]>
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 | ||
dacd2a1a | 56 | static struct kmem_cache *nf_conntrack_cachep __read_mostly; |
77ab9cff | 57 | |
9fb9cbb1 YK |
58 | static int nf_conntrack_hash_rnd_initted; |
59 | static unsigned int nf_conntrack_hash_rnd; | |
60 | ||
61 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
62 | unsigned int size, unsigned int rnd) | |
63 | { | |
0794935e PM |
64 | unsigned int n; |
65 | u_int32_t h; | |
66 | ||
67 | /* The direction must be ignored, so we hash everything up to the | |
68 | * destination ports (which is a multiple of 4) and treat the last | |
69 | * three bytes manually. | |
70 | */ | |
71 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
72 | h = jhash2((u32 *)tuple, n, | |
73 | rnd ^ (((__force __u16)tuple->dst.u.all << 16) | | |
74 | tuple->dst.protonum)); | |
75 | ||
76 | return ((u64)h * size) >> 32; | |
9fb9cbb1 YK |
77 | } |
78 | ||
79 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
80 | { | |
81 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
82 | nf_conntrack_hash_rnd); | |
83 | } | |
84 | ||
5f2b4c90 | 85 | bool |
9fb9cbb1 YK |
86 | nf_ct_get_tuple(const struct sk_buff *skb, |
87 | unsigned int nhoff, | |
88 | unsigned int dataoff, | |
89 | u_int16_t l3num, | |
90 | u_int8_t protonum, | |
91 | struct nf_conntrack_tuple *tuple, | |
92 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 93 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 94 | { |
443a70d5 | 95 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
96 | |
97 | tuple->src.l3num = l3num; | |
98 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 99 | return false; |
9fb9cbb1 YK |
100 | |
101 | tuple->dst.protonum = protonum; | |
102 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
103 | ||
605dcad6 | 104 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 105 | } |
13b18339 | 106 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 107 | |
5f2b4c90 JE |
108 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
109 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
110 | { |
111 | struct nf_conntrack_l3proto *l3proto; | |
112 | struct nf_conntrack_l4proto *l4proto; | |
113 | unsigned int protoff; | |
114 | u_int8_t protonum; | |
115 | int ret; | |
116 | ||
117 | rcu_read_lock(); | |
118 | ||
119 | l3proto = __nf_ct_l3proto_find(l3num); | |
120 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
121 | if (ret != NF_ACCEPT) { | |
122 | rcu_read_unlock(); | |
5f2b4c90 | 123 | return false; |
e2a3123f YK |
124 | } |
125 | ||
126 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
127 | ||
128 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
129 | l3proto, l4proto); | |
130 | ||
131 | rcu_read_unlock(); | |
132 | return ret; | |
133 | } | |
134 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
135 | ||
5f2b4c90 | 136 | bool |
9fb9cbb1 YK |
137 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
138 | const struct nf_conntrack_tuple *orig, | |
139 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 140 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 141 | { |
443a70d5 | 142 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
143 | |
144 | inverse->src.l3num = orig->src.l3num; | |
145 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 146 | return false; |
9fb9cbb1 YK |
147 | |
148 | inverse->dst.dir = !orig->dst.dir; | |
149 | ||
150 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 151 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 152 | } |
13b18339 | 153 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 154 | |
9fb9cbb1 YK |
155 | static void |
156 | clean_from_lists(struct nf_conn *ct) | |
157 | { | |
0d53778e | 158 | pr_debug("clean_from_lists(%p)\n", ct); |
76507f69 PM |
159 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
160 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); | |
9fb9cbb1 YK |
161 | |
162 | /* Destroy all pending expectations */ | |
c1d10adb | 163 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
164 | } |
165 | ||
166 | static void | |
167 | destroy_conntrack(struct nf_conntrack *nfct) | |
168 | { | |
169 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
0d55af87 | 170 | struct net *net = nf_ct_net(ct); |
605dcad6 | 171 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 172 | |
0d53778e | 173 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
