]>
Commit | Line | Data |
---|---|---|
9fb9cbb1 YK |
1 | /* Connection state tracking for netfilter. This is separated from, |
2 | but required by, the NAT layer; it can also be used by an iptables | |
3 | extension. */ | |
4 | ||
5 | /* (C) 1999-2001 Paul `Rusty' Russell | |
dc808fe2 | 6 | * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org> |
9fb9cbb1 YK |
7 | * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org> |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
9fb9cbb1 YK |
12 | */ |
13 | ||
9fb9cbb1 YK |
14 | #include <linux/types.h> |
15 | #include <linux/netfilter.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/skbuff.h> | |
18 | #include <linux/proc_fs.h> | |
19 | #include <linux/vmalloc.h> | |
20 | #include <linux/stddef.h> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/random.h> | |
23 | #include <linux/jhash.h> | |
24 | #include <linux/err.h> | |
25 | #include <linux/percpu.h> | |
26 | #include <linux/moduleparam.h> | |
27 | #include <linux/notifier.h> | |
28 | #include <linux/kernel.h> | |
29 | #include <linux/netdevice.h> | |
30 | #include <linux/socket.h> | |
d7fe0f24 | 31 | #include <linux/mm.h> |
9fb9cbb1 | 32 | |
9fb9cbb1 YK |
33 | #include <net/netfilter/nf_conntrack.h> |
34 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
605dcad6 | 35 | #include <net/netfilter/nf_conntrack_l4proto.h> |
77ab9cff | 36 | #include <net/netfilter/nf_conntrack_expect.h> |
9fb9cbb1 YK |
37 | #include <net/netfilter/nf_conntrack_helper.h> |
38 | #include <net/netfilter/nf_conntrack_core.h> | |
ecfab2c9 | 39 | #include <net/netfilter/nf_conntrack_extend.h> |
58401572 | 40 | #include <net/netfilter/nf_conntrack_acct.h> |
9fb9cbb1 | 41 | |
dc808fe2 | 42 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 43 | |
f8ba1aff | 44 | DEFINE_SPINLOCK(nf_conntrack_lock); |
13b18339 | 45 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 | 46 | |
e2b7606c | 47 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
48 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
49 | ||
94aec08e | 50 | int nf_conntrack_max __read_mostly; |
a999e683 | 51 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 52 | |
e2b7606c | 53 | struct nf_conn nf_conntrack_untracked __read_mostly; |
13b18339 PM |
54 | EXPORT_SYMBOL_GPL(nf_conntrack_untracked); |
55 | ||
94aec08e | 56 | unsigned int nf_ct_log_invalid __read_mostly; |
dacd2a1a | 57 | static struct kmem_cache *nf_conntrack_cachep __read_mostly; |
77ab9cff | 58 | |
9fb9cbb1 YK |
59 | DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); |
60 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); | |
61 | ||
9fb9cbb1 YK |
62 | static int nf_conntrack_hash_rnd_initted; |
63 | static unsigned int nf_conntrack_hash_rnd; | |
64 | ||
65 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
66 | unsigned int size, unsigned int rnd) | |
67 | { | |
0794935e PM |
68 | unsigned int n; |
69 | u_int32_t h; | |
70 | ||
71 | /* The direction must be ignored, so we hash everything up to the | |
72 | * destination ports (which is a multiple of 4) and treat the last | |
73 | * three bytes manually. | |
74 | */ | |
75 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
76 | h = jhash2((u32 *)tuple, n, | |
77 | rnd ^ (((__force __u16)tuple->dst.u.all << 16) | | |
78 | tuple->dst.protonum)); | |
79 | ||
80 | return ((u64)h * size) >> 32; | |
9fb9cbb1 YK |
81 | } |
82 | ||
83 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
84 | { | |
85 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
86 | nf_conntrack_hash_rnd); | |
87 | } | |
88 | ||
5f2b4c90 | 89 | bool |
9fb9cbb1 YK |
90 | nf_ct_get_tuple(const struct sk_buff *skb, |
91 | unsigned int nhoff, | |
92 | unsigned int dataoff, | |
93 | u_int16_t l3num, | |
94 | u_int8_t protonum, | |
95 | struct nf_conntrack_tuple *tuple, | |
96 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 97 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 98 | { |
443a70d5 | 99 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
100 | |
101 | tuple->src.l3num = l3num; | |
102 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 103 | return false; |
9fb9cbb1 YK |
104 | |
105 | tuple->dst.protonum = protonum; | |
106 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
107 | ||
605dcad6 | 108 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 109 | } |
13b18339 | 110 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 111 | |
5f2b4c90 JE |
112 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
113 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
114 | { |
115 | struct nf_conntrack_l3proto *l3proto; | |
116 | struct nf_conntrack_l4proto *l4proto; | |
117 | unsigned int protoff; | |
118 | u_int8_t protonum; | |
119 | int ret; | |
120 | ||
121 | rcu_read_lock(); | |
122 | ||
123 | l3proto = __nf_ct_l3proto_find(l3num); | |
124 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
125 | if (ret != NF_ACCEPT) { | |
126 | rcu_read_unlock(); | |
5f2b4c90 | 127 | return false; |
e2a3123f YK |
128 | } |
129 | ||
130 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
131 | ||
132 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
133 | l3proto, l4proto); | |
134 | ||
135 | rcu_read_unlock(); | |
136 | return ret; | |
137 | } | |
138 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
139 | ||
5f2b4c90 | 140 | bool |
9fb9cbb1 YK |
141 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
142 | const struct nf_conntrack_tuple *orig, | |
143 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 144 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 145 | { |
443a70d5 | 146 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
147 | |
148 | inverse->src.l3num = orig->src.l3num; | |
149 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 150 | return false; |
9fb9cbb1 YK |
151 | |
152 | inverse->dst.dir = !orig->dst.dir; | |
153 | ||
154 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 155 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 156 | } |
13b18339 | 157 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 158 | |
9fb9cbb1 YK |
159 | static void |
