]>
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> | |
41d73ec0 | 28 | #include <net/netfilter/nf_conntrack_seqadj.h> |
5b1158e9 | 29 | #include <net/netfilter/nf_conntrack_l3proto.h> |
5d0aa2cc | 30 | #include <net/netfilter/nf_conntrack_zones.h> |
c7232c99 | 31 | #include <linux/netfilter/nf_nat.h> |
5b1158e9 | 32 | |
fa3e0709 FW |
33 | static DEFINE_SPINLOCK(nf_nat_lock); |
34 | ||
c7232c99 PM |
35 | static DEFINE_MUTEX(nf_nat_proto_mutex); |
36 | static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO] | |
37 | __read_mostly; | |
38 | static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO] | |
ce4b1ceb | 39 | __read_mostly; |
a76ae1c8 | 40 | |
fa3e0709 FW |
41 | static struct hlist_head *nf_nat_bysource __read_mostly; |
42 | static unsigned int nf_nat_htable_size __read_mostly; | |
43 | static unsigned int nf_nat_hash_rnd __read_mostly; | |
c7232c99 PM |
44 | |
45 | inline const struct nf_nat_l3proto * | |
46 | __nf_nat_l3proto_find(u8 family) | |
5b1158e9 | 47 | { |
c7232c99 | 48 | return rcu_dereference(nf_nat_l3protos[family]); |
5b1158e9 JK |
49 | } |
50 | ||
c7232c99 PM |
51 | inline const struct nf_nat_l4proto * |
52 | __nf_nat_l4proto_find(u8 family, u8 protonum) | |
53 | { | |
54 | return rcu_dereference(nf_nat_l4protos[family][protonum]); | |
55 | } | |
56 | EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find); | |
57 | ||
58 | #ifdef CONFIG_XFRM | |
59 | static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl) | |
60 | { | |
61 | const struct nf_nat_l3proto *l3proto; | |
62 | const struct nf_conn *ct; | |
63 | enum ip_conntrack_info ctinfo; | |
64 | enum ip_conntrack_dir dir; | |
65 | unsigned long statusbit; | |
66 | u8 family; | |
67 | ||
68 | ct = nf_ct_get(skb, &ctinfo); | |
69 | if (ct == NULL) | |
70 | return; | |
71 | ||
53890234 | 72 | family = nf_ct_l3num(ct); |
c7232c99 PM |
73 | l3proto = __nf_nat_l3proto_find(family); |
74 | if (l3proto == NULL) | |
53890234 | 75 | return; |
c7232c99 PM |
76 | |
77 | dir = CTINFO2DIR(ctinfo); | |
78 | if (dir == IP_CT_DIR_ORIGINAL) | |
79 | statusbit = IPS_DST_NAT; | |
80 | else | |
81 | statusbit = IPS_SRC_NAT; | |
82 | ||
83 | l3proto->decode_session(skb, ct, dir, statusbit, fl); | |
c7232c99 PM |
84 | } |
85 | ||
c7af6483 | 86 | int nf_xfrm_me_harder(struct net *net, struct sk_buff *skb, unsigned int family) |
c7232c99 PM |
87 | { |
88 | struct flowi fl; | |
89 | unsigned int hh_len; | |
90 | struct dst_entry *dst; | |
aaa795ad | 91 | int err; |
c7232c99 | 92 | |
aaa795ad | 93 | err = xfrm_decode_session(skb, &fl, family); |
e7e6f630 | 94 | if (err < 0) |
aaa795ad | 95 | return err; |
c7232c99 PM |
96 | |
97 | dst = skb_dst(skb); | |
98 | if (dst->xfrm) | |
99 | dst = ((struct xfrm_dst *)dst)->route; | |
100 | dst_hold(dst); | |
101 | ||
c7af6483 | 102 | dst = xfrm_lookup(net, dst, &fl, skb->sk, 0); |
c7232c99 | 103 | if (IS_ERR(dst)) |
aaa795ad | 104 | return PTR_ERR(dst); |
c7232c99 PM |
105 | |
106 | skb_dst_drop(skb); | |
107 | skb_dst_set(skb, dst); | |
108 | ||
109 | /* Change in oif may mean change in hh_len. */ | |
110 | hh_len = skb_dst(skb)->dev->hard_header_len; | |
111 | if (skb_headroom(skb) < hh_len && | |
112 | pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC)) | |
aaa795ad | 113 | return -ENOMEM; |
c7232c99 PM |
114 | return 0; |
115 | } | |
116 | EXPORT_SYMBOL(nf_xfrm_me_harder); | |
117 | #endif /* CONFIG_XFRM */ | |
118 | ||
fa3e0709 FW |
119 | /* We keep an extra hash for each conntrack, for fast searching. */ |
120 | static unsigned int | |
121 | hash_by_src(const struct net *n, const struct nf_conntrack_tuple *tuple) | |
5b1158e9 | 122 | { |
fa3e0709 FW |
123 | unsigned int hash; |
124 | ||
125 | get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd)); | |
7001c6d1 | 126 | |
5b1158e9 | 127 | /* Original src, to ensure we map it consistently if poss. */ |
fa3e0709 FW |
128 | hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32), |
129 | tuple->dst.protonum ^ nf_nat_hash_rnd ^ net_hash_mix(n)); | |
8fc54f68 | 130 | |
fa3e0709 | 131 | return reciprocal_scale(hash, nf_nat_htable_size); |
5b1158e9 JK |
132 | } |
133 | ||
5b1158e9 JK |
134 | /* Is this tuple already taken? (not by us) */ |
