]>
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> |
e6a7d3c0 | 41 | #include <net/netfilter/nf_nat.h> |
e17b666a | 42 | #include <net/netfilter/nf_nat_core.h> |
9fb9cbb1 | 43 | |
dc808fe2 | 44 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 45 | |
e17b666a PM |
46 | int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, |
47 | enum nf_nat_manip_type manip, | |
48 | struct nlattr *attr) __read_mostly; | |
e6a7d3c0 PNA |
49 | EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); |
50 | ||
f8ba1aff | 51 | DEFINE_SPINLOCK(nf_conntrack_lock); |
13b18339 | 52 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 | 53 | |
e2b7606c | 54 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
55 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
56 | ||
94aec08e | 57 | int nf_conntrack_max __read_mostly; |
a999e683 | 58 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 59 | |
e2b7606c | 60 | struct nf_conn nf_conntrack_untracked __read_mostly; |
13b18339 PM |
61 | EXPORT_SYMBOL_GPL(nf_conntrack_untracked); |
62 | ||
dacd2a1a | 63 | static struct kmem_cache *nf_conntrack_cachep __read_mostly; |
77ab9cff | 64 | |
9fb9cbb1 YK |
65 | static int nf_conntrack_hash_rnd_initted; |
66 | static unsigned int nf_conntrack_hash_rnd; | |
67 | ||
68 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
69 | unsigned int size, unsigned int rnd) | |
70 | { | |
0794935e PM |
71 | unsigned int n; |
72 | u_int32_t h; | |
73 | ||
74 | /* The direction must be ignored, so we hash everything up to the | |
75 | * destination ports (which is a multiple of 4) and treat the last | |
76 | * three bytes manually. | |
77 | */ | |
78 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
79 | h = jhash2((u32 *)tuple, n, | |
80 | rnd ^ (((__force __u16)tuple->dst.u.all << 16) | | |
81 | tuple->dst.protonum)); | |
82 | ||
83 | return ((u64)h * size) >> 32; | |
9fb9cbb1 YK |
84 | } |
85 | ||
86 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
87 | { | |
88 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
89 | nf_conntrack_hash_rnd); | |
90 | } | |
91 | ||
5f2b4c90 | 92 | bool |
9fb9cbb1 YK |
93 | nf_ct_get_tuple(const struct sk_buff *skb, |
94 | unsigned int nhoff, | |
95 | unsigned int dataoff, | |
96 | u_int16_t l3num, | |
97 | u_int8_t protonum, | |
98 | struct nf_conntrack_tuple *tuple, | |
99 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 100 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 101 | { |
443a70d5 | 102 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
103 | |
104 | tuple->src.l3num = l3num; | |
105 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 106 | return false; |
9fb9cbb1 YK |
107 | |
108 | tuple->dst.protonum = protonum; | |
109 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
110 | ||
605dcad6 | 111 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 112 | } |
13b18339 | 113 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 114 | |
5f2b4c90 JE |
115 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
116 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
117 | { |
118 | struct nf_conntrack_l3proto *l3proto; | |
119 | struct nf_conntrack_l4proto *l4proto; | |
120 | unsigned int protoff; | |
121 | u_int8_t protonum; | |
122 | int ret; | |
123 | ||
124 | rcu_read_lock(); | |
125 | ||
126 | l3proto = __nf_ct_l3proto_find(l3num); | |
127 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
128 | if (ret != NF_ACCEPT) { | |
129 | rcu_read_unlock(); | |
5f2b4c90 | 130 | return false; |
e2a3123f YK |
131 | } |
132 | ||
133 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
134 | ||
135 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
136 | l3proto, l4proto); | |
137 | ||
138 | rcu_read_unlock(); | |
139 | return ret; | |
140 | } | |
141 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
142 | ||
5f2b4c90 | 143 | bool |
9fb9cbb1 YK |
144 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
145 | const struct nf_conntrack_tuple *orig, | |
146 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 147 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 148 | { |
443a70d5 | 149 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
150 | |
151 | inverse->src.l3num = orig->src.l3num; | |
152 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 153 | return false; |
9fb9cbb1 YK |
154 | |
155 | inverse->dst.dir = !orig->dst.dir; | |
156 | ||
157 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 158 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 159 | } |
13b18339 | 160 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 161 | |
9fb9cbb1 YK |
162 | static void |
163 | clean_from_lists(struct nf_conn *ct) | |
164 | { | |
0d53778e | 165 | pr_debug("clean_from_lists(%p)\n", ct); |
76507f69 PM |
166 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
167 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); | |
9fb9cbb1 YK |
168 | |
169 | /* Destroy all pending expectations */ | |
c1d10adb | 170 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
171 | } |
172 | ||
173 | static void | |
174 | destroy_conntrack(struct nf_conntrack *nfct) | |
175 | { | |
176 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
0d55af87 | 177 | struct net *net = nf_ct_net(ct); |
605dcad6 | 178 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 179 | |
0d53778e | 180 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
181 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
182 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
183 | ||
19abb7b0 PNA |
184 | if (!test_bit(IPS_DYING_BIT, &ct->status)) |
185 | nf_conntrack_event(IPCT_DESTROY, ct); | |
9fb9cbb1 YK |
186 | set_bit(IPS_DYING_BIT, &ct->status); |
187 | ||
188 | /* To make sure we don't get any weird locking issues here: | |
189 | * destroy_conntrack() MUST NOT be called with a write lock | |
190 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 191 | rcu_read_lock(); |
5e8fbe2a | 192 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
193 | if (l4proto && l4proto->destroy) |
194 | l4proto->destroy(ct); | |
9fb9cbb1 | 195 | |