160 | clean_from_lists(struct nf_conn *ct) | |
161 | { | |
0d53778e | 162 | pr_debug("clean_from_lists(%p)\n", ct); |
76507f69 PM |
163 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
164 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); | |
9fb9cbb1 YK |
165 | |
166 | /* Destroy all pending expectations */ | |
c1d10adb | 167 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
168 | } |
169 | ||
170 | static void | |
171 | destroy_conntrack(struct nf_conntrack *nfct) | |
172 | { | |
173 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 174 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 175 | |
0d53778e | 176 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
177 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
178 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
179 | ||
180 | nf_conntrack_event(IPCT_DESTROY, ct); | |
181 | set_bit(IPS_DYING_BIT, &ct->status); | |
182 | ||
183 | /* To make sure we don't get any weird locking issues here: | |
184 | * destroy_conntrack() MUST NOT be called with a write lock | |
185 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 186 | rcu_read_lock(); |
5e8fbe2a | 187 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
188 | if (l4proto && l4proto->destroy) |
189 | l4proto->destroy(ct); | |
9fb9cbb1 | 190 | |
982d9a9c | 191 | rcu_read_unlock(); |
9fb9cbb1 | 192 | |
f8ba1aff | 193 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
194 | /* Expectations will have been removed in clean_from_lists, |
195 | * except TFTP can create an expectation on the first packet, | |
196 | * before connection is in the list, so we need to clean here, | |
197 | * too. */ | |
c1d10adb | 198 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
199 | |
200 | /* We overload first tuple to link into unconfirmed list. */ | |
201 | if (!nf_ct_is_confirmed(ct)) { | |
f205c5e0 PM |
202 | BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode)); |
203 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); | |
9fb9cbb1 YK |
204 | } |
205 | ||
206 | NF_CT_STAT_INC(delete); | |
f8ba1aff | 207 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
208 | |
209 | if (ct->master) | |
210 | nf_ct_put(ct->master); | |
211 | ||
0d53778e | 212 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
213 | nf_conntrack_free(ct); |
214 | } | |
215 | ||
216 | static void death_by_timeout(unsigned long ul_conntrack) | |
217 | { | |
218 | struct nf_conn *ct = (void *)ul_conntrack; | |
5397e97d | 219 | struct nf_conn_help *help = nfct_help(ct); |
3c158f7f | 220 | struct nf_conntrack_helper *helper; |
5397e97d | 221 | |
3c158f7f PM |
222 | if (help) { |
223 | rcu_read_lock(); | |
224 | helper = rcu_dereference(help->helper); | |
225 | if (helper && helper->destroy) | |
226 | helper->destroy(ct); | |
227 | rcu_read_unlock(); | |
228 | } | |
9fb9cbb1 | 229 | |
f8ba1aff | 230 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
231 | /* Inside lock so preempt is disabled on module removal path. |
232 | * Otherwise we can get spurious warnings. */ | |
233 | NF_CT_STAT_INC(delete_list); | |
234 | clean_from_lists(ct); | |
f8ba1aff | 235 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
236 | nf_ct_put(ct); |
237 | } | |
238 | ||
c1d10adb | 239 | struct nf_conntrack_tuple_hash * |
400dad39 | 240 | __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
241 | { |
242 | struct nf_conntrack_tuple_hash *h; | |
f205c5e0 | 243 | struct hlist_node *n; |
9fb9cbb1 YK |
244 | unsigned int hash = hash_conntrack(tuple); |
245 | ||
4e29e9ec PM |
246 | /* Disable BHs the entire time since we normally need to disable them |
247 | * at least once for the stats anyway. | |
248 | */ | |
249 | local_bh_disable(); | |
400dad39 | 250 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) { |
ba419aff | 251 | if (nf_ct_tuple_equal(tuple, &h->tuple)) { |
9fb9cbb1 | 252 | NF_CT_STAT_INC(found); |
4e29e9ec | 253 | local_bh_enable(); |
9fb9cbb1 YK |
254 | return h; |
255 | } | |
256 | NF_CT_STAT_INC(searched); | |
257 | } | |
4e29e9ec | 258 | local_bh_enable(); |
9fb9cbb1 YK |
259 | |
260 | return NULL; | |
261 | } | |
13b18339 | 262 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
263 | |
264 | /* Find a connection corresponding to a tuple. */ | |
265 | struct nf_conntrack_tuple_hash * | |
400dad39 | 266 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
267 | { |
268 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 269 | struct nf_conn *ct; |
9fb9cbb1 | 270 | |
76507f69 | 271 | rcu_read_lock(); |
400dad39 | 272 | h = __nf_conntrack_find(net, tuple); |
76507f69 PM |
273 | if (h) { |
274 | ct = nf_ct_tuplehash_to_ctrack(h); | |
275 | if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
276 | h = NULL; | |
277 | } | |
278 | rcu_read_unlock(); | |
9fb9cbb1 YK |
279 | |
280 | return h; | |
281 | } | |
13b18339 | 282 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 283 | |
c1d10adb PNA |
284 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
285 | unsigned int hash, | |
601e68e1 | 286 | unsigned int repl_hash) |
c1d10adb | 287 | { |
400dad39 AD |
288 | struct net *net = nf_ct_net(ct); |
289 | ||
76507f69 | 290 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
400dad39 | 291 | &net->ct.hash[hash]); |
76507f69 | 292 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, |
400dad39 | 293 | &net->ct.hash[repl_hash]); |
c1d10adb PNA |
294 | } |
295 | ||
296 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
297 | { | |
298 | unsigned int hash, repl_hash; | |
299 | ||
300 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
301 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
302 | ||
f8ba1aff | 303 | spin_lock_bh(&nf_conntrack_lock); |
c1d10adb | 304 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
f8ba1aff | 305 | spin_unlock_bh(&nf_conntrack_lock); |
c1d10adb | 306 | } |
13b18339 | 307 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 308 | |
9fb9cbb1 YK |
309 | /* Confirm a connection given skb; places it in hash table */ |
310 | int | |
3db05fea | 311 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
312 | { |