135 | int | |
136 | nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple, | |
137 | const struct nf_conn *ignored_conntrack) | |
138 | { | |
139 | /* Conntrack tracking doesn't keep track of outgoing tuples; only | |
c7232c99 PM |
140 | * incoming ones. NAT means they don't have a fixed mapping, |
141 | * so we invert the tuple and look for the incoming reply. | |
142 | * | |
143 | * We could keep a separate hash if this proves too slow. | |
144 | */ | |
5b1158e9 JK |
145 | struct nf_conntrack_tuple reply; |
146 | ||
147 | nf_ct_invert_tuplepr(&reply, tuple); | |
148 | return nf_conntrack_tuple_taken(&reply, ignored_conntrack); | |
149 | } | |
150 | EXPORT_SYMBOL(nf_nat_used_tuple); | |
151 | ||
152 | /* If we source map this tuple so reply looks like reply_tuple, will | |
c7232c99 PM |
153 | * that meet the constraints of range. |
154 | */ | |
155 | static int in_range(const struct nf_nat_l3proto *l3proto, | |
156 | const struct nf_nat_l4proto *l4proto, | |
157 | const struct nf_conntrack_tuple *tuple, | |
158 | const struct nf_nat_range *range) | |
5b1158e9 | 159 | { |
5b1158e9 | 160 | /* If we are supposed to map IPs, then we must be in the |
c7232c99 PM |
161 | * range specified, otherwise let this drag us onto a new src IP. |
162 | */ | |
163 | if (range->flags & NF_NAT_RANGE_MAP_IPS && | |
164 | !l3proto->in_range(tuple, range)) | |
165 | return 0; | |
5b1158e9 | 166 | |
cbc9f2f4 | 167 | if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) || |
c7232c99 PM |
168 | l4proto->in_range(tuple, NF_NAT_MANIP_SRC, |
169 | &range->min_proto, &range->max_proto)) | |
170 | return 1; | |
5b1158e9 | 171 | |
c7232c99 | 172 | return 0; |
5b1158e9 JK |
173 | } |
174 | ||
175 | static inline int | |
176 | same_src(const struct nf_conn *ct, | |
177 | const struct nf_conntrack_tuple *tuple) | |
178 | { | |
179 | const struct nf_conntrack_tuple *t; | |
180 | ||
181 | t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple; | |
182 | return (t->dst.protonum == tuple->dst.protonum && | |
c7232c99 | 183 | nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) && |
5b1158e9 JK |
184 | t->src.u.all == tuple->src.u.all); |
185 | } | |
186 | ||
187 | /* Only called for SRC manip */ | |
188 | static int | |
308ac914 DB |
189 | find_appropriate_src(struct net *net, |
190 | const struct nf_conntrack_zone *zone, | |
c7232c99 PM |
191 | const struct nf_nat_l3proto *l3proto, |
192 | const struct nf_nat_l4proto *l4proto, | |
0c4c9288 | 193 | const struct nf_conntrack_tuple *tuple, |
5b1158e9 | 194 | struct nf_conntrack_tuple *result, |
c7232c99 | 195 | const struct nf_nat_range *range) |
5b1158e9 | 196 | { |
fa3e0709 | 197 | unsigned int h = hash_by_src(net, tuple); |
72b72949 | 198 | const struct nf_conn *ct; |
870190a9 | 199 | |
fa3e0709 FW |
200 | hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) { |
201 | if (same_src(ct, tuple) && | |
202 | net_eq(net, nf_ct_net(ct)) && | |
203 | nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) { | |
204 | /* Copy source part from reply tuple. */ | |
205 | nf_ct_invert_tuplepr(result, | |
206 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
207 | result->dst = tuple->dst; | |
208 | ||
209 | if (in_range(l3proto, l4proto, result, range)) | |
210 | return 1; | |
211 | } | |
97772bcd | 212 | } |
97772bcd | 213 | return 0; |
5b1158e9 JK |
214 | } |
215 | ||
216 | /* For [FUTURE] fragmentation handling, we want the least-used | |
c7232c99 PM |
217 | * src-ip/dst-ip/proto triple. Fairness doesn't come into it. Thus |
218 | * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports | |
219 | * 1-65535, we don't do pro-rata allocation based on ports; we choose | |
220 | * the ip with the lowest src-ip/dst-ip/proto usage. | |
221 | */ | |
5b1158e9 | 222 | static void |
308ac914 DB |
223 | find_best_ips_proto(const struct nf_conntrack_zone *zone, |
224 | struct nf_conntrack_tuple *tuple, | |
c7232c99 | 225 | const struct nf_nat_range *range, |
5b1158e9 JK |
226 | const struct nf_conn *ct, |
227 | enum nf_nat_manip_type maniptype) | |
228 | { | |
c7232c99 PM |
229 | union nf_inet_addr *var_ipp; |
230 | unsigned int i, max; | |
5b1158e9 | 231 | /* Host order */ |