174 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
175 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
176 | ||
177 | nf_conntrack_event(IPCT_DESTROY, ct); | |
178 | set_bit(IPS_DYING_BIT, &ct->status); | |
179 | ||
180 | /* To make sure we don't get any weird locking issues here: | |
181 | * destroy_conntrack() MUST NOT be called with a write lock | |
182 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 183 | rcu_read_lock(); |
5e8fbe2a | 184 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
185 | if (l4proto && l4proto->destroy) |
186 | l4proto->destroy(ct); | |
9fb9cbb1 | 187 | |
982d9a9c | 188 | rcu_read_unlock(); |
9fb9cbb1 | 189 | |
f8ba1aff | 190 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
191 | /* Expectations will have been removed in clean_from_lists, |
192 | * except TFTP can create an expectation on the first packet, | |
193 | * before connection is in the list, so we need to clean here, | |
194 | * too. */ | |
c1d10adb | 195 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
196 | |
197 | /* We overload first tuple to link into unconfirmed list. */ | |
198 | if (!nf_ct_is_confirmed(ct)) { | |
f205c5e0 PM |
199 | BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode)); |
200 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); | |
9fb9cbb1 YK |
201 | } |
202 | ||
0d55af87 | 203 | NF_CT_STAT_INC(net, delete); |
f8ba1aff | 204 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
205 | |
206 | if (ct->master) | |
207 | nf_ct_put(ct->master); | |
208 | ||
0d53778e | 209 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
210 | nf_conntrack_free(ct); |
211 | } | |
212 | ||
213 | static void death_by_timeout(unsigned long ul_conntrack) | |
214 | { | |
215 | struct nf_conn *ct = (void *)ul_conntrack; | |
0d55af87 | 216 | struct net *net = nf_ct_net(ct); |
5397e97d | 217 | struct nf_conn_help *help = nfct_help(ct); |
3c158f7f | 218 | struct nf_conntrack_helper *helper; |
5397e97d | 219 | |
3c158f7f PM |
220 | if (help) { |
221 | rcu_read_lock(); | |
222 | helper = rcu_dereference(help->helper); | |
223 | if (helper && helper->destroy) | |
224 | helper->destroy(ct); | |
225 | rcu_read_unlock(); | |
226 | } | |
9fb9cbb1 | 227 | |
f8ba1aff | 228 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
229 | /* Inside lock so preempt is disabled on module removal path. |
230 | * Otherwise we can get spurious warnings. */ | |
0d55af87 | 231 | NF_CT_STAT_INC(net, delete_list); |
9fb9cbb1 | 232 | clean_from_lists(ct); |
f8ba1aff | 233 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
234 | nf_ct_put(ct); |
235 | } | |
236 | ||
c1d10adb | 237 | struct nf_conntrack_tuple_hash * |
400dad39 | 238 | __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
239 | { |
240 | struct nf_conntrack_tuple_hash *h; | |
f205c5e0 | 241 | struct hlist_node *n; |
9fb9cbb1 YK |
242 | unsigned int hash = hash_conntrack(tuple); |
243 | ||
4e29e9ec PM |
244 | /* Disable BHs the entire time since we normally need to disable them |
245 | * at least once for the stats anyway. | |
246 | */ | |
247 | local_bh_disable(); | |
400dad39 | 248 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) { |
ba419aff | 249 | if (nf_ct_tuple_equal(tuple, &h->tuple)) { |
0d55af87 | 250 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 251 | local_bh_enable(); |
9fb9cbb1 YK |
252 | return h; |
253 | } | |
0d55af87 | 254 | NF_CT_STAT_INC(net, searched); |
9fb9cbb1 | 255 | } |
4e29e9ec | 256 | local_bh_enable(); |
9fb9cbb1 YK |
257 | |
258 | return NULL; | |
259 | } | |
13b18339 | 260 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
261 | |
262 | /* Find a connection corresponding to a tuple. */ | |
263 | struct nf_conntrack_tuple_hash * | |
400dad39 | 264 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
265 | { |
266 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 267 | struct nf_conn *ct; |
9fb9cbb1 | 268 | |
76507f69 | 269 | rcu_read_lock(); |
400dad39 | 270 | h = __nf_conntrack_find(net, tuple); |
76507f69 PM |
271 | if (h) { |
272 | ct = nf_ct_tuplehash_to_ctrack(h); | |
273 | if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
274 | h = NULL; | |
275 | } | |
276 | rcu_read_unlock(); | |
9fb9cbb1 YK |
277 | |
278 | return h; | |
279 | } | |
13b18339 | 280 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 281 | |
c1d10adb PNA |
282 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