313 | unsigned int hash, repl_hash; | |
df0933dc | 314 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 315 | struct nf_conn *ct; |
df0933dc | 316 | struct nf_conn_help *help; |
f205c5e0 | 317 | struct hlist_node *n; |
9fb9cbb1 | 318 | enum ip_conntrack_info ctinfo; |
400dad39 | 319 | struct net *net; |
9fb9cbb1 | 320 | |
3db05fea | 321 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 322 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
323 | |
324 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
325 | ICMP/TCP RST packets in other direction. Actual packet | |
326 | which created connection will be IP_CT_NEW or for an | |
327 | expected connection, IP_CT_RELATED. */ | |
328 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
329 | return NF_ACCEPT; | |
330 | ||
331 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
332 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
333 | ||
334 | /* We're not in hash table, and we refuse to set up related | |
335 | connections for unconfirmed conns. But packet copies and | |
336 | REJECT will give spurious warnings here. */ | |
337 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
338 | ||
339 | /* No external references means noone else could have | |
340 | confirmed us. */ | |
341 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 342 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 343 | |
f8ba1aff | 344 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
345 | |
346 | /* See if there's one in the list already, including reverse: | |
347 | NAT could have grabbed it without realizing, since we're | |
348 | not in the hash. If there is, we lost race. */ | |
400dad39 | 349 | hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode) |
df0933dc PM |
350 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
351 | &h->tuple)) | |
352 | goto out; | |
400dad39 | 353 | hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode) |
df0933dc PM |
354 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
355 | &h->tuple)) | |
356 | goto out; | |
9fb9cbb1 | 357 | |
df0933dc | 358 | /* Remove from unconfirmed list */ |
f205c5e0 | 359 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
df0933dc PM |
360 | |
361 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
362 | /* Timer relative to confirmation time, not original | |
363 | setting time, otherwise we'd get timer wrap in | |
364 | weird delay cases. */ | |
365 | ct->timeout.expires += jiffies; | |
366 | add_timer(&ct->timeout); | |
367 | atomic_inc(&ct->ct_general.use); | |
368 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
369 | NF_CT_STAT_INC(insert); | |
f8ba1aff | 370 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
371 | help = nfct_help(ct); |
372 | if (help && help->helper) | |
a71996fc | 373 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
9fb9cbb1 | 374 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
375 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
376 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
a71996fc | 377 | nf_conntrack_event_cache(IPCT_NATINFO, ct); |
9fb9cbb1 | 378 | #endif |
df0933dc | 379 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 380 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 381 | return NF_ACCEPT; |
9fb9cbb1 | 382 | |
df0933dc | 383 | out: |
9fb9cbb1 | 384 | NF_CT_STAT_INC(insert_failed); |
f8ba1aff | 385 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
386 | return NF_DROP; |
387 | } | |
13b18339 | 388 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
389 | |
390 | /* Returns true if a connection correspondings to the tuple (required | |
391 | for NAT). */ | |
392 | int | |
393 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
394 | const struct nf_conn *ignored_conntrack) | |
395 | { | |
400dad39 | 396 | struct net *net = nf_ct_net(ignored_conntrack); |
9fb9cbb1 | 397 | struct nf_conntrack_tuple_hash *h; |
ba419aff PM |
398 | struct hlist_node *n; |
399 | unsigned int hash = hash_conntrack(tuple); | |
9fb9cbb1 | 400 | |
4e29e9ec PM |
401 | /* Disable BHs the entire time since we need to disable them at |
402 | * least once for the stats anyway. | |
403 | */ | |
404 | rcu_read_lock_bh(); | |
400dad39 | 405 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) { |
ba419aff PM |
406 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
407 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
408 | NF_CT_STAT_INC(found); | |
4e29e9ec | 409 | rcu_read_unlock_bh(); |
ba419aff PM |
410 | return 1; |
411 | } | |
412 | NF_CT_STAT_INC(searched); | |
413 | } | |
4e29e9ec | 414 | rcu_read_unlock_bh(); |
9fb9cbb1 | 415 | |
ba419aff | 416 | return 0; |
9fb9cbb1 | 417 | } |
13b18339 | 418 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 419 | |
7ae7730f PM |
420 | #define NF_CT_EVICTION_RANGE 8 |
421 | ||
9fb9cbb1 YK |
422 | /* There's a small race here where we may free a just-assured |
423 | connection. Too bad: we're in trouble anyway. */ | |
400dad39 | 424 | static noinline int early_drop(struct net *net, unsigned int hash) |
9fb9cbb1 | 425 | { |
f205c5e0 | 426 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 427 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 428 | struct nf_conn *ct = NULL, *tmp; |
f205c5e0 | 429 | struct hlist_node *n; |
7ae7730f | 430 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
431 | int dropped = 0; |
432 | ||
76507f69 | 433 | rcu_read_lock(); |
7ae7730f | 434 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
400dad39 | 435 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], |
76507f69 | 436 | hnode) { |
7ae7730f PM |
437 | tmp = nf_ct_tuplehash_to_ctrack(h); |
438 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
439 | ct = tmp; | |
440 | cnt++; | |
441 | } | |
76507f69 PM |
442 | |
443 | if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
444 | ct = NULL; | |
7ae7730f PM |
445 | if (ct || cnt >= NF_CT_EVICTION_RANGE) |
446 | break; | |
447 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 448 | } |
76507f69 | 449 | rcu_read_unlock(); |
9fb9cbb1 YK |
450 | |
451 | if (!ct) | |
