 | const struct nf_nat_l4proto **l4protos; |
547 | unsigned int i; | |
5b1158e9 JK |
548 | int ret = 0; |
549 | ||
c7232c99 PM |
550 | mutex_lock(&nf_nat_proto_mutex); |
551 | if (nf_nat_l4protos[l3proto] == NULL) { | |
552 | l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *), | |
553 | GFP_KERNEL); | |
554 | if (l4protos == NULL) { | |
555 | ret = -ENOMEM; | |
556 | goto out; | |
557 | } | |
558 | ||
559 | for (i = 0; i < IPPROTO_MAX; i++) | |
560 | RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown); | |
561 | ||
562 | /* Before making proto_array visible to lockless readers, | |
563 | * we must make sure its content is committed to memory. | |
564 | */ | |
565 | smp_wmb(); | |
566 | ||
567 | nf_nat_l4protos[l3proto] = l4protos; | |
568 | } | |
569 | ||
eb733162 | 570 | if (rcu_dereference_protected( |
c7232c99 PM |
571 | nf_nat_l4protos[l3proto][l4proto->l4proto], |
572 | lockdep_is_held(&nf_nat_proto_mutex) | |
573 | ) != &nf_nat_l4proto_unknown) { | |
5b1158e9 JK |
574 | ret = -EBUSY; |
575 | goto out; | |
576 | } | |
c7232c99 | 577 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto); |
5b1158e9 | 578 | out: |
c7232c99 | 579 | mutex_unlock(&nf_nat_proto_mutex); |
5b1158e9 JK |
580 | return ret; |
581 | } | |
c7232c99 | 582 | EXPORT_SYMBOL_GPL(nf_nat_l4proto_register); |
5b1158e9 | 583 | |
25985edc | 584 | /* No one stores the protocol anywhere; simply delete it. */ |
c7232c99 | 585 | void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto) |
5b1158e9 | 586 | { |
c7232c99 PM |
587 | mutex_lock(&nf_nat_proto_mutex); |
588 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], | |
589 | &nf_nat_l4proto_unknown); | |
590 | mutex_unlock(&nf_nat_proto_mutex); | |
e22a0548 | 591 | synchronize_rcu(); |
c7232c99 PM |
592 | |
593 | nf_nat_l4proto_clean(l3proto, l4proto->l4proto); | |
5b1158e9 | 594 | } |
c7232c99 PM |
595 | EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister); |
596 | ||
597 | int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto) | |
598 | { | |
599 | int err; | |
600 | ||
601 | err = nf_ct_l3proto_try_module_get(l3proto->l3proto); | |
602 | if (err < 0) | |
603 | return err; | |
604 | ||
605 | mutex_lock(&nf_nat_proto_mutex); | |
606 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP], | |
607 | &nf_nat_l4proto_tcp); | |
608 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP], | |
609 | &nf_nat_l4proto_udp); | |
610 | mutex_unlock(&nf_nat_proto_mutex); | |
611 | ||
612 | RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto); | |
613 | return 0; | |
614 | } | |
615 | EXPORT_SYMBOL_GPL(nf_nat_l3proto_register); | |
616 | ||
617 | void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto) | |
618 | { | |
619 | mutex_lock(&nf_nat_proto_mutex); | |
620 | RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL); | |
621 | mutex_unlock(&nf_nat_proto_mutex); | |
622 | synchronize_rcu(); | |
623 | ||
624 | nf_nat_l3proto_clean(l3proto->l3proto); | |
625 | nf_ct_l3proto_module_put(l3proto->l3proto); | |
626 | } | |
627 | EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister); | |
5b1158e9 | 628 | |
25985edc | 629 | /* No one using conntrack by the time this called. */ |
d8a0509a YK |
630 | static void nf_nat_cleanup_conntrack(struct nf_conn *ct) |
631 | { | |
632 | struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT); | |
633 | ||
b6b84d4a | 634 | if (nat == NULL || nat->ct == NULL) |
d8a0509a YK |
635 | return; |
636 | ||
41a7cab6 | 637 | NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE); |
d8a0509a | 638 | |
02502f62 | 639 | spin_lock_bh(&nf_nat_lock); |
4d354c57 | 640 | hlist_del_rcu(&nat->bysource); |
02502f62 | 641 | spin_unlock_bh(&nf_nat_lock); |
d8a0509a YK |
642 | } |
643 | ||
86577c66 | 644 | static void nf_nat_move_storage(void *new, void *old) |
2d59e5ca | 645 | { |
86577c66 PM |
646 | struct nf_conn_nat *new_nat = new; |
647 | struct nf_conn_nat *old_nat = old; | |
b6b84d4a | 648 | struct nf_conn *ct = old_nat->ct; |