c7232c99 PM |
232 | u32 minip, maxip, j, dist; |
233 | bool full_range; | |
5b1158e9 JK |
234 | |
235 | /* No IP mapping? Do nothing. */ | |
cbc9f2f4 | 236 | if (!(range->flags & NF_NAT_RANGE_MAP_IPS)) |
5b1158e9 JK |
237 | return; |
238 | ||
cbc9f2f4 | 239 | if (maniptype == NF_NAT_MANIP_SRC) |
c7232c99 | 240 | var_ipp = &tuple->src.u3; |
5b1158e9 | 241 | else |
c7232c99 | 242 | var_ipp = &tuple->dst.u3; |
5b1158e9 JK |
243 | |
244 | /* Fast path: only one choice. */ | |
c7232c99 PM |
245 | if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) { |
246 | *var_ipp = range->min_addr; | |
5b1158e9 JK |
247 | return; |
248 | } | |
249 | ||
c7232c99 PM |
250 | if (nf_ct_l3num(ct) == NFPROTO_IPV4) |
251 | max = sizeof(var_ipp->ip) / sizeof(u32) - 1; | |
252 | else | |
253 | max = sizeof(var_ipp->ip6) / sizeof(u32) - 1; | |
254 | ||
5b1158e9 JK |
255 | /* Hashing source and destination IPs gives a fairly even |
256 | * spread in practice (if there are a small number of IPs | |
257 | * involved, there usually aren't that many connections | |
258 | * anyway). The consistency means that servers see the same | |
259 | * client coming from the same IP (some Internet Banking sites | |
c7232c99 PM |
260 | * like this), even across reboots. |
261 | */ | |
5693d68d | 262 | j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32), |
c7232c99 | 263 | range->flags & NF_NAT_RANGE_PERSISTENT ? |
308ac914 | 264 | 0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id); |
c7232c99 PM |
265 | |
266 | full_range = false; | |
267 | for (i = 0; i <= max; i++) { | |
268 | /* If first bytes of the address are at the maximum, use the | |
269 | * distance. Otherwise use the full range. | |
270 | */ | |
271 | if (!full_range) { | |
272 | minip = ntohl((__force __be32)range->min_addr.all[i]); | |
273 | maxip = ntohl((__force __be32)range->max_addr.all[i]); | |
274 | dist = maxip - minip + 1; | |
275 | } else { | |
276 | minip = 0; | |
277 | dist = ~0; | |
278 | } | |
279 | ||
280 | var_ipp->all[i] = (__force __u32) | |
8fc54f68 | 281 | htonl(minip + reciprocal_scale(j, dist)); |
c7232c99 PM |
282 | if (var_ipp->all[i] != range->max_addr.all[i]) |
283 | full_range = true; | |
284 | ||
285 | if (!(range->flags & NF_NAT_RANGE_PERSISTENT)) | |
286 | j ^= (__force u32)tuple->dst.u3.all[i]; | |
287 | } | |
5b1158e9 JK |
288 | } |
289 | ||
c7232c99 PM |
290 | /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING, |
291 | * we change the source to map into the range. For NF_INET_PRE_ROUTING | |
6e23ae2a | 292 | * and NF_INET_LOCAL_OUT, we change the destination to map into the |
c7232c99 | 293 | * range. It might not be possible to get a unique tuple, but we try. |
5b1158e9 JK |
294 | * At worst (or if we race), we will end up with a final duplicate in |
295 | * __ip_conntrack_confirm and drop the packet. */ | |
296 | static void | |
297 | get_unique_tuple(struct nf_conntrack_tuple *tuple, | |
298 | const struct nf_conntrack_tuple *orig_tuple, | |
c7232c99 | 299 | const struct nf_nat_range *range, |
5b1158e9 JK |
300 | struct nf_conn *ct, |
301 | enum nf_nat_manip_type maniptype) | |
302 | { | |
308ac914 | 303 | const struct nf_conntrack_zone *zone; |
c7232c99 PM |
304 | const struct nf_nat_l3proto *l3proto; |
305 | const struct nf_nat_l4proto *l4proto; | |
0c4c9288 | 306 | struct net *net = nf_ct_net(ct); |
308ac914 DB |
307 | |
308 | zone = nf_ct_zone(ct); | |
5b1158e9 | 309 | |
c7232c99 PM |
310 | rcu_read_lock(); |
311 | l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num); | |
312 | l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num, | |
313 | orig_tuple->dst.protonum); | |
5b1158e9 | 314 | |
c7232c99 PM |
315 | /* 1) If this srcip/proto/src-proto-part is currently mapped, |
316 | * and that same mapping gives a unique tuple within the given | |
317 | * range, use that. | |
318 | * | |
319 | * This is only required for source (ie. NAT/masq) mappings. | |
320 | * So far, we don't do local source mappings, so multiple | |
321 | * manips not an issue. | |
322 | */ | |
cbc9f2f4 | 323 | if (maniptype == NF_NAT_MANIP_SRC && |
34ce3240 | 324 | !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) { |