452 | return dropped; | |
453 | ||
454 | if (del_timer(&ct->timeout)) { | |
455 | death_by_timeout((unsigned long)ct); | |
456 | dropped = 1; | |
c0e912d7 | 457 | NF_CT_STAT_INC_ATOMIC(early_drop); |
9fb9cbb1 YK |
458 | } |
459 | nf_ct_put(ct); | |
460 | return dropped; | |
461 | } | |
462 | ||
5a1fb391 AD |
463 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
464 | const struct nf_conntrack_tuple *orig, | |
b891c5a8 PNA |
465 | const struct nf_conntrack_tuple *repl, |
466 | gfp_t gfp) | |
9fb9cbb1 | 467 | { |
c88130bc | 468 | struct nf_conn *ct = NULL; |
9fb9cbb1 | 469 | |
dc808fe2 | 470 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
471 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
472 | nf_conntrack_hash_rnd_initted = 1; | |
473 | } | |
474 | ||
5251e2d2 | 475 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 476 | atomic_inc(&net->ct.count); |
5251e2d2 | 477 | |
76eb9460 | 478 | if (nf_conntrack_max && |
49ac8713 | 479 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
9fb9cbb1 | 480 | unsigned int hash = hash_conntrack(orig); |
400dad39 | 481 | if (!early_drop(net, hash)) { |
49ac8713 | 482 | atomic_dec(&net->ct.count); |
9fb9cbb1 YK |
483 | if (net_ratelimit()) |
484 | printk(KERN_WARNING | |
485 | "nf_conntrack: table full, dropping" | |
486 | " packet.\n"); | |
487 | return ERR_PTR(-ENOMEM); | |
488 | } | |
489 | } | |
490 | ||
b891c5a8 | 491 | ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp); |
c88130bc | 492 | if (ct == NULL) { |
0d53778e | 493 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
49ac8713 | 494 | atomic_dec(&net->ct.count); |
dacd2a1a | 495 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 YK |
496 | } |
497 | ||
c88130bc PM |
498 | atomic_set(&ct->ct_general.use, 1); |
499 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; | |
500 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
9fb9cbb1 | 501 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 502 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
5a1fb391 AD |
503 | #ifdef CONFIG_NET_NS |
504 | ct->ct_net = net; | |
505 | #endif | |
c88130bc | 506 | INIT_RCU_HEAD(&ct->rcu); |
9fb9cbb1 | 507 | |
c88130bc | 508 | return ct; |
9fb9cbb1 | 509 | } |
13b18339 | 510 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 511 | |
76507f69 | 512 | static void nf_conntrack_free_rcu(struct rcu_head *head) |
9fb9cbb1 | 513 | { |
76507f69 | 514 | struct nf_conn *ct = container_of(head, struct nf_conn, rcu); |
49ac8713 | 515 | struct net *net = nf_ct_net(ct); |
76507f69 PM |
516 | |
517 | nf_ct_ext_free(ct); | |
518 | kmem_cache_free(nf_conntrack_cachep, ct); | |
49ac8713 | 519 | atomic_dec(&net->ct.count); |
9fb9cbb1 | 520 | } |
76507f69 | 521 | |
c88130bc | 522 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 523 | { |
ceeff754 | 524 | nf_ct_ext_destroy(ct); |
c88130bc | 525 | call_rcu(&ct->rcu, nf_conntrack_free_rcu); |
76507f69 | 526 | } |
13b18339 | 527 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
528 | |
529 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
530 | failed due to stress. Otherwise it really is unclassifiable. */ | |
531 | static struct nf_conntrack_tuple_hash * | |
5a1fb391 AD |
532 | init_conntrack(struct net *net, |
533 | const struct nf_conntrack_tuple *tuple, | |
9fb9cbb1 | 534 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 535 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
536 | struct sk_buff *skb, |
537 | unsigned int dataoff) | |
538 | { | |
c88130bc | 539 | struct nf_conn *ct; |
3c158f7f | 540 | struct nf_conn_help *help; |
9fb9cbb1 YK |
541 | struct nf_conntrack_tuple repl_tuple; |
542 | struct nf_conntrack_expect *exp; | |
543 | ||
605dcad6 | 544 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 545 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
546 | return NULL; |
547 | } | |
548 | ||
5a1fb391 | 549 | ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC); |
c88130bc | 550 | if (ct == NULL || IS_ERR(ct)) { |
0d53778e | 551 | pr_debug("Can't allocate conntrack.\n"); |
c88130bc | 552 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 YK |
553 | } |
554 | ||
c88130bc PM |
555 | if (!l4proto->new(ct, skb, dataoff)) { |
556 | nf_conntrack_free(ct); | |
0d53778e | 557 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
558 | return NULL; |
559 | } | |
560 | ||
58401572 KPO |
561 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
562 | ||
f8ba1aff | 563 | spin_lock_bh(&nf_conntrack_lock); |
9b03f38d | 564 | exp = nf_ct_find_expectation(net, tuple); |
9fb9cbb1 | 565 | if (exp) { |
0d53778e | 566 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 567 | ct, exp); |
9fb9cbb1 | 568 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
569 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
570 | ct->master = exp->master; | |
ceceae1b | 571 | if (exp->helper) { |
c88130bc | 572 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b YK |
573 | if (help) |
574 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
575 | } |
576 | ||
9fb9cbb1 | 577 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 578 | ct->mark = exp->master->mark; |
7c9728c3 JM |
579 | #endif |
580 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 581 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 582 | #endif |
c88130bc | 583 | nf_conntrack_get(&ct->master->ct_general); |
9fb9cbb1 | 584 | NF_CT_STAT_INC(expect_new); |
22e7410b | 585 | } else { |
ceceae1b YK |
586 | struct nf_conntrack_helper *helper; |
587 | ||
588 | helper = __nf_ct_helper_find(&repl_tuple); | |
589 | if (helper) { | |
c88130bc | 590 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b | 591 | if (help) |
ceceae1b | 592 | rcu_assign_pointer(help->helper, helper); |
3c158f7f | 593 | } |
9fb9cbb1 | 594 | NF_CT_STAT_INC(new); |
22e7410b | 595 | } |
9fb9cbb1 YK |
596 | |
597 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
63c9a262 AD |
598 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
599 | &net->ct.unconfirmed); | |
9fb9cbb1 | 600 | |
f8ba1aff | 601 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
602 | |