283 | unsigned int hash, | |
601e68e1 | 284 | unsigned int repl_hash) |
c1d10adb | 285 | { |
400dad39 AD |
286 | struct net *net = nf_ct_net(ct); |
287 | ||
76507f69 | 288 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
400dad39 | 289 | &net->ct.hash[hash]); |
76507f69 | 290 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, |
400dad39 | 291 | &net->ct.hash[repl_hash]); |
c1d10adb PNA |
292 | } |
293 | ||
294 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
295 | { | |
296 | unsigned int hash, repl_hash; | |
297 | ||
298 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
299 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
300 | ||
f8ba1aff | 301 | spin_lock_bh(&nf_conntrack_lock); |
c1d10adb | 302 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
f8ba1aff | 303 | spin_unlock_bh(&nf_conntrack_lock); |
c1d10adb | 304 | } |
13b18339 | 305 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 306 | |
9fb9cbb1 YK |
307 | /* Confirm a connection given skb; places it in hash table */ |
308 | int | |
3db05fea | 309 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
310 | { |
311 | unsigned int hash, repl_hash; | |
df0933dc | 312 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 313 | struct nf_conn *ct; |
df0933dc | 314 | struct nf_conn_help *help; |
f205c5e0 | 315 | struct hlist_node *n; |
9fb9cbb1 | 316 | enum ip_conntrack_info ctinfo; |
400dad39 | 317 | struct net *net; |
9fb9cbb1 | 318 | |
3db05fea | 319 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 320 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
321 | |
322 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
323 | ICMP/TCP RST packets in other direction. Actual packet | |
324 | which created connection will be IP_CT_NEW or for an | |
325 | expected connection, IP_CT_RELATED. */ | |
326 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
327 | return NF_ACCEPT; | |
328 | ||
329 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
330 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
331 | ||
332 | /* We're not in hash table, and we refuse to set up related | |
333 | connections for unconfirmed conns. But packet copies and | |
334 | REJECT will give spurious warnings here. */ | |
335 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
336 | ||
337 | /* No external references means noone else could have | |
338 | confirmed us. */ | |
339 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 340 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 341 | |
f8ba1aff | 342 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
343 | |
344 | /* See if there's one in the list already, including reverse: | |
345 | NAT could have grabbed it without realizing, since we're | |
346 | not in the hash. If there is, we lost race. */ | |
400dad39 | 347 | hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode) |
df0933dc PM |
348 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
349 | &h->tuple)) | |
350 | goto out; | |
400dad39 | 351 | hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode) |
df0933dc PM |
352 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
353 | &h->tuple)) | |
354 | goto out; | |
9fb9cbb1 | 355 | |
df0933dc | 356 | /* Remove from unconfirmed list */ |
f205c5e0 | 357 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
df0933dc PM |
358 | |
359 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
360 | /* Timer relative to confirmation time, not original | |
361 | setting time, otherwise we'd get timer wrap in | |
362 | weird delay cases. */ | |
363 | ct->timeout.expires += jiffies; | |
364 | add_timer(&ct->timeout); | |
365 | atomic_inc(&ct->ct_general.use); | |
366 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
0d55af87 | 367 | NF_CT_STAT_INC(net, insert); |
f8ba1aff | 368 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
369 | help = nfct_help(ct); |
370 | if (help && help->helper) | |
a71996fc | 371 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
9fb9cbb1 | 372 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
373 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
374 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
a71996fc | 375 | nf_conntrack_event_cache(IPCT_NATINFO, ct); |
9fb9cbb1 | 376 | #endif |
df0933dc | 377 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 378 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 379 | return NF_ACCEPT; |