41a7cab6 | 325 | /* try the original tuple first */ |
c7232c99 | 326 | if (in_range(l3proto, l4proto, orig_tuple, range)) { |
41a7cab6 CG |
327 | if (!nf_nat_used_tuple(orig_tuple, ct)) { |
328 | *tuple = *orig_tuple; | |
c7232c99 | 329 | goto out; |
41a7cab6 | 330 | } |
c7232c99 PM |
331 | } else if (find_appropriate_src(net, zone, l3proto, l4proto, |
332 | orig_tuple, tuple, range)) { | |
0d53778e | 333 | pr_debug("get_unique_tuple: Found current src map\n"); |
0dbff689 | 334 | if (!nf_nat_used_tuple(tuple, ct)) |
c7232c99 | 335 | goto out; |
5b1158e9 JK |
336 | } |
337 | } | |
338 | ||
c7232c99 | 339 | /* 2) Select the least-used IP/proto combination in the given range */ |
5b1158e9 | 340 | *tuple = *orig_tuple; |
5d0aa2cc | 341 | find_best_ips_proto(zone, tuple, range, ct, maniptype); |
5b1158e9 JK |
342 | |
343 | /* 3) The per-protocol part of the manip is made to map into | |
c7232c99 PM |
344 | * the range to make a unique tuple. |
345 | */ | |
5b1158e9 JK |
346 | |
347 | /* Only bother mapping if it's not already in range and unique */ | |
34ce3240 | 348 | if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) { |
cbc9f2f4 | 349 | if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) { |
c7232c99 PM |
350 | if (l4proto->in_range(tuple, maniptype, |
351 | &range->min_proto, | |
352 | &range->max_proto) && | |
353 | (range->min_proto.all == range->max_proto.all || | |
99ad3c53 CG |
354 | !nf_nat_used_tuple(tuple, ct))) |
355 | goto out; | |
356 | } else if (!nf_nat_used_tuple(tuple, ct)) { | |
357 | goto out; | |
358 | } | |
359 | } | |
5b1158e9 JK |
360 | |
361 | /* Last change: get protocol to try to obtain unique tuple. */ | |
c7232c99 | 362 | l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct); |
e22a0548 PM |
363 | out: |
364 | rcu_read_unlock(); | |
5b1158e9 JK |
365 | } |
366 | ||
f768e5bd FW |
367 | struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct) |
368 | { | |
369 | struct nf_conn_nat *nat = nfct_nat(ct); | |
370 | if (nat) | |
371 | return nat; | |
372 | ||
373 | if (!nf_ct_is_confirmed(ct)) | |
374 | nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC); | |
375 | ||
376 | return nat; | |
377 | } | |
378 | EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add); | |
379 | ||
5b1158e9 JK |
380 | unsigned int |
381 | nf_nat_setup_info(struct nf_conn *ct, | |
c7232c99 | 382 | const struct nf_nat_range *range, |
cc01dcbd | 383 | enum nf_nat_manip_type maniptype) |
5b1158e9 | 384 | { |
fa3e0709 | 385 | struct net *net = nf_ct_net(ct); |
5b1158e9 | 386 | struct nf_conntrack_tuple curr_tuple, new_tuple; |
2d59e5ca | 387 | |
d110a394 LZ |
388 | /* Can't setup nat info for confirmed ct. */ |
389 | if (nf_ct_is_confirmed(ct)) | |
390 | return NF_ACCEPT; | |
391 | ||
cbc9f2f4 PM |
392 | NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC || |
393 | maniptype == NF_NAT_MANIP_DST); | |
5b1158e9 JK |
394 | BUG_ON(nf_nat_initialized(ct, maniptype)); |
395 | ||
396 | /* What we've got will look like inverse of reply. Normally | |
c7232c99 PM |
397 | * this is what is in the conntrack, except for prior |
398 | * manipulations (future optimization: if num_manips == 0, | |
399 | * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple) | |
400 | */ | |
5b1158e9 JK |
401 | nf_ct_invert_tuplepr(&curr_tuple, |
402 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
403 | ||
404 | get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype); | |
405 | ||
406 | if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) { | |
407 | struct nf_conntrack_tuple reply; | |
408 | ||
409 | /* Alter conntrack table so will recognize replies. */ | |
410 | nf_ct_invert_tuplepr(&reply, &new_tuple); | |
411 | nf_conntrack_alter_reply(ct, &reply); | |
412 | ||
413 | /* Non-atomic: we own this at the moment. */ | |
cbc9f2f4 | 414 | if (maniptype == NF_NAT_MANIP_SRC) |
5b1158e9 JK |
415 | ct->status |= IPS_SRC_NAT; |
416 | else | |
417 | ct->status |= IPS_DST_NAT; | |
41d73ec0 | 418 | |
ab6dd1be | 419 | if (nfct_help(ct) && !nfct_seqadj(ct)) |
4440a2ab GF |
420 | if (!nfct_seqadj_ext_add(ct)) |
421 | return NF_DROP; | |
5b1158e9 JK |
422 | } |
423 | ||
cbc9f2f4 | 424 | if (maniptype == NF_NAT_MANIP_SRC) { |