603 | if (exp) { | |
604 | if (exp->expectfn) | |
c88130bc | 605 | exp->expectfn(ct, exp); |
6823645d | 606 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
607 | } |
608 | ||
c88130bc | 609 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
610 | } |
611 | ||
612 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
613 | static inline struct nf_conn * | |
a702a65f AD |
614 | resolve_normal_ct(struct net *net, |
615 | struct sk_buff *skb, | |
9fb9cbb1 YK |
616 | unsigned int dataoff, |
617 | u_int16_t l3num, | |
618 | u_int8_t protonum, | |
619 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 620 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
621 | int *set_reply, |
622 | enum ip_conntrack_info *ctinfo) | |
623 | { | |
624 | struct nf_conntrack_tuple tuple; | |
625 | struct nf_conntrack_tuple_hash *h; | |
626 | struct nf_conn *ct; | |
627 | ||
bbe735e4 | 628 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 629 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 630 | l4proto)) { |
0d53778e | 631 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
632 | return NULL; |
633 | } | |
634 | ||
635 | /* look for tuple match */ | |
a702a65f | 636 | h = nf_conntrack_find_get(net, &tuple); |
9fb9cbb1 | 637 | if (!h) { |
a702a65f | 638 | h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
639 | if (!h) |
640 | return NULL; | |
641 | if (IS_ERR(h)) | |
642 | return (void *)h; | |
643 | } | |
644 | ct = nf_ct_tuplehash_to_ctrack(h); | |
645 | ||
646 | /* It exists; we have (non-exclusive) reference. */ | |
647 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
648 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
649 | /* Please set reply bit if this packet OK */ | |
650 | *set_reply = 1; | |
651 | } else { | |
652 | /* Once we've had two way comms, always ESTABLISHED. */ | |
653 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 654 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
655 | *ctinfo = IP_CT_ESTABLISHED; |
656 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
657 | pr_debug("nf_conntrack_in: related packet for %p\n", |
658 | ct); | |
9fb9cbb1 YK |
659 | *ctinfo = IP_CT_RELATED; |
660 | } else { | |
0d53778e | 661 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
662 | *ctinfo = IP_CT_NEW; |
663 | } | |
664 | *set_reply = 0; | |
665 | } | |
666 | skb->nfct = &ct->ct_general; | |
667 | skb->nfctinfo = *ctinfo; | |
668 | return ct; | |
669 | } | |
670 | ||
671 | unsigned int | |
a702a65f AD |
672 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
673 | struct sk_buff *skb) | |
9fb9cbb1 YK |
674 | { |
675 | struct nf_conn *ct; | |
676 | enum ip_conntrack_info ctinfo; | |
677 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 678 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
679 | unsigned int dataoff; |
680 | u_int8_t protonum; | |
681 | int set_reply = 0; | |
682 | int ret; | |
683 | ||
684 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 685 | if (skb->nfct) { |
c0e912d7 | 686 | NF_CT_STAT_INC_ATOMIC(ignore); |
9fb9cbb1 YK |
687 | return NF_ACCEPT; |
688 | } | |
689 | ||
923f4902 | 690 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 691 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 692 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
693 | &dataoff, &protonum); |
694 | if (ret <= 0) { | |
0d53778e | 695 | pr_debug("not prepared to track yet or error occured\n"); |
d87d8469 YK |
696 | NF_CT_STAT_INC_ATOMIC(error); |
697 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
698 | return -ret; |
699 | } | |
700 | ||
76108cea | 701 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
702 | |
703 | /* It may be an special packet, error, unclean... | |
704 | * inverse of the return code tells to the netfilter | |
705 | * core what to do with the packet. */ | |
74c51a14 AD |
706 | if (l4proto->error != NULL) { |
707 | ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum); | |
708 | if (ret <= 0) { | |
709 | NF_CT_STAT_INC_ATOMIC(error); | |
710 | NF_CT_STAT_INC_ATOMIC(invalid); | |
711 | return -ret; | |
712 | } | |
9fb9cbb1 YK |
713 | } |
714 | ||
a702a65f AD |
715 | ct = resolve_normal_ct(net, skb, dataoff, pf, protonum, |
716 | l3proto, l4proto, &set_reply, &ctinfo); | |
9fb9cbb1 YK |
717 | if (!ct) { |
718 | /* Not valid part of a connection */ | |
c0e912d7 | 719 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
720 | return NF_ACCEPT; |
721 | } | |
722 | ||
723 | if (IS_ERR(ct)) { | |
724 | /* Too stressed to deal. */ | |
c0e912d7 | 725 | NF_CT_STAT_INC_ATOMIC(drop); |
9fb9cbb1 YK |
726 | return NF_DROP; |
727 | } | |
728 | ||
3db05fea | 729 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 730 | |
3db05fea | 731 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
732 | if (ret < 0) { |
733 | /* Invalid: inverse of the return code tells | |
734 | * the netfilter core what to do */ | |
0d53778e | 735 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
736 | nf_conntrack_put(skb->nfct); |
737 | skb->nfct = NULL; | |
c0e912d7 | 738 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
739 | return -ret; |
740 | } | |
741 | ||
742 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
a71996fc | 743 | nf_conntrack_event_cache(IPCT_STATUS, ct); |
9fb9cbb1 YK |
744 | |
745 | return ret; | |
746 | } | |
13b18339 | 747 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 748 | |
5f2b4c90 JE |
749 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
750 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 751 | { |
5f2b4c90 | 752 | bool ret; |
923f4902 PM |
753 | |
754 | rcu_read_lock(); | |
755 | ret = nf_ct_invert_tuple(inverse, orig, | |
756 | __nf_ct_l3proto_find(orig->src.l3num), | |
757 | __nf_ct_l4proto_find(orig->src.l3num, | |
758 | orig->dst.protonum)); | |
759 | rcu_read_unlock(); | |
760 | return ret; | |
9fb9cbb1 | 761 | } |
13b18339 | 762 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 763 | |
5b1158e9 JK |
764 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
765 | implicitly racy: see __nf_conntrack_confirm */ | |
766 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