9fb9cbb1 | 380 | |
df0933dc | 381 | out: |
0d55af87 | 382 | NF_CT_STAT_INC(net, insert_failed); |
f8ba1aff | 383 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
384 | return NF_DROP; |
385 | } | |
13b18339 | 386 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
387 | |
388 | /* Returns true if a connection correspondings to the tuple (required | |
389 | for NAT). */ | |
390 | int | |
391 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
392 | const struct nf_conn *ignored_conntrack) | |
393 | { | |
400dad39 | 394 | struct net *net = nf_ct_net(ignored_conntrack); |
9fb9cbb1 | 395 | struct nf_conntrack_tuple_hash *h; |
ba419aff PM |
396 | struct hlist_node *n; |
397 | unsigned int hash = hash_conntrack(tuple); | |
9fb9cbb1 | 398 | |
4e29e9ec PM |
399 | /* Disable BHs the entire time since we need to disable them at |
400 | * least once for the stats anyway. | |
401 | */ | |
402 | rcu_read_lock_bh(); | |
400dad39 | 403 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) { |
ba419aff PM |
404 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
405 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
0d55af87 | 406 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 407 | rcu_read_unlock_bh(); |
ba419aff PM |
408 | return 1; |
409 | } | |
0d55af87 | 410 | NF_CT_STAT_INC(net, searched); |
ba419aff | 411 | } |
4e29e9ec | 412 | rcu_read_unlock_bh(); |
9fb9cbb1 | 413 | |
ba419aff | 414 | return 0; |
9fb9cbb1 | 415 | } |
13b18339 | 416 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 417 | |
7ae7730f PM |
418 | #define NF_CT_EVICTION_RANGE 8 |
419 | ||
9fb9cbb1 YK |
420 | /* There's a small race here where we may free a just-assured |
421 | connection. Too bad: we're in trouble anyway. */ | |
400dad39 | 422 | static noinline int early_drop(struct net *net, unsigned int hash) |
9fb9cbb1 | 423 | { |
f205c5e0 | 424 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 425 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 426 | struct nf_conn *ct = NULL, *tmp; |
f205c5e0 | 427 | struct hlist_node *n; |
7ae7730f | 428 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
429 | int dropped = 0; |
430 | ||
76507f69 | 431 | rcu_read_lock(); |
7ae7730f | 432 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
400dad39 | 433 | hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], |
76507f69 | 434 | hnode) { |
7ae7730f PM |
435 | tmp = nf_ct_tuplehash_to_ctrack(h); |
436 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
437 | ct = tmp; | |
438 | cnt++; | |
439 | } | |
76507f69 PM |
440 | |
441 | if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
442 | ct = NULL; | |
7ae7730f PM |
443 | if (ct || cnt >= NF_CT_EVICTION_RANGE) |
444 | break; | |
445 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 446 | } |
76507f69 | 447 | rcu_read_unlock(); |
9fb9cbb1 YK |
448 | |
449 | if (!ct) | |
450 | return dropped; | |
451 | ||
452 | if (del_timer(&ct->timeout)) { | |
453 | death_by_timeout((unsigned long)ct); | |
454 | dropped = 1; | |
0d55af87 | 455 | NF_CT_STAT_INC_ATOMIC(net, early_drop); |
9fb9cbb1 YK |
456 | } |
457 | nf_ct_put(ct); | |
458 | return dropped; | |
459 | } | |
460 | ||
5a1fb391 AD |
461 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
462 | const struct nf_conntrack_tuple *orig, | |
b891c5a8 PNA |
463 | const struct nf_conntrack_tuple *repl, |
464 | gfp_t gfp) | |
9fb9cbb1 | 465 | { |
c88130bc | 466 | struct nf_conn *ct = NULL; |
9fb9cbb1 | 467 | |
dc808fe2 | 468 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
469 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
470 | nf_conntrack_hash_rnd_initted = 1; | |
471 | } | |
472 | ||
5251e2d2 | 473 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 474 | atomic_inc(&net->ct.count); |
5251e2d2 | 475 | |
76eb9460 | 476 | if (nf_conntrack_max && |
49ac8713 | 477 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
9fb9cbb1 | 478 | unsigned int hash = hash_conntrack(orig); |
400dad39 | 479 | if (!early_drop(net, hash)) { |
49ac8713 | 480 | atomic_dec(&net->ct.count); |
9fb9cbb1 YK |
481 | if (net_ratelimit()) |
482 | printk(KERN_WARNING | |
483 | "nf_conntrack: table full, dropping" | |
484 | " packet.\n"); | |
485 | return ERR_PTR(-ENOMEM); | |
486 | } | |
487 | } | |
488 | ||
b891c5a8 | 489 | ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp); |
c88130bc | 490 | if (ct == NULL) { |