982d9a9c | 196 | rcu_read_unlock(); |
9fb9cbb1 | 197 | |
f8ba1aff | 198 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
199 | /* Expectations will have been removed in clean_from_lists, |
200 | * except TFTP can create an expectation on the first packet, | |
201 | * before connection is in the list, so we need to clean here, | |
202 | * too. */ | |
c1d10adb | 203 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
204 | |
205 | /* We overload first tuple to link into unconfirmed list. */ | |
206 | if (!nf_ct_is_confirmed(ct)) { | |
f205c5e0 PM |
207 | BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode)); |
208 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); | |
9fb9cbb1 YK |
209 | } |
210 | ||
0d55af87 | 211 | NF_CT_STAT_INC(net, delete); |
f8ba1aff | 212 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
213 | |
214 | if (ct->master) | |
215 | nf_ct_put(ct->master); | |
216 | ||
0d53778e | 217 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
218 | nf_conntrack_free(ct); |
219 | } | |
220 | ||
221 | static void death_by_timeout(unsigned long ul_conntrack) | |
222 | { | |
223 | struct nf_conn *ct = (void *)ul_conntrack; | |
0d55af87 | 224 | struct net *net = nf_ct_net(ct); |
5397e97d | 225 | struct nf_conn_help *help = nfct_help(ct); |
3c158f7f | 226 | struct nf_conntrack_helper *helper; |
5397e97d | 227 | |
3c158f7f PM |
228 | if (help) { |
229 | rcu_read_lock(); | |
230 | helper = rcu_dereference(help->helper); | |
231 | if (helper && helper->destroy) | |
232 | helper->destroy(ct); | |
233 | rcu_read_unlock(); | |
234 | } | |
9fb9cbb1 | 235 | |
f8ba1aff | 236 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
237 | /* Inside lock so preempt is disabled on module removal path. |
238 | * Otherwise we can get spurious warnings. */ | |
0d55af87 | 239 | NF_CT_STAT_INC(net, delete_list); |
9fb9cbb1 | 240 | clean_from_lists(ct); |
f8ba1aff | 241 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
242 | nf_ct_put(ct); |
243 | } | |
244 | ||
c1d10adb | 245 | struct nf_conntrack_tuple_hash * |
400dad39 | 246 | __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
247 | { |
248 | struct nf_conntrack_tuple_hash *h; | |
f205c5e0 | 249 | struct hlist_node *n; |
9fb9cbb1 YK |
250 | unsigned int hash = hash_conntrack(tuple); |
251 | ||
4e29e9ec PM |
252 | /* Disable BHs the entire time since we normally need to disable them |
253 | * at least once for the stats anyway. | |
254 | */ | |
255 | local_bh_disable(); | |
400dad39 | 256 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) { |
ba419aff | 257 | if (nf_ct_tuple_equal(tuple, &h->tuple)) { |
0d55af87 | 258 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 259 | local_bh_enable(); |
9fb9cbb1 YK |
260 | return h; |
261 | } | |
0d55af87 | 262 | NF_CT_STAT_INC(net, searched); |
9fb9cbb1 | 263 | } |
4e29e9ec | 264 | local_bh_enable(); |
9fb9cbb1 YK |
265 | |
266 | return NULL; | |
267 | } | |
13b18339 | 268 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
269 | |
270 | /* Find a connection corresponding to a tuple. */ | |
271 | struct nf_conntrack_tuple_hash * | |
400dad39 | 272 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
273 | { |
274 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 275 | struct nf_conn *ct; |
9fb9cbb1 | 276 | |
76507f69 | 277 | rcu_read_lock(); |
400dad39 | 278 | h = __nf_conntrack_find(net, tuple); |
76507f69 PM |
279 | if (h) { |
280 | ct = nf_ct_tuplehash_to_ctrack(h); | |
281 | if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
282 | h = NULL; | |
283 | } | |
284 | rcu_read_unlock(); | |
9fb9cbb1 YK |
285 | |
286 | return h; | |
287 | } | |
13b18339 | 288 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 289 | |
c1d10adb PNA |
290 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
291 | unsigned int hash, | |
601e68e1 | 292 | unsigned int repl_hash) |
c1d10adb | 293 | { |
400dad39 AD |
294 | struct net *net = nf_ct_net(ct); |
295 | ||
76507f69 | 296 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
400dad39 | 297 | &net->ct.hash[hash]); |
76507f69 | 298 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, |
400dad39 | 299 | &net->ct.hash[repl_hash]); |
c1d10adb PNA |
300 | } |
301 | ||
302 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
303 | { | |
304 | unsigned int hash, repl_hash; | |
305 | ||
306 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
307 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
308 | ||
c1d10adb | 309 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
c1d10adb | 310 | } |
13b18339 | 311 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 312 | |
9fb9cbb1 YK |
313 | /* Confirm a connection given skb; places it in hash table */ |
314 | int | |
3db05fea | 315 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
316 | { |
317 | unsigned int hash, repl_hash; | |
df0933dc | 318 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 319 | struct nf_conn *ct; |
df0933dc | 320 | struct nf_conn_help *help; |
f205c5e0 | 321 | struct hlist_node *n; |
9fb9cbb1 | 322 | enum ip_conntrack_info ctinfo; |
400dad39 | 323 | struct net *net; |
9fb9cbb1 | 324 | |
3db05fea | 325 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 326 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
327 | |
328 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
329 | ICMP/TCP RST packets in other direction. Actual packet | |
330 | which created connection will be IP_CT_NEW or for an | |
331 | expected connection, IP_CT_RELATED. */ | |
332 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
333 | return NF_ACCEPT; | |
334 | ||
335 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
336 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
337 | ||
338 | /* We're not in hash table, and we refuse to set up related | |
339 | connections for unconfirmed conns. But packet copies and | |
340 | REJECT will give spurious warnings here. */ | |
341 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
342 | ||
343 | /* No external references means noone else could have | |
344 | confirmed us. */ | |