767 | const struct nf_conntrack_tuple *newreply) | |
768 | { | |
769 | struct nf_conn_help *help = nfct_help(ct); | |
ceceae1b | 770 | struct nf_conntrack_helper *helper; |
5b1158e9 | 771 | |
5b1158e9 JK |
772 | /* Should be unconfirmed, so not in hash table yet */ |
773 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
774 | ||
0d53778e | 775 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 776 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
777 | |
778 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 779 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 780 | return; |
ceceae1b | 781 | |
c52fbb41 | 782 | rcu_read_lock(); |
ceceae1b YK |
783 | helper = __nf_ct_helper_find(newreply); |
784 | if (helper == NULL) { | |
785 | if (help) | |
786 | rcu_assign_pointer(help->helper, NULL); | |
787 | goto out; | |
5d78a849 | 788 | } |
ceceae1b YK |
789 | |
790 | if (help == NULL) { | |
b560580a PM |
791 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
792 | if (help == NULL) | |
ceceae1b | 793 | goto out; |
ceceae1b YK |
794 | } else { |
795 | memset(&help->help, 0, sizeof(help->help)); | |
796 | } | |
797 | ||
798 | rcu_assign_pointer(help->helper, helper); | |
799 | out: | |
c52fbb41 | 800 | rcu_read_unlock(); |
5b1158e9 | 801 | } |
13b18339 | 802 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 803 | |
9fb9cbb1 YK |
804 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
805 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
806 | enum ip_conntrack_info ctinfo, | |
807 | const struct sk_buff *skb, | |
808 | unsigned long extra_jiffies, | |
809 | int do_acct) | |
810 | { | |
811 | int event = 0; | |
812 | ||
813 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
814 | NF_CT_ASSERT(skb); | |
815 | ||
f8ba1aff | 816 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 817 | |
997ae831 | 818 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
819 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
820 | goto acct; | |
997ae831 | 821 | |
9fb9cbb1 YK |
822 | /* If not in hash table, timer will not be active yet */ |
823 | if (!nf_ct_is_confirmed(ct)) { | |
824 | ct->timeout.expires = extra_jiffies; | |
825 | event = IPCT_REFRESH; | |
826 | } else { | |
be00c8e4 MJ |
827 | unsigned long newtime = jiffies + extra_jiffies; |
828 | ||
829 | /* Only update the timeout if the new timeout is at least | |
830 | HZ jiffies from the old timeout. Need del_timer for race | |
831 | avoidance (may already be dying). */ | |
832 | if (newtime - ct->timeout.expires >= HZ | |
833 | && del_timer(&ct->timeout)) { | |
834 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
835 | add_timer(&ct->timeout); |
836 | event = IPCT_REFRESH; | |
837 | } | |
838 | } | |
839 | ||
47d95045 | 840 | acct: |
9fb9cbb1 | 841 | if (do_acct) { |
58401572 | 842 | struct nf_conn_counter *acct; |
3ffd5eeb | 843 | |
58401572 KPO |
844 | acct = nf_conn_acct_find(ct); |
845 | if (acct) { | |
846 | acct[CTINFO2DIR(ctinfo)].packets++; | |
847 | acct[CTINFO2DIR(ctinfo)].bytes += | |
848 | skb->len - skb_network_offset(skb); | |
849 | } | |
9fb9cbb1 | 850 | } |
9fb9cbb1 | 851 | |
f8ba1aff | 852 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
853 | |
854 | /* must be unlocked when calling event cache */ | |
855 | if (event) | |
a71996fc | 856 | nf_conntrack_event_cache(event, ct); |
9fb9cbb1 | 857 | } |
13b18339 | 858 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 859 | |
4c889498 DM |
860 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
861 | enum ip_conntrack_info ctinfo, | |
862 | const struct sk_buff *skb, | |
863 | int do_acct) | |
51091764 | 864 | { |
718d4ad9 | 865 | if (do_acct) { |
58401572 KPO |
866 | struct nf_conn_counter *acct; |
867 | ||
718d4ad9 | 868 | spin_lock_bh(&nf_conntrack_lock); |
58401572 KPO |
869 | acct = nf_conn_acct_find(ct); |
870 | if (acct) { | |
871 | acct[CTINFO2DIR(ctinfo)].packets++; | |
872 | acct[CTINFO2DIR(ctinfo)].bytes += | |
873 | skb->len - skb_network_offset(skb); | |
874 | } | |
718d4ad9 FH |
875 | spin_unlock_bh(&nf_conntrack_lock); |
876 | } | |
58401572 | 877 | |
4c889498 | 878 | if (del_timer(&ct->timeout)) { |
51091764 | 879 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
880 | return true; |
881 | } | |
882 | return false; | |
51091764 | 883 | } |
718d4ad9 | 884 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 885 | |
e281db5c | 886 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
887 | |
888 | #include <linux/netfilter/nfnetlink.h> | |
889 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
890 | #include <linux/mutex.h> |
891 | ||
c1d10adb PNA |
892 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
893 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
894 | * or depend on ctnetlink */ | |
fdf70832 | 895 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
896 | const struct nf_conntrack_tuple *tuple) |
897 | { | |
77236b6e PM |
898 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
899 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
900 | return 0; |
901 | ||
df6fb868 | 902 | nla_put_failure: |
c1d10adb PNA |
903 | return -1; |
904 | } | |
fdf70832 | 905 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 906 | |
f73e924c PM |
907 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
908 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
909 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 910 | }; |
f73e924c | 911 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 912 | |
fdf70832 | 913 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
914 | struct nf_conntrack_tuple *t) |
915 | { | |
df6fb868 | 916 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
917 | return -EINVAL; |
918 | ||
77236b6e PM |
919 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
920 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
921 | |
922 | return 0; | |
923 | } | |
fdf70832 | 924 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
c1d10adb PNA |
925 | #endif |
926 | ||
9fb9cbb1 | 927 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 928 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