0d53778e | 491 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
49ac8713 | 492 | atomic_dec(&net->ct.count); |
dacd2a1a | 493 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 YK |
494 | } |
495 | ||
c88130bc PM |
496 | atomic_set(&ct->ct_general.use, 1); |
497 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; | |
498 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
9fb9cbb1 | 499 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 500 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
5a1fb391 AD |
501 | #ifdef CONFIG_NET_NS |
502 | ct->ct_net = net; | |
503 | #endif | |
c88130bc | 504 | INIT_RCU_HEAD(&ct->rcu); |
9fb9cbb1 | 505 | |
c88130bc | 506 | return ct; |
9fb9cbb1 | 507 | } |
13b18339 | 508 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 509 | |
76507f69 | 510 | static void nf_conntrack_free_rcu(struct rcu_head *head) |
9fb9cbb1 | 511 | { |
76507f69 | 512 | struct nf_conn *ct = container_of(head, struct nf_conn, rcu); |
49ac8713 | 513 | struct net *net = nf_ct_net(ct); |
76507f69 PM |
514 | |
515 | nf_ct_ext_free(ct); | |
516 | kmem_cache_free(nf_conntrack_cachep, ct); | |
49ac8713 | 517 | atomic_dec(&net->ct.count); |
9fb9cbb1 | 518 | } |
76507f69 | 519 | |
c88130bc | 520 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 521 | { |
ceeff754 | 522 | nf_ct_ext_destroy(ct); |
c88130bc | 523 | call_rcu(&ct->rcu, nf_conntrack_free_rcu); |
76507f69 | 524 | } |
13b18339 | 525 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
526 | |
527 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
528 | failed due to stress. Otherwise it really is unclassifiable. */ | |
529 | static struct nf_conntrack_tuple_hash * | |
5a1fb391 AD |
530 | init_conntrack(struct net *net, |
531 | const struct nf_conntrack_tuple *tuple, | |
9fb9cbb1 | 532 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 533 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
534 | struct sk_buff *skb, |
535 | unsigned int dataoff) | |
536 | { | |
c88130bc | 537 | struct nf_conn *ct; |
3c158f7f | 538 | struct nf_conn_help *help; |
9fb9cbb1 YK |
539 | struct nf_conntrack_tuple repl_tuple; |
540 | struct nf_conntrack_expect *exp; | |
541 | ||
605dcad6 | 542 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 543 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
544 | return NULL; |
545 | } | |
546 | ||
5a1fb391 | 547 | ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC); |
c88130bc | 548 | if (ct == NULL || IS_ERR(ct)) { |
0d53778e | 549 | pr_debug("Can't allocate conntrack.\n"); |
c88130bc | 550 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 YK |
551 | } |
552 | ||
c88130bc PM |
553 | if (!l4proto->new(ct, skb, dataoff)) { |
554 | nf_conntrack_free(ct); | |
0d53778e | 555 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
556 | return NULL; |
557 | } | |
558 | ||
58401572 KPO |
559 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
560 | ||
f8ba1aff | 561 | spin_lock_bh(&nf_conntrack_lock); |
9b03f38d | 562 | exp = nf_ct_find_expectation(net, tuple); |
9fb9cbb1 | 563 | if (exp) { |
0d53778e | 564 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 565 | ct, exp); |
9fb9cbb1 | 566 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
567 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
568 | ct->master = exp->master; | |
ceceae1b | 569 | if (exp->helper) { |
c88130bc | 570 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b YK |
571 | if (help) |
572 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
573 | } |
574 | ||
9fb9cbb1 | 575 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 576 | ct->mark = exp->master->mark; |
7c9728c3 JM |
577 | #endif |
578 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 579 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 580 | #endif |
c88130bc | 581 | nf_conntrack_get(&ct->master->ct_general); |
0d55af87 | 582 | NF_CT_STAT_INC(net, expect_new); |
22e7410b | 583 | } else { |
ceceae1b YK |
584 | struct nf_conntrack_helper *helper; |
585 | ||
586 | helper = __nf_ct_helper_find(&repl_tuple); | |
587 | if (helper) { | |
c88130bc | 588 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b | 589 | if (help) |
ceceae1b | 590 | rcu_assign_pointer(help->helper, helper); |
3c158f7f | 591 | } |
0d55af87 | 592 | NF_CT_STAT_INC(net, new); |
22e7410b | 593 | } |
9fb9cbb1 YK |
594 | |
595 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
63c9a262 AD |