2d59e5ca | 649 | |
41a7cab6 | 650 | if (!ct || !(ct->status & IPS_SRC_NAT_DONE)) |
2d59e5ca YK |
651 | return; |
652 | ||
02502f62 | 653 | spin_lock_bh(&nf_nat_lock); |
68b80f11 | 654 | hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource); |
02502f62 | 655 | spin_unlock_bh(&nf_nat_lock); |
2d59e5ca YK |
656 | } |
657 | ||
61eb3107 | 658 | static struct nf_ct_ext_type nat_extend __read_mostly = { |
d8a0509a YK |
659 | .len = sizeof(struct nf_conn_nat), |
660 | .align = __alignof__(struct nf_conn_nat), | |
661 | .destroy = nf_nat_cleanup_conntrack, | |
662 | .move = nf_nat_move_storage, | |
663 | .id = NF_CT_EXT_NAT, | |
664 | .flags = NF_CT_EXT_F_PREALLOC, | |
2d59e5ca YK |
665 | }; |
666 | ||
e6a7d3c0 PNA |
667 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
668 | ||
669 | #include <linux/netfilter/nfnetlink.h> | |
670 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
671 | ||
672 | static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = { | |
673 | [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 }, | |
674 | [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 }, | |
675 | }; | |
676 | ||
677 | static int nfnetlink_parse_nat_proto(struct nlattr *attr, | |
678 | const struct nf_conn *ct, | |
c7232c99 | 679 | struct nf_nat_range *range) |
e6a7d3c0 PNA |
680 | { |
681 | struct nlattr *tb[CTA_PROTONAT_MAX+1]; | |
c7232c99 | 682 | const struct nf_nat_l4proto *l4proto; |
e6a7d3c0 PNA |
683 | int err; |
684 | ||
685 | err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy); | |
686 | if (err < 0) | |
687 | return err; | |
688 | ||
c7232c99 PM |
689 | l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
690 | if (l4proto->nlattr_to_range) | |
691 | err = l4proto->nlattr_to_range(tb, range); | |
692 | ||
e6a7d3c0 PNA |
693 | return err; |
694 | } | |
695 | ||
696 | static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = { | |
c7232c99 PM |
697 | [CTA_NAT_V4_MINIP] = { .type = NLA_U32 }, |
698 | [CTA_NAT_V4_MAXIP] = { .type = NLA_U32 }, | |
58a317f1 PM |
699 | [CTA_NAT_V6_MINIP] = { .len = sizeof(struct in6_addr) }, |
700 | [CTA_NAT_V6_MAXIP] = { .len = sizeof(struct in6_addr) }, | |
329fb58a | 701 | [CTA_NAT_PROTO] = { .type = NLA_NESTED }, |
e6a7d3c0 PNA |
702 | }; |
703 | ||
704 | static int | |
39938324 | 705 | nfnetlink_parse_nat(const struct nlattr *nat, |
c7232c99 | 706 | const struct nf_conn *ct, struct nf_nat_range *range) |
e6a7d3c0 | 707 | { |
c7232c99 | 708 | const struct nf_nat_l3proto *l3proto; |
e6a7d3c0 PNA |
709 | struct nlattr *tb[CTA_NAT_MAX+1]; |
710 | int err; | |
711 | ||
712 | memset(range, 0, sizeof(*range)); | |
713 | ||
714 | err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy); | |
715 | if (err < 0) | |
716 | return err; | |
717 | ||
c7232c99 PM |
718 | rcu_read_lock(); |
719 | l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct)); | |
720 | if (l3proto == NULL) { | |
721 | err = -EAGAIN; | |
722 | goto out; | |
723 | } | |
724 | err = l3proto->nlattr_to_range(tb, range); | |
725 | if (err < 0) | |
726 | goto out; | |
e6a7d3c0 PNA |
727 | |
728 | if (!tb[CTA_NAT_PROTO]) | |
c7232c99 | 729 | goto out; |
e6a7d3c0 PNA |
730 | |
731 | err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range); | |
c7232c99 PM |
732 | out: |
733 | rcu_read_unlock(); | |
734 | return err; | |
e6a7d3c0 PNA |
735 | } |
736 | ||
737 | static int | |
738 | nfnetlink_parse_nat_setup(struct nf_conn *ct, | |
739 | enum nf_nat_manip_type manip, | |
39938324 | 740 | const struct nlattr *attr) |
e6a7d3c0 | 741 | { |
c7232c99 PM |
742 | struct nf_nat_range range; |
743 | int err; | |
e6a7d3c0 | 744 | |
c7232c99 PM |
745 | err = nfnetlink_parse_nat(attr, ct, &range); |
746 | if (err < 0) | |
747 | return err; | |
e6a7d3c0 PNA |
748 | if (nf_nat_initialized(ct, manip)) |
749 | return -EEXIST; | |
750 | ||
751 | return nf_nat_setup_info(ct, &range, manip); | |
752 | } | |
753 | #else | |
754 | static int | |
755 | nfnetlink_parse_nat_setup(struct nf_conn *ct, | |