345 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 346 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 347 | |
f8ba1aff | 348 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
349 | |
350 | /* See if there's one in the list already, including reverse: | |
351 | NAT could have grabbed it without realizing, since we're | |
352 | not in the hash. If there is, we lost race. */ | |
400dad39 | 353 | hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode) |
df0933dc PM |
354 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
355 | &h->tuple)) | |
356 | goto out; | |
400dad39 | 357 | hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode) |
df0933dc PM |
358 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
359 | &h->tuple)) | |
360 | goto out; | |
9fb9cbb1 | 361 | |
df0933dc | 362 | /* Remove from unconfirmed list */ |
f205c5e0 | 363 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
df0933dc PM |
364 | |
365 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
366 | /* Timer relative to confirmation time, not original | |
367 | setting time, otherwise we'd get timer wrap in | |
368 | weird delay cases. */ | |
369 | ct->timeout.expires += jiffies; | |
370 | add_timer(&ct->timeout); | |
371 | atomic_inc(&ct->ct_general.use); | |
372 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
0d55af87 | 373 | NF_CT_STAT_INC(net, insert); |
f8ba1aff | 374 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
375 | help = nfct_help(ct); |
376 | if (help && help->helper) | |
a71996fc | 377 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
9fb9cbb1 | 378 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
379 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
380 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
a71996fc | 381 | nf_conntrack_event_cache(IPCT_NATINFO, ct); |
9fb9cbb1 | 382 | #endif |
df0933dc | 383 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 384 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 385 | return NF_ACCEPT; |
9fb9cbb1 | 386 | |
df0933dc | 387 | out: |
0d55af87 | 388 | NF_CT_STAT_INC(net, insert_failed); |
f8ba1aff | 389 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
390 | return NF_DROP; |
391 | } | |
13b18339 | 392 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
393 | |
394 | /* Returns true if a connection correspondings to the tuple (required | |
395 | for NAT). */ | |
396 | int | |
397 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
398 | const struct nf_conn *ignored_conntrack) | |
399 | { | |
400dad39 | 400 | struct net *net = nf_ct_net(ignored_conntrack); |
9fb9cbb1 | 401 | struct nf_conntrack_tuple_hash *h; |
ba419aff PM |
402 | struct hlist_node *n; |
403 | unsigned int hash = hash_conntrack(tuple); | |
9fb9cbb1 | 404 | |
4e29e9ec PM |
405 | /* Disable BHs the entire time since we need to disable them at |
406 | * least once for the stats anyway. | |
407 | */ | |
408 | rcu_read_lock_bh(); | |
400dad39 | 409 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) { |
ba419aff PM |
410 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
411 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
0d55af87 | 412 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 413 | rcu_read_unlock_bh(); |
ba419aff PM |
414 | return 1; |
415 | } | |
0d55af87 | 416 | NF_CT_STAT_INC(net, searched); |
ba419aff | 417 | } |
4e29e9ec | 418 | rcu_read_unlock_bh(); |
9fb9cbb1 | 419 | |
ba419aff | 420 | return 0; |
9fb9cbb1 | 421 | } |
13b18339 | 422 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 423 | |
7ae7730f PM |
424 | #define NF_CT_EVICTION_RANGE 8 |
425 | ||
9fb9cbb1 YK |
426 | /* There's a small race here where we may free a just-assured |
427 | connection. Too bad: we're in trouble anyway. */ | |
400dad39 | 428 | static noinline int early_drop(struct net *net, unsigned int hash) |
9fb9cbb1 | 429 | { |
f205c5e0 | 430 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 431 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 432 | struct nf_conn *ct = NULL, *tmp; |
f205c5e0 | 433 | struct hlist_node *n; |
7ae7730f | 434 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
435 | int dropped = 0; |
436 | ||
76507f69 | 437 | rcu_read_lock(); |
7ae7730f | 438 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
400dad39 | 439 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], |
76507f69 | 440 | hnode) { |
7ae7730f PM |
441 | tmp = nf_ct_tuplehash_to_ctrack(h); |
442 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
443 | ct = tmp; | |
444 | cnt++; | |
445 | } | |
76507f69 PM |
446 | |
447 | if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
448 | ct = NULL; | |
7ae7730f PM |
449 | if (ct || cnt >= NF_CT_EVICTION_RANGE) |
450 | break; | |
451 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 452 | } |
76507f69 | 453 | rcu_read_unlock(); |
9fb9cbb1 YK |
454 | |
455 | if (!ct) | |
456 | return dropped; | |
457 | ||
458 | if (del_timer(&ct->timeout)) { | |
459 | death_by_timeout((unsigned long)ct); | |
460 | dropped = 1; | |
0d55af87 | 461 | NF_CT_STAT_INC_ATOMIC(net, early_drop); |
9fb9cbb1 YK |
462 | } |
463 | nf_ct_put(ct); | |
464 | return dropped; | |
465 | } | |
466 | ||
5a1fb391 AD |
467 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
468 | const struct nf_conntrack_tuple *orig, | |
b891c5a8 PNA |
469 | const struct nf_conntrack_tuple *repl, |
470 | gfp_t gfp) | |
9fb9cbb1 | 471 | { |
cd7fcbf1 | 472 | struct nf_conn *ct; |
9fb9cbb1 | 473 | |
dc808fe2 | 474 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
475 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
476 | nf_conntrack_hash_rnd_initted = 1; | |
477 | } | |
478 | ||
5251e2d2 | 479 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 480 | atomic_inc(&net->ct.count); |
5251e2d2 | 481 | |
76eb9460 | 482 | if (nf_conntrack_max && |
49ac8713 | 483 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
9fb9cbb1 | 484 | unsigned int hash = hash_conntrack(orig); |
400dad39 | 485 | if (!early_drop(net, hash)) { |
49ac8713 | 486 | atomic_dec(&net->ct.count); |