fa3e0709 FW |
425 | unsigned int srchash; |
426 | ||
427 | srchash = hash_by_src(net, | |
428 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
429 | spin_lock_bh(&nf_nat_lock); | |
430 | hlist_add_head_rcu(&ct->nat_bysource, | |
431 | &nf_nat_bysource[srchash]); | |
432 | spin_unlock_bh(&nf_nat_lock); | |
5b1158e9 JK |
433 | } |
434 | ||
435 | /* It's done. */ | |
cbc9f2f4 | 436 | if (maniptype == NF_NAT_MANIP_DST) |
a7c2f4d7 | 437 | ct->status |= IPS_DST_NAT_DONE; |
5b1158e9 | 438 | else |
a7c2f4d7 | 439 | ct->status |= IPS_SRC_NAT_DONE; |
5b1158e9 JK |
440 | |
441 | return NF_ACCEPT; | |
442 | } | |
443 | EXPORT_SYMBOL(nf_nat_setup_info); | |
444 | ||
0eba801b PNA |
445 | static unsigned int |
446 | __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip) | |
f59cb045 PNA |
447 | { |
448 | /* Force range to this IP; let proto decide mapping for | |
449 | * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED). | |
450 | * Use reply in case it's already been mangled (eg local packet). | |
451 | */ | |
452 | union nf_inet_addr ip = | |
0eba801b | 453 | (manip == NF_NAT_MANIP_SRC ? |
f59cb045 PNA |
454 | ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 : |
455 | ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3); | |
456 | struct nf_nat_range range = { | |
457 | .flags = NF_NAT_RANGE_MAP_IPS, | |
458 | .min_addr = ip, | |
459 | .max_addr = ip, | |
460 | }; | |
0eba801b PNA |
461 | return nf_nat_setup_info(ct, &range, manip); |
462 | } | |
463 | ||
464 | unsigned int | |
465 | nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum) | |
466 | { | |
467 | return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum)); | |
f59cb045 PNA |
468 | } |
469 | EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding); | |
470 | ||
5b1158e9 JK |
471 | /* Do packet manipulations according to nf_nat_setup_info. */ |
472 | unsigned int nf_nat_packet(struct nf_conn *ct, | |
473 | enum ip_conntrack_info ctinfo, | |
474 | unsigned int hooknum, | |
3db05fea | 475 | struct sk_buff *skb) |
5b1158e9 | 476 | { |
c7232c99 PM |
477 | const struct nf_nat_l3proto *l3proto; |
478 | const struct nf_nat_l4proto *l4proto; | |
5b1158e9 JK |
479 | enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo); |
480 | unsigned long statusbit; | |
481 | enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum); | |
482 | ||
cbc9f2f4 | 483 | if (mtype == NF_NAT_MANIP_SRC) |
5b1158e9 JK |
484 | statusbit = IPS_SRC_NAT; |
485 | else | |
486 | statusbit = IPS_DST_NAT; | |
487 | ||
488 | /* Invert if this is reply dir. */ | |
489 | if (dir == IP_CT_DIR_REPLY) | |
490 | statusbit ^= IPS_NAT_MASK; | |
491 | ||
492 | /* Non-atomic: these bits don't change. */ | |
493 | if (ct->status & statusbit) { | |
494 | struct nf_conntrack_tuple target; | |
495 | ||
496 | /* We are aiming to look like inverse of other direction. */ | |
497 | nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple); | |
498 | ||
c7232c99 PM |
499 | l3proto = __nf_nat_l3proto_find(target.src.l3num); |
500 | l4proto = __nf_nat_l4proto_find(target.src.l3num, | |
501 | target.dst.protonum); | |
502 | if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype)) | |
5b1158e9 JK |
503 | return NF_DROP; |
504 | } | |
505 | return NF_ACCEPT; | |
506 | } | |
507 | EXPORT_SYMBOL_GPL(nf_nat_packet); | |
508 | ||
c7232c99 PM |
509 | struct nf_nat_proto_clean { |
510 | u8 l3proto; | |
511 | u8 l4proto; | |
c7232c99 PM |
512 | }; |
513 | ||
c2d421e1 FW |
514 | /* kill conntracks with affected NAT section */ |
515 | static int nf_nat_proto_remove(struct nf_conn *i, void *data) | |
5b1158e9 | 516 | { |
c7232c99 | 517 | const struct nf_nat_proto_clean *clean = data; |
c2d421e1 | 518 | |
c7232c99 PM |
519 | if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) || |
520 | (clean->l4proto && nf_ct_protonum(i) != clean->l4proto)) | |
5b1158e9 JK |
521 | return 0; |
522 | ||
c2d421e1 | 523 | return i->status & IPS_NAT_MASK ? 1 : 0; |
c7232c99 | 524 | } |
5b1158e9 | 525 | |
945b2b2d FW |
526 | static int nf_nat_proto_clean(struct nf_conn *ct, void *data) |
527 | { | |
945b2b2d FW |
528 | if (nf_nat_proto_remove(ct, data)) |
529 | return 1; | |