756 | enum nf_nat_manip_type manip, | |
39938324 | 757 | const struct nlattr *attr) |
e6a7d3c0 PNA |
758 | { |
759 | return -EOPNOTSUPP; | |
760 | } | |
761 | #endif | |
762 | ||
0c4c9288 AD |
763 | static int __net_init nf_nat_net_init(struct net *net) |
764 | { | |
d696c7bd | 765 | /* Leave them the same for the moment. */ |
c7232c99 PM |
766 | net->ct.nat_htable_size = net->ct.htable_size; |
767 | net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0); | |
768 | if (!net->ct.nat_bysource) | |
0c4c9288 AD |
769 | return -ENOMEM; |
770 | return 0; | |
771 | } | |
772 | ||
0c4c9288 AD |
773 | static void __net_exit nf_nat_net_exit(struct net *net) |
774 | { | |
c7232c99 PM |
775 | struct nf_nat_proto_clean clean = {}; |
776 | ||
777 | nf_ct_iterate_cleanup(net, &nf_nat_proto_clean, &clean); | |
0c4c9288 | 778 | synchronize_rcu(); |
c7232c99 | 779 | nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size); |
0c4c9288 AD |
780 | } |
781 | ||
782 | static struct pernet_operations nf_nat_net_ops = { | |
783 | .init = nf_nat_net_init, | |
784 | .exit = nf_nat_net_exit, | |
785 | }; | |
786 | ||
544d5c7d PNA |
787 | static struct nf_ct_helper_expectfn follow_master_nat = { |
788 | .name = "nat-follow-master", | |
789 | .expectfn = nf_nat_follow_master, | |
790 | }; | |
791 | ||
d584a61a PNA |
792 | static struct nfq_ct_nat_hook nfq_ct_nat = { |
793 | .seq_adjust = nf_nat_tcp_seq_adjust, | |
794 | }; | |
795 | ||
5b1158e9 JK |
796 | static int __init nf_nat_init(void) |
797 | { | |
2d59e5ca YK |
798 | int ret; |
799 | ||
800 | ret = nf_ct_extend_register(&nat_extend); | |
801 | if (ret < 0) { | |
802 | printk(KERN_ERR "nf_nat_core: Unable to register extension\n"); | |
803 | return ret; | |
804 | } | |
5b1158e9 | 805 | |
0c4c9288 AD |
806 | ret = register_pernet_subsys(&nf_nat_net_ops); |
807 | if (ret < 0) | |
2d59e5ca | 808 | goto cleanup_extend; |
5b1158e9 | 809 | |
c7232c99 | 810 | nf_ct_helper_expectfn_register(&follow_master_nat); |
5b1158e9 | 811 | |
5b1158e9 | 812 | /* Initialize fake conntrack so that NAT will skip it */ |
5bfddbd4 | 813 | nf_ct_untracked_status_or(IPS_NAT_DONE_MASK); |
5b1158e9 | 814 | |
dd13b010 | 815 | BUG_ON(nf_nat_seq_adjust_hook != NULL); |
a9b3cd7f | 816 | RCU_INIT_POINTER(nf_nat_seq_adjust_hook, nf_nat_seq_adjust); |
e6a7d3c0 | 817 | BUG_ON(nfnetlink_parse_nat_setup_hook != NULL); |
a9b3cd7f | 818 | RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, |
e6a7d3c0 | 819 | nfnetlink_parse_nat_setup); |
f9dd09c7 | 820 | BUG_ON(nf_ct_nat_offset != NULL); |
a9b3cd7f | 821 | RCU_INIT_POINTER(nf_ct_nat_offset, nf_nat_get_offset); |
d584a61a | 822 | RCU_INIT_POINTER(nfq_ct_nat_hook, &nfq_ct_nat); |
c7232c99 PM |
823 | #ifdef CONFIG_XFRM |
824 | BUG_ON(nf_nat_decode_session_hook != NULL); | |
825 | RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session); | |
826 | #endif | |
5b1158e9 | 827 | return 0; |
2d59e5ca YK |
828 | |
829 | cleanup_extend: | |
830 | nf_ct_extend_unregister(&nat_extend); | |
831 | return ret; | |
5b1158e9 JK |
832 | } |
833 | ||
5b1158e9 JK |
834 | static void __exit nf_nat_cleanup(void) |
835 | { | |
c7232c99 PM |
836 | unsigned int i; |
837 | ||
0c4c9288 | 838 | unregister_pernet_subsys(&nf_nat_net_ops); |
2d59e5ca | 839 | nf_ct_extend_unregister(&nat_extend); |
544d5c7d | 840 | nf_ct_helper_expectfn_unregister(&follow_master_nat); |
a9b3cd7f SH |
841 | RCU_INIT_POINTER(nf_nat_seq_adjust_hook, NULL); |
842 | RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL); | |
843 | RCU_INIT_POINTER(nf_ct_nat_offset, NULL); | |
d584a61a | 844 | RCU_INIT_POINTER(nfq_ct_nat_hook, NULL); |
c7232c99 PM |
845 | #ifdef CONFIG_XFRM |
846 | RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL); | |
847 | #endif | |
848 | for (i = 0; i < NFPROTO_NUMPROTO; i++) | |
849 | kfree(nf_nat_l4protos[i]); | |
dd13b010 | 850 | synchronize_net(); |
5b1158e9 JK |
851 | } |
852 | ||
853 | MODULE_LICENSE("GPL"); | |
854 | ||
855 | module_init(nf_nat_init); | |
856 | module_exit(nf_nat_cleanup); |