9fb9cbb1 YK |
487 | if (net_ratelimit()) |
488 | printk(KERN_WARNING | |
489 | "nf_conntrack: table full, dropping" | |
490 | " packet.\n"); | |
491 | return ERR_PTR(-ENOMEM); | |
492 | } | |
493 | } | |
494 | ||
b891c5a8 | 495 | ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp); |
c88130bc | 496 | if (ct == NULL) { |
0d53778e | 497 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
49ac8713 | 498 | atomic_dec(&net->ct.count); |
dacd2a1a | 499 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 YK |
500 | } |
501 | ||
c88130bc PM |
502 | atomic_set(&ct->ct_general.use, 1); |
503 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; | |
504 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
9fb9cbb1 | 505 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 506 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
5a1fb391 AD |
507 | #ifdef CONFIG_NET_NS |
508 | ct->ct_net = net; | |
509 | #endif | |
c88130bc | 510 | INIT_RCU_HEAD(&ct->rcu); |
9fb9cbb1 | 511 | |
c88130bc | 512 | return ct; |
9fb9cbb1 | 513 | } |
13b18339 | 514 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 515 | |
76507f69 | 516 | static void nf_conntrack_free_rcu(struct rcu_head *head) |
9fb9cbb1 | 517 | { |
76507f69 | 518 | struct nf_conn *ct = container_of(head, struct nf_conn, rcu); |
49ac8713 | 519 | struct net *net = nf_ct_net(ct); |
76507f69 PM |
520 | |
521 | nf_ct_ext_free(ct); | |
522 | kmem_cache_free(nf_conntrack_cachep, ct); | |
49ac8713 | 523 | atomic_dec(&net->ct.count); |
9fb9cbb1 | 524 | } |
76507f69 | 525 | |
c88130bc | 526 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 527 | { |
ceeff754 | 528 | nf_ct_ext_destroy(ct); |
c88130bc | 529 | call_rcu(&ct->rcu, nf_conntrack_free_rcu); |
76507f69 | 530 | } |
13b18339 | 531 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
532 | |
533 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
534 | failed due to stress. Otherwise it really is unclassifiable. */ | |
535 | static struct nf_conntrack_tuple_hash * | |
5a1fb391 AD |
536 | init_conntrack(struct net *net, |
537 | const struct nf_conntrack_tuple *tuple, | |
9fb9cbb1 | 538 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 539 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
540 | struct sk_buff *skb, |
541 | unsigned int dataoff) | |
542 | { | |
c88130bc | 543 | struct nf_conn *ct; |
3c158f7f | 544 | struct nf_conn_help *help; |
9fb9cbb1 YK |
545 | struct nf_conntrack_tuple repl_tuple; |
546 | struct nf_conntrack_expect *exp; | |
547 | ||
605dcad6 | 548 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 549 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
550 | return NULL; |
551 | } | |
552 | ||
5a1fb391 | 553 | ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC); |
cd7fcbf1 | 554 | if (IS_ERR(ct)) { |
0d53778e | 555 | pr_debug("Can't allocate conntrack.\n"); |
c88130bc | 556 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 YK |
557 | } |
558 | ||
c88130bc PM |
559 | if (!l4proto->new(ct, skb, dataoff)) { |
560 | nf_conntrack_free(ct); | |
0d53778e | 561 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
562 | return NULL; |
563 | } | |
564 | ||
58401572 KPO |
565 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
566 | ||
f8ba1aff | 567 | spin_lock_bh(&nf_conntrack_lock); |
9b03f38d | 568 | exp = nf_ct_find_expectation(net, tuple); |
9fb9cbb1 | 569 | if (exp) { |
0d53778e | 570 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 571 | ct, exp); |
9fb9cbb1 | 572 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
573 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
574 | ct->master = exp->master; | |
ceceae1b | 575 | if (exp->helper) { |
c88130bc | 576 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b YK |
577 | if (help) |
578 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
579 | } |
580 | ||
9fb9cbb1 | 581 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 582 | ct->mark = exp->master->mark; |
7c9728c3 JM |
583 | #endif |
584 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 585 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 586 | #endif |
c88130bc | 587 | nf_conntrack_get(&ct->master->ct_general); |
0d55af87 | 588 | NF_CT_STAT_INC(net, expect_new); |
22e7410b | 589 | } else { |
226c0c0e | 590 | __nf_ct_try_assign_helper(ct, GFP_ATOMIC); |
0d55af87 | 591 | NF_CT_STAT_INC(net, new); |
22e7410b | 592 | } |
9fb9cbb1 YK |
593 | |
594 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
63c9a262 AD |
595 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
596 | &net->ct.unconfirmed); | |
9fb9cbb1 | 597 | |
f8ba1aff | 598 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
599 | |
600 | if (exp) { | |
601 | if (exp->expectfn) | |
c88130bc | 602 | exp->expectfn(ct, exp); |
6823645d | 603 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
604 | } |
605 | ||
c88130bc | 606 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
607 | } |
608 | ||
609 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
610 | static inline struct nf_conn * | |
a702a65f AD |
611 | resolve_normal_ct(struct net *net, |
612 | struct sk_buff *skb, | |
9fb9cbb1 YK |
613 | unsigned int dataoff, |
614 | u_int16_t l3num, | |
615 | u_int8_t protonum, | |
616 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 617 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
618 | int *set_reply, |
619 | enum ip_conntrack_info *ctinfo) | |
620 | { | |
621 | struct nf_conntrack_tuple tuple; | |
622 | struct nf_conntrack_tuple_hash *h; | |
623 | struct nf_conn *ct; | |
624 | ||
bbe735e4 | 625 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 626 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 627 | l4proto)) { |
0d53778e | 628 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
629 | return NULL; |
630 | } | |
631 | ||
632 | /* look for tuple match */ | |
a702a65f | 633 | h = nf_conntrack_find_get(net, &tuple); |
9fb9cbb1 | 634 | if (!h) { |
a702a65f | 635 | h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
636 | if (!h) |
637 | return NULL; | |
638 | if (IS_ERR(h)) | |