929 | { |
930 | struct nf_conn *ct; | |
931 | enum ip_conntrack_info ctinfo; | |
932 | ||
933 | /* This ICMP is in reverse direction to the packet which caused it */ | |
934 | ct = nf_ct_get(skb, &ctinfo); | |
935 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
936 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
937 | else | |
938 | ctinfo = IP_CT_RELATED; | |
939 | ||
940 | /* Attach to new skbuff, and increment count */ | |
941 | nskb->nfct = &ct->ct_general; | |
942 | nskb->nfctinfo = ctinfo; | |
943 | nf_conntrack_get(nskb->nfct); | |
944 | } | |
945 | ||
9fb9cbb1 | 946 | /* Bring out ya dead! */ |
df0933dc | 947 | static struct nf_conn * |
400dad39 | 948 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
949 | void *data, unsigned int *bucket) |
950 | { | |
df0933dc PM |
951 | struct nf_conntrack_tuple_hash *h; |
952 | struct nf_conn *ct; | |
f205c5e0 | 953 | struct hlist_node *n; |
9fb9cbb1 | 954 | |
f8ba1aff | 955 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 956 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
400dad39 | 957 | hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) { |
df0933dc PM |
958 | ct = nf_ct_tuplehash_to_ctrack(h); |
959 | if (iter(ct, data)) | |
960 | goto found; | |
961 | } | |
601e68e1 | 962 | } |
63c9a262 | 963 | hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) { |
df0933dc PM |
964 | ct = nf_ct_tuplehash_to_ctrack(h); |
965 | if (iter(ct, data)) | |
ec68e97d | 966 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 967 | } |
f8ba1aff | 968 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
969 | return NULL; |
970 | found: | |
c073e3fa | 971 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 972 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 973 | return ct; |
9fb9cbb1 YK |
974 | } |
975 | ||
400dad39 AD |
976 | void nf_ct_iterate_cleanup(struct net *net, |
977 | int (*iter)(struct nf_conn *i, void *data), | |
978 | void *data) | |
9fb9cbb1 | 979 | { |
df0933dc | 980 | struct nf_conn *ct; |
9fb9cbb1 YK |
981 | unsigned int bucket = 0; |
982 | ||
400dad39 | 983 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
984 | /* Time to push up daises... */ |
985 | if (del_timer(&ct->timeout)) | |
986 | death_by_timeout((unsigned long)ct); | |
987 | /* ... else the timer will get him soon. */ | |
988 | ||
989 | nf_ct_put(ct); | |
990 | } | |
991 | } | |
13b18339 | 992 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 YK |
993 | |
994 | static int kill_all(struct nf_conn *i, void *data) | |
995 | { | |
996 | return 1; | |
997 | } | |
998 | ||
96eb24d7 | 999 | void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
1000 | { |
1001 | if (vmalloced) | |
1002 | vfree(hash); | |
1003 | else | |
601e68e1 | 1004 | free_pages((unsigned long)hash, |
f205c5e0 | 1005 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1006 | } |
ac565e5f | 1007 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1008 | |
400dad39 | 1009 | void nf_conntrack_flush(struct net *net) |
c1d10adb | 1010 | { |
400dad39 | 1011 | nf_ct_iterate_cleanup(net, kill_all, NULL); |
c1d10adb | 1012 | } |
13b18339 | 1013 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 1014 | |
9fb9cbb1 YK |
1015 | /* Mishearing the voices in his head, our hero wonders how he's |
1016 | supposed to kill the mall. */ | |
dfdb8d79 | 1017 | void nf_conntrack_cleanup(struct net *net) |
9fb9cbb1 | 1018 | { |
c3a47ab3 | 1019 | rcu_assign_pointer(ip_ct_attach, NULL); |
7d3cdc6b | 1020 | |
9fb9cbb1 YK |
1021 | /* This makes sure all current packets have passed through |
1022 | netfilter framework. Roll on, two-stage module | |
1023 | delete... */ | |
1024 | synchronize_net(); | |
1025 | ||
6058fa6b AD |
1026 | nf_ct_event_cache_flush(net); |
1027 | nf_conntrack_ecache_fini(net); | |
9fb9cbb1 | 1028 | i_see_dead_people: |
400dad39 | 1029 | nf_conntrack_flush(net); |
49ac8713 | 1030 | if (atomic_read(&net->ct.count) != 0) { |
9fb9cbb1 YK |
1031 | schedule(); |
1032 | goto i_see_dead_people; | |
1033 | } | |
6636568c PM |
1034 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1035 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1036 | schedule(); | |
9fb9cbb1 | 1037 | |
de6e05c4 YK |
1038 | rcu_assign_pointer(nf_ct_destroy, NULL); |
1039 | ||
dacd2a1a | 1040 | kmem_cache_destroy(nf_conntrack_cachep); |
400dad39 | 1041 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1042 | nf_conntrack_htable_size); |
5a6f294e | 1043 | |
58401572 | 1044 | nf_conntrack_acct_fini(); |
9b03f38d | 1045 | nf_conntrack_expect_fini(net); |
9714be7d KPO |
1046 | nf_conntrack_helper_fini(); |
1047 | nf_conntrack_proto_fini(); | |
9fb9cbb1 YK |
1048 | } |
1049 | ||
96eb24d7 | 1050 | struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced) |
9fb9cbb1 | 1051 | { |
f205c5e0 | 1052 | struct hlist_head *hash; |
8e5105a0 | 1053 | unsigned int size, i; |
9fb9cbb1 | 1054 | |
601e68e1 | 1055 | *vmalloced = 0; |
8e5105a0 | 1056 | |
f205c5e0 | 1057 | size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); |
29b67497 | 1058 | hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN, |
f205c5e0 | 1059 | get_order(sizeof(struct hlist_head) |
9fb9cbb1 | 1060 | * size)); |
601e68e1 | 1061 | if (!hash) { |
9fb9cbb1 YK |
1062 | *vmalloced = 1; |
1063 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
f205c5e0 | 1064 | hash = vmalloc(sizeof(struct hlist_head) * size); |
9fb9cbb1 YK |
1065 | } |
1066 | ||
1067 | if (hash) | |
601e68e1 | 1068 | for (i = 0; i < size; i++) |
f205c5e0 | 1069 | INIT_HLIST_HEAD(&hash[i]); |
9fb9cbb1 YK |
1070 | |
1071 | return hash; | |
1072 | } | |
ac565e5f | 1073 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1074 | |
fae718dd | 1075 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1076 | { |
96eb24d7 SH |
1077 | int i, bucket, vmalloced, old_vmalloced; |
1078 | unsigned int hashsize, old_size; | |
9fb9cbb1 | 1079 | int rnd; |
f205c5e0 | 1080 | struct hlist_head *hash, *old_hash; |
9fb9cbb1 YK |
1081 | struct nf_conntrack_tuple_hash *h; |
1082 | ||