530 | ||
6e699867 | 531 | if ((ct->status & IPS_SRC_NAT_DONE) == 0) |
945b2b2d FW |
532 | return 0; |
533 | ||
534 | /* This netns is being destroyed, and conntrack has nat null binding. | |
535 | * Remove it from bysource hash, as the table will be freed soon. | |
536 | * | |
537 | * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack() | |
538 | * will delete entry from already-freed table. | |
539 | */ | |
124dffea | 540 | clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status); |
fa3e0709 FW |
541 | spin_lock_bh(&nf_nat_lock); |
542 | hlist_del_rcu(&ct->nat_bysource); | |
543 | spin_unlock_bh(&nf_nat_lock); | |
945b2b2d | 544 | |
945b2b2d FW |
545 | /* don't delete conntrack. Although that would make things a lot |
546 | * simpler, we'd end up flushing all conntracks on nat rmmod. | |
547 | */ | |
548 | return 0; | |
549 | } | |
550 | ||
c7232c99 PM |
551 | static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto) |
552 | { | |
553 | struct nf_nat_proto_clean clean = { | |
554 | .l3proto = l3proto, | |
555 | .l4proto = l4proto, | |
556 | }; | |
c7232c99 | 557 | |
8f23f35f | 558 | nf_ct_iterate_destroy(nf_nat_proto_remove, &clean); |
c7232c99 | 559 | } |
5b1158e9 | 560 | |
c7232c99 PM |
561 | static void nf_nat_l3proto_clean(u8 l3proto) |
562 | { | |
563 | struct nf_nat_proto_clean clean = { | |
564 | .l3proto = l3proto, | |
565 | }; | |
c7232c99 | 566 | |
8f23f35f | 567 | nf_ct_iterate_destroy(nf_nat_proto_remove, &clean); |
5b1158e9 | 568 | } |
5b1158e9 JK |
569 | |
570 | /* Protocol registration. */ | |
c7232c99 | 571 | int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto) |
5b1158e9 | 572 | { |
c7232c99 PM |
573 | const struct nf_nat_l4proto **l4protos; |
574 | unsigned int i; | |
5b1158e9 JK |
575 | int ret = 0; |
576 | ||
c7232c99 PM |
577 | mutex_lock(&nf_nat_proto_mutex); |
578 | if (nf_nat_l4protos[l3proto] == NULL) { | |
579 | l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *), | |
580 | GFP_KERNEL); | |
581 | if (l4protos == NULL) { | |
582 | ret = -ENOMEM; | |
583 | goto out; | |
584 | } | |
585 | ||
586 | for (i = 0; i < IPPROTO_MAX; i++) | |
587 | RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown); | |
588 | ||
589 | /* Before making proto_array visible to lockless readers, | |
590 | * we must make sure its content is committed to memory. | |
591 | */ | |
592 | smp_wmb(); | |
593 | ||
594 | nf_nat_l4protos[l3proto] = l4protos; | |
595 | } | |
596 | ||
eb733162 | 597 | if (rcu_dereference_protected( |
c7232c99 PM |
598 | nf_nat_l4protos[l3proto][l4proto->l4proto], |
599 | lockdep_is_held(&nf_nat_proto_mutex) | |
600 | ) != &nf_nat_l4proto_unknown) { | |
5b1158e9 JK |
601 | ret = -EBUSY; |
602 | goto out; | |
603 | } | |
c7232c99 | 604 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto); |
5b1158e9 | 605 | out: |
c7232c99 | 606 | mutex_unlock(&nf_nat_proto_mutex); |
5b1158e9 JK |
607 | return ret; |
608 | } | |
c7232c99 | 609 | EXPORT_SYMBOL_GPL(nf_nat_l4proto_register); |
5b1158e9 | 610 | |
25985edc | 611 | /* No one stores the protocol anywhere; simply delete it. */ |
c7232c99 | 612 | void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto) |
5b1158e9 | 613 | { |
c7232c99 PM |
614 | mutex_lock(&nf_nat_proto_mutex); |
615 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], | |
616 | &nf_nat_l4proto_unknown); | |
617 | mutex_unlock(&nf_nat_proto_mutex); | |
e22a0548 | 618 | synchronize_rcu(); |
c7232c99 PM |
619 | |
620 | nf_nat_l4proto_clean(l3proto, l4proto->l4proto); | |
5b1158e9 | 621 | } |
c7232c99 PM |
622 | EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister); |
623 | ||
624 | int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto) | |
625 | { | |
626 | int err; | |
627 | ||
628 | err = nf_ct_l3proto_try_module_get(l3proto->l3proto); | |
629 | if (err < 0) | |
630 | return err; | |
631 | ||
632 | mutex_lock(&nf_nat_proto_mutex); | |
633 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP], | |
634 | &nf_nat_l4proto_tcp); | |
635 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP], | |
636 | &nf_nat_l4proto_udp); | |
0c4e966e DC |
637 | #ifdef CONFIG_NF_NAT_PROTO_DCCP |
638 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP], | |