639 | return (void *)h; | |
640 | } | |
641 | ct = nf_ct_tuplehash_to_ctrack(h); | |
642 | ||
643 | /* It exists; we have (non-exclusive) reference. */ | |
644 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
645 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
646 | /* Please set reply bit if this packet OK */ | |
647 | *set_reply = 1; | |
648 | } else { | |
649 | /* Once we've had two way comms, always ESTABLISHED. */ | |
650 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 651 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
652 | *ctinfo = IP_CT_ESTABLISHED; |
653 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
654 | pr_debug("nf_conntrack_in: related packet for %p\n", |
655 | ct); | |
9fb9cbb1 YK |
656 | *ctinfo = IP_CT_RELATED; |
657 | } else { | |
0d53778e | 658 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
659 | *ctinfo = IP_CT_NEW; |
660 | } | |
661 | *set_reply = 0; | |
662 | } | |
663 | skb->nfct = &ct->ct_general; | |
664 | skb->nfctinfo = *ctinfo; | |
665 | return ct; | |
666 | } | |
667 | ||
668 | unsigned int | |
a702a65f AD |
669 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
670 | struct sk_buff *skb) | |
9fb9cbb1 YK |
671 | { |
672 | struct nf_conn *ct; | |
673 | enum ip_conntrack_info ctinfo; | |
674 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 675 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
676 | unsigned int dataoff; |
677 | u_int8_t protonum; | |
678 | int set_reply = 0; | |
679 | int ret; | |
680 | ||
681 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 682 | if (skb->nfct) { |
0d55af87 | 683 | NF_CT_STAT_INC_ATOMIC(net, ignore); |
9fb9cbb1 YK |
684 | return NF_ACCEPT; |
685 | } | |
686 | ||
923f4902 | 687 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 688 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 689 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
690 | &dataoff, &protonum); |
691 | if (ret <= 0) { | |
0d53778e | 692 | pr_debug("not prepared to track yet or error occured\n"); |
0d55af87 AD |
693 | NF_CT_STAT_INC_ATOMIC(net, error); |
694 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
9fb9cbb1 YK |
695 | return -ret; |
696 | } | |
697 | ||
76108cea | 698 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
699 | |
700 | /* It may be an special packet, error, unclean... | |
701 | * inverse of the return code tells to the netfilter | |
702 | * core what to do with the packet. */ | |
74c51a14 AD |
703 | if (l4proto->error != NULL) { |
704 | ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum); | |
705 | if (ret <= 0) { | |
0d55af87 AD |
706 | NF_CT_STAT_INC_ATOMIC(net, error); |
707 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
74c51a14 AD |
708 | return -ret; |
709 | } | |
9fb9cbb1 YK |
710 | } |
711 | ||
a702a65f AD |
712 | ct = resolve_normal_ct(net, skb, dataoff, pf, protonum, |
713 | l3proto, l4proto, &set_reply, &ctinfo); | |
9fb9cbb1 YK |
714 | if (!ct) { |
715 | /* Not valid part of a connection */ | |
0d55af87 | 716 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
9fb9cbb1 YK |
717 | return NF_ACCEPT; |
718 | } | |
719 | ||
720 | if (IS_ERR(ct)) { | |
721 | /* Too stressed to deal. */ | |
0d55af87 | 722 | NF_CT_STAT_INC_ATOMIC(net, drop); |
9fb9cbb1 YK |
723 | return NF_DROP; |
724 | } | |
725 | ||
3db05fea | 726 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 727 | |
3db05fea | 728 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
ec8d5409 | 729 | if (ret <= 0) { |
9fb9cbb1 YK |
730 | /* Invalid: inverse of the return code tells |
731 | * the netfilter core what to do */ | |
0d53778e | 732 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
733 | nf_conntrack_put(skb->nfct); |
734 | skb->nfct = NULL; | |
0d55af87 | 735 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
9fb9cbb1 YK |
736 | return -ret; |
737 | } | |
738 | ||
739 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
a71996fc | 740 | nf_conntrack_event_cache(IPCT_STATUS, ct); |
9fb9cbb1 YK |
741 | |
742 | return ret; | |
743 | } | |
13b18339 | 744 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 745 | |
5f2b4c90 JE |
746 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
747 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 748 | { |
5f2b4c90 | 749 | bool ret; |
923f4902 PM |
750 | |
751 | rcu_read_lock(); | |
752 | ret = nf_ct_invert_tuple(inverse, orig, | |
753 | __nf_ct_l3proto_find(orig->src.l3num), | |
754 | __nf_ct_l4proto_find(orig->src.l3num, | |
755 | orig->dst.protonum)); | |
756 | rcu_read_unlock(); | |
757 | return ret; | |
9fb9cbb1 | 758 | } |
13b18339 | 759 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 760 | |
5b1158e9 JK |
761 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
762 | implicitly racy: see __nf_conntrack_confirm */ | |
763 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
764 | const struct nf_conntrack_tuple *newreply) | |
765 | { | |
766 | struct nf_conn_help *help = nfct_help(ct); | |
767 | ||
5b1158e9 JK |
768 | /* Should be unconfirmed, so not in hash table yet */ |
769 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
770 | ||
0d53778e | 771 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 772 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
773 | |
774 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 775 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 776 | return; |
ceceae1b | 777 | |
c52fbb41 | 778 | rcu_read_lock(); |
226c0c0e | 779 | __nf_ct_try_assign_helper(ct, GFP_ATOMIC); |
c52fbb41 | 780 | rcu_read_unlock(); |
5b1158e9 | 781 | } |
13b18339 | 782 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 783 | |
9fb9cbb1 YK |
784 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
785 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
786 | enum ip_conntrack_info ctinfo, | |
787 | const struct sk_buff *skb, | |
788 | unsigned long extra_jiffies, | |
789 | int do_acct) | |
790 | { | |
791 | int event = 0; | |
792 | ||
793 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
794 | NF_CT_ASSERT(skb); | |
795 | ||