1083 | /* On boot, we can set this without any fancy locking. */ | |
1084 | if (!nf_conntrack_htable_size) | |
1085 | return param_set_uint(val, kp); | |
1086 | ||
96eb24d7 | 1087 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1088 | if (!hashsize) |
1089 | return -EINVAL; | |
1090 | ||
ac565e5f | 1091 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); |
9fb9cbb1 YK |
1092 | if (!hash) |
1093 | return -ENOMEM; | |
1094 | ||
1095 | /* We have to rehahs for the new table anyway, so we also can | |
1096 | * use a newrandom seed */ | |
1097 | get_random_bytes(&rnd, 4); | |
1098 | ||
76507f69 PM |
1099 | /* Lookups in the old hash might happen in parallel, which means we |
1100 | * might get false negatives during connection lookup. New connections | |
1101 | * created because of a false negative won't make it into the hash | |
1102 | * though since that required taking the lock. | |
1103 | */ | |
f8ba1aff | 1104 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1105 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
400dad39 AD |
1106 | while (!hlist_empty(&init_net.ct.hash[i])) { |
1107 | h = hlist_entry(init_net.ct.hash[i].first, | |
f205c5e0 | 1108 | struct nf_conntrack_tuple_hash, hnode); |
76507f69 | 1109 | hlist_del_rcu(&h->hnode); |
9fb9cbb1 | 1110 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
f205c5e0 | 1111 | hlist_add_head(&h->hnode, &hash[bucket]); |
9fb9cbb1 YK |
1112 | } |
1113 | } | |
1114 | old_size = nf_conntrack_htable_size; | |
400dad39 AD |
1115 | old_vmalloced = init_net.ct.hash_vmalloc; |
1116 | old_hash = init_net.ct.hash; | |
9fb9cbb1 YK |
1117 | |
1118 | nf_conntrack_htable_size = hashsize; | |
400dad39 AD |
1119 | init_net.ct.hash_vmalloc = vmalloced; |
1120 | init_net.ct.hash = hash; | |
9fb9cbb1 | 1121 | nf_conntrack_hash_rnd = rnd; |
f8ba1aff | 1122 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1123 | |
ac565e5f | 1124 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1125 | return 0; |
1126 | } | |
fae718dd | 1127 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1128 | |
fae718dd | 1129 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1130 | &nf_conntrack_htable_size, 0600); |
1131 | ||
dfdb8d79 | 1132 | int nf_conntrack_init(struct net *net) |
9fb9cbb1 | 1133 | { |
f205c5e0 | 1134 | int max_factor = 8; |
9fb9cbb1 YK |
1135 | int ret; |
1136 | ||
1137 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1138 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1139 | if (!nf_conntrack_htable_size) { |
1140 | nf_conntrack_htable_size | |
1141 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
f205c5e0 | 1142 | / sizeof(struct hlist_head)); |
9fb9cbb1 | 1143 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1144 | nf_conntrack_htable_size = 16384; |
1145 | if (nf_conntrack_htable_size < 32) | |
1146 | nf_conntrack_htable_size = 32; | |
1147 | ||
1148 | /* Use a max. factor of four by default to get the same max as | |
1149 | * with the old struct list_heads. When a table size is given | |
1150 | * we use the old value of 8 to avoid reducing the max. | |
1151 | * entries. */ | |
1152 | max_factor = 4; | |
9fb9cbb1 | 1153 | } |
49ac8713 | 1154 | atomic_set(&net->ct.count, 0); |
6058fa6b AD |
1155 | ret = nf_conntrack_ecache_init(net); |
1156 | if (ret < 0) | |
1157 | goto err_ecache; | |
400dad39 AD |
1158 | net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
1159 | &net->ct.hash_vmalloc); | |
1160 | if (!net->ct.hash) { | |
9fb9cbb1 | 1161 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); |
6058fa6b | 1162 | goto err_hash; |
9fb9cbb1 | 1163 | } |
63c9a262 | 1164 | INIT_HLIST_HEAD(&net->ct.unconfirmed); |
9fb9cbb1 | 1165 | |
f205c5e0 | 1166 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1167 | |
1168 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1169 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1170 | nf_conntrack_max); | |
1171 | ||
dacd2a1a YK |
1172 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1173 | sizeof(struct nf_conn), | |
20c2df83 | 1174 | 0, 0, NULL); |
dacd2a1a | 1175 | if (!nf_conntrack_cachep) { |
9fb9cbb1 YK |
1176 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
1177 | goto err_free_hash; | |
1178 | } | |
1179 | ||
e9c1b084 PM |
1180 | ret = nf_conntrack_proto_init(); |
1181 | if (ret < 0) | |
9fb9cbb1 | 1182 | goto err_free_conntrack_slab; |
9fb9cbb1 | 1183 | |
9b03f38d | 1184 | ret = nf_conntrack_expect_init(net); |
933a41e7 | 1185 | if (ret < 0) |
e9c1b084 | 1186 | goto out_fini_proto; |
933a41e7 | 1187 | |
ceceae1b YK |
1188 | ret = nf_conntrack_helper_init(); |
1189 | if (ret < 0) | |
e9c1b084 | 1190 | goto out_fini_expect; |
ceceae1b | 1191 | |
58401572 KPO |
1192 | ret = nf_conntrack_acct_init(); |
1193 | if (ret < 0) | |
1194 | goto out_fini_helper; | |
1195 | ||
7d3cdc6b | 1196 | /* For use by REJECT target */ |
b334aadc | 1197 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); |
de6e05c4 | 1198 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); |
7d3cdc6b | 1199 | |
9fb9cbb1 YK |
1200 | /* Set up fake conntrack: |
1201 | - to never be deleted, not in any hashes */ | |
5a1fb391 AD |
1202 | #ifdef CONFIG_NET_NS |
1203 | nf_conntrack_untracked.ct_net = &init_net; | |
1204 | #endif | |
9fb9cbb1 YK |
1205 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); |
1206 | /* - and look it like as a confirmed connection */ | |
1207 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1208 | ||
1209 | return ret; | |
1210 | ||
58401572 KPO |
1211 | out_fini_helper: |
1212 | nf_conntrack_helper_fini(); | |
e9c1b084 | 1213 | out_fini_expect: |
9b03f38d | 1214 | nf_conntrack_expect_fini(net); |
ceceae1b YK |
1215 | out_fini_proto: |
1216 | nf_conntrack_proto_fini(); | |
9fb9cbb1 | 1217 | err_free_conntrack_slab: |
dacd2a1a | 1218 | kmem_cache_destroy(nf_conntrack_cachep); |
9fb9cbb1 | 1219 | err_free_hash: |
400dad39 | 1220 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1221 | nf_conntrack_htable_size); |
6058fa6b AD |
1222 | err_hash: |
1223 | nf_conntrack_ecache_fini(net); | |
1224 | err_ecache: | |
9fb9cbb1 YK |
1225 | return -ENOMEM; |
1226 | } |