639 | &nf_nat_l4proto_dccp); | |
7a2dd28c DC |
640 | #endif |
641 | #ifdef CONFIG_NF_NAT_PROTO_SCTP | |
642 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP], | |
643 | &nf_nat_l4proto_sctp); | |
b8ad652f DC |
644 | #endif |
645 | #ifdef CONFIG_NF_NAT_PROTO_UDPLITE | |
646 | RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE], | |
647 | &nf_nat_l4proto_udplite); | |
0c4e966e | 648 | #endif |
c7232c99 PM |
649 | mutex_unlock(&nf_nat_proto_mutex); |
650 | ||
651 | RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto); | |
652 | return 0; | |
653 | } | |
654 | EXPORT_SYMBOL_GPL(nf_nat_l3proto_register); | |
655 | ||
656 | void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto) | |
657 | { | |
658 | mutex_lock(&nf_nat_proto_mutex); | |
659 | RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL); | |
660 | mutex_unlock(&nf_nat_proto_mutex); | |
661 | synchronize_rcu(); | |
662 | ||
663 | nf_nat_l3proto_clean(l3proto->l3proto); | |
664 | nf_ct_l3proto_module_put(l3proto->l3proto); | |
665 | } | |
666 | EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister); | |
5b1158e9 | 667 | |
25985edc | 668 | /* No one using conntrack by the time this called. */ |
d8a0509a YK |
669 | static void nf_nat_cleanup_conntrack(struct nf_conn *ct) |
670 | { | |
fa3e0709 FW |
671 | if (ct->status & IPS_SRC_NAT_DONE) { |
672 | spin_lock_bh(&nf_nat_lock); | |
673 | hlist_del_rcu(&ct->nat_bysource); | |
674 | spin_unlock_bh(&nf_nat_lock); | |
675 | } | |
2d59e5ca YK |
676 | } |
677 | ||
61eb3107 | 678 | static struct nf_ct_ext_type nat_extend __read_mostly = { |
d8a0509a YK |
679 | .len = sizeof(struct nf_conn_nat), |
680 | .align = __alignof__(struct nf_conn_nat), | |
681 | .destroy = nf_nat_cleanup_conntrack, | |
d8a0509a | 682 | .id = NF_CT_EXT_NAT, |
2d59e5ca YK |
683 | }; |
684 | ||
24de3d37 | 685 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
e6a7d3c0 PNA |
686 | |
687 | #include <linux/netfilter/nfnetlink.h> | |
688 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
689 | ||
690 | static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = { | |
691 | [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 }, | |
692 | [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 }, | |
693 | }; | |
694 | ||
695 | static int nfnetlink_parse_nat_proto(struct nlattr *attr, | |
696 | const struct nf_conn *ct, | |
c7232c99 | 697 | struct nf_nat_range *range) |
e6a7d3c0 PNA |
698 | { |
699 | struct nlattr *tb[CTA_PROTONAT_MAX+1]; | |
c7232c99 | 700 | const struct nf_nat_l4proto *l4proto; |
e6a7d3c0 PNA |
701 | int err; |
702 | ||
fceb6435 JB |
703 | err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, |
704 | protonat_nla_policy, NULL); | |
e6a7d3c0 PNA |
705 | if (err < 0) |
706 | return err; | |
707 | ||
c7232c99 PM |
708 | l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
709 | if (l4proto->nlattr_to_range) | |
710 | err = l4proto->nlattr_to_range(tb, range); | |
711 | ||
e6a7d3c0 PNA |
712 | return err; |
713 | } | |
714 | ||
715 | static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = { | |
c7232c99 PM |
716 | [CTA_NAT_V4_MINIP] = { .type = NLA_U32 }, |
717 | [CTA_NAT_V4_MAXIP] = { .type = NLA_U32 }, | |
58a317f1 PM |
718 | [CTA_NAT_V6_MINIP] = { .len = sizeof(struct in6_addr) }, |
719 | [CTA_NAT_V6_MAXIP] = { .len = sizeof(struct in6_addr) }, | |
329fb58a | 720 | [CTA_NAT_PROTO] = { .type = NLA_NESTED }, |
e6a7d3c0 PNA |
721 | }; |
722 | ||
723 | static int | |
39938324 | 724 | nfnetlink_parse_nat(const struct nlattr *nat, |
0eba801b PNA |
725 | const struct nf_conn *ct, struct nf_nat_range *range, |
726 | const struct nf_nat_l3proto *l3proto) | |
e6a7d3c0 PNA |
727 | { |
728 | struct nlattr *tb[CTA_NAT_MAX+1]; | |
729 | int err; | |
730 | ||
731 | memset(range, 0, sizeof(*range)); | |
732 | ||
fceb6435 | 733 | err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL); |
e6a7d3c0 PNA |
734 | if (err < 0) |
735 | return err; | |
736 | ||
c7232c99 PM |
737 | err = l3proto->nlattr_to_range(tb, range); |
738 | if (err < 0) | |
0eba801b | 739 | return err; |
e6a7d3c0 PNA |
740 | |
741 | if (!tb[CTA_NAT_PROTO]) | |
0eba801b | 742 | return 0; |