f8ba1aff | 796 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 797 | |
997ae831 | 798 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
799 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
800 | goto acct; | |
997ae831 | 801 | |
9fb9cbb1 YK |
802 | /* If not in hash table, timer will not be active yet */ |
803 | if (!nf_ct_is_confirmed(ct)) { | |
804 | ct->timeout.expires = extra_jiffies; | |
805 | event = IPCT_REFRESH; | |
806 | } else { | |
be00c8e4 MJ |
807 | unsigned long newtime = jiffies + extra_jiffies; |
808 | ||
809 | /* Only update the timeout if the new timeout is at least | |
810 | HZ jiffies from the old timeout. Need del_timer for race | |
811 | avoidance (may already be dying). */ | |
812 | if (newtime - ct->timeout.expires >= HZ | |
813 | && del_timer(&ct->timeout)) { | |
814 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
815 | add_timer(&ct->timeout); |
816 | event = IPCT_REFRESH; | |
817 | } | |
818 | } | |
819 | ||
47d95045 | 820 | acct: |
9fb9cbb1 | 821 | if (do_acct) { |
58401572 | 822 | struct nf_conn_counter *acct; |
3ffd5eeb | 823 | |
58401572 KPO |
824 | acct = nf_conn_acct_find(ct); |
825 | if (acct) { | |
826 | acct[CTINFO2DIR(ctinfo)].packets++; | |
827 | acct[CTINFO2DIR(ctinfo)].bytes += | |
828 | skb->len - skb_network_offset(skb); | |
829 | } | |
9fb9cbb1 | 830 | } |
9fb9cbb1 | 831 | |
f8ba1aff | 832 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
833 | |
834 | /* must be unlocked when calling event cache */ | |
835 | if (event) | |
a71996fc | 836 | nf_conntrack_event_cache(event, ct); |
9fb9cbb1 | 837 | } |
13b18339 | 838 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 839 | |
4c889498 DM |
840 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
841 | enum ip_conntrack_info ctinfo, | |
842 | const struct sk_buff *skb, | |
843 | int do_acct) | |
51091764 | 844 | { |
718d4ad9 | 845 | if (do_acct) { |
58401572 KPO |
846 | struct nf_conn_counter *acct; |
847 | ||
718d4ad9 | 848 | spin_lock_bh(&nf_conntrack_lock); |
58401572 KPO |
849 | acct = nf_conn_acct_find(ct); |
850 | if (acct) { | |
851 | acct[CTINFO2DIR(ctinfo)].packets++; | |
852 | acct[CTINFO2DIR(ctinfo)].bytes += | |
853 | skb->len - skb_network_offset(skb); | |
854 | } | |
718d4ad9 FH |
855 | spin_unlock_bh(&nf_conntrack_lock); |
856 | } | |
58401572 | 857 | |
4c889498 | 858 | if (del_timer(&ct->timeout)) { |
51091764 | 859 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
860 | return true; |
861 | } | |
862 | return false; | |
51091764 | 863 | } |
718d4ad9 | 864 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 865 | |
e281db5c | 866 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
867 | |
868 | #include <linux/netfilter/nfnetlink.h> | |
869 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
870 | #include <linux/mutex.h> |
871 | ||
c1d10adb PNA |
872 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
873 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
874 | * or depend on ctnetlink */ | |
fdf70832 | 875 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
876 | const struct nf_conntrack_tuple *tuple) |
877 | { | |
77236b6e PM |
878 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
879 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
880 | return 0; |
881 | ||
df6fb868 | 882 | nla_put_failure: |
c1d10adb PNA |
883 | return -1; |
884 | } | |
fdf70832 | 885 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 886 | |
f73e924c PM |
887 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
888 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
889 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 890 | }; |
f73e924c | 891 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 892 | |
fdf70832 | 893 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
894 | struct nf_conntrack_tuple *t) |
895 | { | |
df6fb868 | 896 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
897 | return -EINVAL; |
898 | ||
77236b6e PM |
899 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
900 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
901 | |
902 | return 0; | |
903 | } | |
fdf70832 | 904 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
c1d10adb PNA |
905 | #endif |
906 | ||
9fb9cbb1 | 907 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 908 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
909 | { |
910 | struct nf_conn *ct; | |
911 | enum ip_conntrack_info ctinfo; | |
912 | ||
913 | /* This ICMP is in reverse direction to the packet which caused it */ | |
914 | ct = nf_ct_get(skb, &ctinfo); | |
915 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
916 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
917 | else | |
918 | ctinfo = IP_CT_RELATED; | |
919 | ||
920 | /* Attach to new skbuff, and increment count */ | |
921 | nskb->nfct = &ct->ct_general; | |
922 | nskb->nfctinfo = ctinfo; | |
923 | nf_conntrack_get(nskb->nfct); | |
924 | } | |
925 | ||
9fb9cbb1 | 926 | /* Bring out ya dead! */ |
df0933dc | 927 | static struct nf_conn * |
400dad39 | 928 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
929 | void *data, unsigned int *bucket) |
930 | { | |
df0933dc PM |
931 | struct nf_conntrack_tuple_hash *h; |
932 | struct nf_conn *ct; | |
f205c5e0 | 933 | struct hlist_node *n; |
9fb9cbb1 | 934 | |
f8ba1aff | 935 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 936 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
400dad39 | 937 | hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) { |
df0933dc PM |
938 | ct = nf_ct_tuplehash_to_ctrack(h); |
939 | if (iter(ct, data)) | |
940 | goto found; | |
941 | } | |
601e68e1 | 942 | } |
63c9a262 | 943 | hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) { |
df0933dc PM |
944 | ct = nf_ct_tuplehash_to_ctrack(h); |
945 | if (iter(ct, data)) | |
ec68e97d | 946 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 947 | } |
f8ba1aff | 948 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
949 | return NULL; |
950 | found: | |
c073e3fa | 951 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 952 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 953 | return ct; |
9fb9cbb1 YK |
954 | } |
955 | ||
400dad39 AD |
956 | void nf_ct_iterate_cleanup(struct net *net, |
957 | int (*iter)(struct nf_conn *i, void *data), | |
958 | void *data) | |
9fb9cbb1 | 959 | { |
df0933dc | 960 | struct nf_conn *ct; |
9fb9cbb1 YK |
961 | unsigned int bucket = 0; |
962 | ||
400dad39 | 963 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
964 | /* Time to push up daises... */ |
965 | if (del_timer(&ct->timeout)) | |
966 | death_by_timeout((unsigned long)ct); | |
967 | /* ... else the timer will get him soon. */ | |
968 | ||
969 | nf_ct_put(ct); | |
970 | } | |
971 | } | |
13b18339 | 972 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 973 | |
19abb7b0 PNA |
974 | struct __nf_ct_flush_report { |
975 | u32 pid; | |
976 | int report; | |
977 | }; | |
978 | ||
9fb9cbb1 YK |
979 | static int kill_all(struct nf_conn *i, void *data) |
980 | { | |
19abb7b0 PNA |
981 | struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data; |
982 | ||
983 | /* get_next_corpse sets the dying bit for us */ | |
984 | nf_conntrack_event_report(IPCT_DESTROY, | |
985 | i, | |
986 | fr->pid, | |
987 | fr->report); | |
9fb9cbb1 YK |
988 | return 1; |
989 | } | |
990 | ||
96eb24d7 | 991 | void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
992 | { |
993 | if (vmalloced) | |
994 | vfree(hash); | |
995 | else | |
601e68e1 | 996 | free_pages((unsigned long)hash, |
f205c5e0 | 997 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 998 | } |
ac565e5f | 999 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1000 | |
19abb7b0 | 1001 | void nf_conntrack_flush(struct net *net, u32 pid, int report) |
c1d10adb | 1002 | { |
19abb7b0 PNA |
1003 | struct __nf_ct_flush_report fr = { |
1004 | .pid = pid, | |
1005 | .report = report, | |
1006 | }; | |
1007 | nf_ct_iterate_cleanup(net, kill_all, &fr); | |
c1d10adb | 1008 | } |
13b18339 | 1009 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 1010 | |
08f6547d | 1011 | static void nf_conntrack_cleanup_init_net(void) |
9fb9cbb1 | 1012 | { |
08f6547d AD |
1013 | nf_conntrack_helper_fini(); |
1014 | nf_conntrack_proto_fini(); | |
1015 | kmem_cache_destroy(nf_conntrack_cachep); | |
1016 | } | |
9fb9cbb1 | 1017 | |
08f6547d AD |
1018 | static void nf_conntrack_cleanup_net(struct net *net) |
1019 | { | |
6058fa6b AD |
1020 | nf_ct_event_cache_flush(net); |
1021 | nf_conntrack_ecache_fini(net); | |
9fb9cbb1 | 1022 | i_see_dead_people: |
19abb7b0 | 1023 | nf_conntrack_flush(net, 0, 0); |
49ac8713 | 1024 | if (atomic_read(&net->ct.count) != 0) { |
9fb9cbb1 YK |
1025 | schedule(); |
1026 | goto i_see_dead_people; | |
1027 | } | |
6636568c PM |
1028 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1029 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1030 | schedule(); | |
9fb9cbb1 | 1031 | |
400dad39 | 1032 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1033 | nf_conntrack_htable_size); |
d716a4df | 1034 | nf_conntrack_acct_fini(net); |
9b03f38d | 1035 | nf_conntrack_expect_fini(net); |
0d55af87 | 1036 | free_percpu(net->ct.stat); |
08f6547d AD |
1037 | } |
1038 | ||
1039 | /* Mishearing the voices in his head, our hero wonders how he's | |
1040 | supposed to kill the mall. */ | |
1041 | void nf_conntrack_cleanup(struct net *net) | |
1042 | { | |
1043 | if (net_eq(net, &init_net)) | |
1044 | rcu_assign_pointer(ip_ct_attach, NULL); | |
1045 | ||
1046 | /* This makes sure all current packets have passed through | |
1047 | netfilter framework. Roll on, two-stage module | |
1048 | delete... */ | |
1049 | synchronize_net(); | |
1050 | ||
1051 | nf_conntrack_cleanup_net(net); | |
1052 | ||
1053 | if (net_eq(net, &init_net)) { | |
1054 | rcu_assign_pointer(nf_ct_destroy, NULL); | |
1055 | nf_conntrack_cleanup_init_net(); | |
1056 | } | |
9fb9cbb1 YK |
1057 | } |
1058 | ||
96eb24d7 | 1059 | struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced) |
9fb9cbb1 | 1060 | { |
f205c5e0 | 1061 | struct hlist_head *hash; |
8e5105a0 | 1062 | unsigned int size, i; |
9fb9cbb1 | 1063 | |
601e68e1 | 1064 | *vmalloced = 0; |
8e5105a0 | 1065 | |
f205c5e0 | 1066 | size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); |
29b67497 | 1067 | hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN, |
f205c5e0 | 1068 | get_order(sizeof(struct hlist_head) |
9fb9cbb1 | 1069 | * size)); |
601e68e1 | 1070 | if (!hash) { |
9fb9cbb1 YK |
1071 | *vmalloced = 1; |
1072 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
f205c5e0 | 1073 | hash = vmalloc(sizeof(struct hlist_head) * size); |
9fb9cbb1 YK |
1074 | } |
1075 | ||
1076 | if (hash) | |
601e68e1 | 1077 | for (i = 0; i < size; i++) |
f205c5e0 | 1078 | INIT_HLIST_HEAD(&hash[i]); |
9fb9cbb1 YK |
1079 | |
1080 | return hash; | |
1081 | } | |
ac565e5f | 1082 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1083 | |
fae718dd | 1084 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1085 | { |
96eb24d7 SH |
1086 | int i, bucket, vmalloced, old_vmalloced; |
1087 | unsigned int hashsize, old_size; | |
9fb9cbb1 | 1088 | int rnd; |
f205c5e0 | 1089 | struct hlist_head *hash, *old_hash; |
9fb9cbb1 YK |
1090 | struct nf_conntrack_tuple_hash *h; |
1091 | ||
1092 | /* On boot, we can set this without any fancy locking. */ | |
1093 | if (!nf_conntrack_htable_size) | |
1094 | return param_set_uint(val, kp); | |
1095 | ||
96eb24d7 | 1096 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1097 | if (!hashsize) |
1098 | return -EINVAL; | |
1099 | ||
ac565e5f | 1100 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); |
9fb9cbb1 YK |
1101 | if (!hash) |
1102 | return -ENOMEM; | |
1103 | ||
1104 | /* We have to rehahs for the new table anyway, so we also can | |
1105 | * use a newrandom seed */ | |
1106 | get_random_bytes(&rnd, 4); | |
1107 | ||
76507f69 PM |
1108 | /* Lookups in the old hash might happen in parallel, which means we |
1109 | * might get false negatives during connection lookup. New connections | |
1110 | * created because of a false negative won't make it into the hash | |
1111 | * though since that required taking the lock. | |