e6a7d3c0 | 743 | |
0eba801b | 744 | return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range); |
e6a7d3c0 PNA |
745 | } |
746 | ||
0eba801b | 747 | /* This function is called under rcu_read_lock() */ |
e6a7d3c0 PNA |
748 | static int |
749 | nfnetlink_parse_nat_setup(struct nf_conn *ct, | |
750 | enum nf_nat_manip_type manip, | |
39938324 | 751 | const struct nlattr *attr) |
e6a7d3c0 | 752 | { |
c7232c99 | 753 | struct nf_nat_range range; |
0eba801b | 754 | const struct nf_nat_l3proto *l3proto; |
c7232c99 | 755 | int err; |
e6a7d3c0 | 756 | |
0eba801b PNA |
757 | /* Should not happen, restricted to creating new conntracks |
758 | * via ctnetlink. | |
759 | */ | |
760 | if (WARN_ON_ONCE(nf_nat_initialized(ct, manip))) | |
761 | return -EEXIST; | |
762 | ||
763 | /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to | |
764 | * attach the null binding, otherwise this may oops. | |
765 | */ | |
766 | l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct)); | |
767 | if (l3proto == NULL) | |
768 | return -EAGAIN; | |
769 | ||
770 | /* No NAT information has been passed, allocate the null-binding */ | |
771 | if (attr == NULL) | |
7025bac4 | 772 | return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0; |
0eba801b PNA |
773 | |
774 | err = nfnetlink_parse_nat(attr, ct, &range, l3proto); | |
c7232c99 PM |
775 | if (err < 0) |
776 | return err; | |
e6a7d3c0 | 777 | |
ecfcdfec | 778 | return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0; |
e6a7d3c0 PNA |
779 | } |
780 | #else | |
781 | static int | |
782 | nfnetlink_parse_nat_setup(struct nf_conn *ct, | |
783 | enum nf_nat_manip_type manip, | |
39938324 | 784 | const struct nlattr *attr) |
e6a7d3c0 PNA |
785 | { |
786 | return -EOPNOTSUPP; | |
787 | } | |
788 | #endif | |
789 | ||
544d5c7d PNA |
790 | static struct nf_ct_helper_expectfn follow_master_nat = { |
791 | .name = "nat-follow-master", | |
792 | .expectfn = nf_nat_follow_master, | |
793 | }; | |
794 | ||
5b1158e9 JK |
795 | static int __init nf_nat_init(void) |
796 | { | |
2d59e5ca YK |
797 | int ret; |
798 | ||
fa3e0709 FW |
799 | /* Leave them the same for the moment. */ |
800 | nf_nat_htable_size = nf_conntrack_htable_size; | |
801 | ||
802 | nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0); | |
803 | if (!nf_nat_bysource) | |
804 | return -ENOMEM; | |
a76ae1c8 | 805 | |
2d59e5ca YK |
806 | ret = nf_ct_extend_register(&nat_extend); |
807 | if (ret < 0) { | |
fa3e0709 | 808 | nf_ct_free_hashtable(nf_nat_bysource, nf_nat_htable_size); |
2d59e5ca YK |
809 | printk(KERN_ERR "nf_nat_core: Unable to register extension\n"); |
810 | return ret; | |
811 | } | |
5b1158e9 | 812 | |
c7232c99 | 813 | nf_ct_helper_expectfn_register(&follow_master_nat); |
5b1158e9 | 814 | |
e6a7d3c0 | 815 | BUG_ON(nfnetlink_parse_nat_setup_hook != NULL); |
a9b3cd7f | 816 | RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, |
e6a7d3c0 | 817 | nfnetlink_parse_nat_setup); |
c7232c99 PM |
818 | #ifdef CONFIG_XFRM |
819 | BUG_ON(nf_nat_decode_session_hook != NULL); | |
820 | RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session); | |
821 | #endif | |
5b1158e9 JK |
822 | return 0; |
823 | } | |
824 | ||
5b1158e9 JK |
825 | static void __exit nf_nat_cleanup(void) |
826 | { | |
8f23f35f | 827 | struct nf_nat_proto_clean clean = {}; |
c7232c99 PM |
828 | unsigned int i; |
829 | ||
8f23f35f FW |
830 | nf_ct_iterate_destroy(nf_nat_proto_clean, &clean); |
831 | ||
2d59e5ca | 832 | nf_ct_extend_unregister(&nat_extend); |
544d5c7d | 833 | nf_ct_helper_expectfn_unregister(&follow_master_nat); |
a9b3cd7f | 834 | RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL); |
c7232c99 PM |
835 | #ifdef CONFIG_XFRM |
836 | RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL); | |
837 | #endif | |
3b7dabf0 LZ |
838 | synchronize_rcu(); |
839 | ||
c7232c99 PM |
840 | for (i = 0; i < NFPROTO_NUMPROTO; i++) |
841 | kfree(nf_nat_l4protos[i]); | |
fa3e0709 FW |
842 | synchronize_net(); |
843 | nf_ct_free_hashtable(nf_nat_bysource, nf_nat_htable_size); | |
5b1158e9 JK |
844 | } |
845 | ||
846 | MODULE_LICENSE("GPL"); | |
847 | ||
848 | module_init(nf_nat_init); | |
849 | module_exit(nf_nat_cleanup); |