1112 | */ | |
f8ba1aff | 1113 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1114 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
400dad39 AD |
1115 | while (!hlist_empty(&init_net.ct.hash[i])) { |
1116 | h = hlist_entry(init_net.ct.hash[i].first, | |
f205c5e0 | 1117 | struct nf_conntrack_tuple_hash, hnode); |
76507f69 | 1118 | hlist_del_rcu(&h->hnode); |
9fb9cbb1 | 1119 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
f205c5e0 | 1120 | hlist_add_head(&h->hnode, &hash[bucket]); |
9fb9cbb1 YK |
1121 | } |
1122 | } | |
1123 | old_size = nf_conntrack_htable_size; | |
400dad39 AD |
1124 | old_vmalloced = init_net.ct.hash_vmalloc; |
1125 | old_hash = init_net.ct.hash; | |
9fb9cbb1 YK |
1126 | |
1127 | nf_conntrack_htable_size = hashsize; | |
400dad39 AD |
1128 | init_net.ct.hash_vmalloc = vmalloced; |
1129 | init_net.ct.hash = hash; | |
9fb9cbb1 | 1130 | nf_conntrack_hash_rnd = rnd; |
f8ba1aff | 1131 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1132 | |
ac565e5f | 1133 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1134 | return 0; |
1135 | } | |
fae718dd | 1136 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1137 | |
fae718dd | 1138 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1139 | &nf_conntrack_htable_size, 0600); |
1140 | ||
08f6547d | 1141 | static int nf_conntrack_init_init_net(void) |
9fb9cbb1 | 1142 | { |
f205c5e0 | 1143 | int max_factor = 8; |
9fb9cbb1 YK |
1144 | int ret; |
1145 | ||
1146 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1147 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1148 | if (!nf_conntrack_htable_size) { |
1149 | nf_conntrack_htable_size | |
1150 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
f205c5e0 | 1151 | / sizeof(struct hlist_head)); |
9fb9cbb1 | 1152 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1153 | nf_conntrack_htable_size = 16384; |
1154 | if (nf_conntrack_htable_size < 32) | |
1155 | nf_conntrack_htable_size = 32; | |
1156 | ||
1157 | /* Use a max. factor of four by default to get the same max as | |
1158 | * with the old struct list_heads. When a table size is given | |
1159 | * we use the old value of 8 to avoid reducing the max. | |
1160 | * entries. */ | |
1161 | max_factor = 4; | |
9fb9cbb1 | 1162 | } |
f205c5e0 | 1163 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1164 | |
1165 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1166 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1167 | nf_conntrack_max); | |
1168 | ||
dacd2a1a YK |
1169 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1170 | sizeof(struct nf_conn), | |
20c2df83 | 1171 | 0, 0, NULL); |
dacd2a1a | 1172 | if (!nf_conntrack_cachep) { |
9fb9cbb1 | 1173 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
08f6547d AD |
1174 | ret = -ENOMEM; |
1175 | goto err_cache; | |
9fb9cbb1 YK |
1176 | } |
1177 | ||
e9c1b084 PM |
1178 | ret = nf_conntrack_proto_init(); |
1179 | if (ret < 0) | |
08f6547d | 1180 | goto err_proto; |
933a41e7 | 1181 | |
ceceae1b YK |
1182 | ret = nf_conntrack_helper_init(); |
1183 | if (ret < 0) | |
08f6547d AD |
1184 | goto err_helper; |
1185 | ||
1186 | return 0; | |
1187 | ||
1188 | err_helper: | |
1189 | nf_conntrack_proto_fini(); | |
1190 | err_proto: | |
1191 | kmem_cache_destroy(nf_conntrack_cachep); | |
1192 | err_cache: | |
1193 | return ret; | |
1194 | } | |
1195 | ||
1196 | static int nf_conntrack_init_net(struct net *net) | |
1197 | { | |
1198 | int ret; | |
ceceae1b | 1199 | |
08f6547d AD |
1200 | atomic_set(&net->ct.count, 0); |
1201 | INIT_HLIST_HEAD(&net->ct.unconfirmed); | |
1202 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); | |
1203 | if (!net->ct.stat) { | |
1204 | ret = -ENOMEM; | |
1205 | goto err_stat; | |
1206 | } | |
1207 | ret = nf_conntrack_ecache_init(net); | |
1208 | if (ret < 0) | |
1209 | goto err_ecache; | |
1210 | net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, | |
1211 | &net->ct.hash_vmalloc); | |
1212 | if (!net->ct.hash) { | |
1213 | ret = -ENOMEM; | |
1214 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1215 | goto err_hash; | |
1216 | } | |
1217 | ret = nf_conntrack_expect_init(net); | |
1218 | if (ret < 0) | |
1219 | goto err_expect; | |
d716a4df | 1220 | ret = nf_conntrack_acct_init(net); |
58401572 | 1221 | if (ret < 0) |
08f6547d | 1222 | goto err_acct; |
7d3cdc6b | 1223 | |
9fb9cbb1 YK |
1224 | /* Set up fake conntrack: |
1225 | - to never be deleted, not in any hashes */ | |
5a1fb391 AD |
1226 | #ifdef CONFIG_NET_NS |
1227 | nf_conntrack_untracked.ct_net = &init_net; | |
1228 | #endif | |
9fb9cbb1 YK |
1229 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); |
1230 | /* - and look it like as a confirmed connection */ | |
1231 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1232 | ||
08f6547d | 1233 | return 0; |
9fb9cbb1 | 1234 | |
08f6547d | 1235 | err_acct: |
9b03f38d | 1236 | nf_conntrack_expect_fini(net); |
08f6547d | 1237 | err_expect: |
400dad39 | 1238 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1239 | nf_conntrack_htable_size); |
6058fa6b AD |
1240 | err_hash: |
1241 | nf_conntrack_ecache_fini(net); | |
1242 | err_ecache: | |
0d55af87 AD |
1243 | free_percpu(net->ct.stat); |
1244 | err_stat: | |
08f6547d AD |
1245 | return ret; |
1246 | } | |
1247 | ||
1248 | int nf_conntrack_init(struct net *net) | |
1249 | { | |
1250 | int ret; | |
1251 | ||
1252 | if (net_eq(net, &init_net)) { | |
1253 | ret = nf_conntrack_init_init_net(); | |
1254 | if (ret < 0) | |
1255 | goto out_init_net; | |
1256 | } | |
1257 | ret = nf_conntrack_init_net(net); | |
1258 | if (ret < 0) | |
1259 | goto out_net; | |
1260 | ||
1261 | if (net_eq(net, &init_net)) { | |
1262 | /* For use by REJECT target */ | |
1263 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); | |
1264 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); | |
1265 | } | |
1266 | return 0; | |
1267 | ||
1268 | out_net: | |
1269 | if (net_eq(net, &init_net)) | |
1270 | nf_conntrack_cleanup_init_net(); | |
1271 | out_init_net: | |
1272 | return ret; | |
9fb9cbb1 | 1273 | } |