596 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
597 | &net->ct.unconfirmed); | |
9fb9cbb1 | 598 | |
f8ba1aff | 599 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
600 | |
601 | if (exp) { | |
602 | if (exp->expectfn) | |
c88130bc | 603 | exp->expectfn(ct, exp); |
6823645d | 604 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
605 | } |
606 | ||
c88130bc | 607 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
608 | } |
609 | ||
610 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
611 | static inline struct nf_conn * | |
a702a65f AD |
612 | resolve_normal_ct(struct net *net, |
613 | struct sk_buff *skb, | |
9fb9cbb1 YK |
614 | unsigned int dataoff, |
615 | u_int16_t l3num, | |
616 | u_int8_t protonum, | |
617 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 618 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
619 | int *set_reply, |
620 | enum ip_conntrack_info *ctinfo) | |
621 | { | |
622 | struct nf_conntrack_tuple tuple; | |
623 | struct nf_conntrack_tuple_hash *h; | |
624 | struct nf_conn *ct; | |
625 | ||
bbe735e4 | 626 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 627 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 628 | l4proto)) { |
0d53778e | 629 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
630 | return NULL; |
631 | } | |
632 | ||
633 | /* look for tuple match */ | |
a702a65f | 634 | h = nf_conntrack_find_get(net, &tuple); |
9fb9cbb1 | 635 | if (!h) { |
a702a65f | 636 | h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
637 | if (!h) |
638 | return NULL; | |
639 | if (IS_ERR(h)) | |
640 | return (void *)h; | |
641 | } | |
642 | ct = nf_ct_tuplehash_to_ctrack(h); | |
643 | ||
644 | /* It exists; we have (non-exclusive) reference. */ | |
645 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
646 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
647 | /* Please set reply bit if this packet OK */ | |
648 | *set_reply = 1; | |
649 | } else { | |
650 | /* Once we've had two way comms, always ESTABLISHED. */ | |
651 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 652 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
653 | *ctinfo = IP_CT_ESTABLISHED; |
654 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
655 | pr_debug("nf_conntrack_in: related packet for %p\n", |
656 | ct); | |
9fb9cbb1 YK |
657 | *ctinfo = IP_CT_RELATED; |
658 | } else { | |
0d53778e | 659 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
660 | *ctinfo = IP_CT_NEW; |
661 | } | |
662 | *set_reply = 0; | |
663 | } | |
664 | skb->nfct = &ct->ct_general; | |
665 | skb->nfctinfo = *ctinfo; | |
666 | return ct; | |
667 | } | |
668 | ||
669 | unsigned int | |
a702a65f AD |
670 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
671 | struct sk_buff *skb) | |
9fb9cbb1 YK |
672 | { |
673 | struct nf_conn *ct; | |
674 | enum ip_conntrack_info ctinfo; | |
675 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 676 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
677 | unsigned int dataoff; |
678 | u_int8_t protonum; | |
679 | int set_reply = 0; | |
680 | int ret; | |
681 | ||
682 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 683 | if (skb->nfct) { |
0d55af87 | 684 | NF_CT_STAT_INC_ATOMIC(net, ignore); |
9fb9cbb1 YK |
685 | return NF_ACCEPT; |
686 | } | |
687 | ||
923f4902 | 688 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 689 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 690 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
691 | &dataoff, &protonum); |
692 | if (ret <= 0) { | |
0d53778e | 693 | pr_debug("not prepared to track yet or error occured\n"); |
0d55af87 AD |
694 | NF_CT_STAT_INC_ATOMIC(net, error); |
695 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
9fb9cbb1 YK |
696 | return -ret; |
697 | } | |
698 | ||
76108cea | 699 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
700 | |
701 | /* It may be an special packet, error, unclean... | |
702 | * inverse of the return code tells to the netfilter | |
703 | * core what to do with the packet. */ | |
74c51a14 AD |
704 | if (l4proto->error != NULL) { |
705 | ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum); | |
706 | if (ret <= 0) { | |
0d55af87 AD |
707 | NF_CT_STAT_INC_ATOMIC(net, error); |
708 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
74c51a14 AD |
709 | return -ret; |
710 | } | |
9fb9cbb1 YK |
711 | } |
712 | ||
a702a65f AD |
713 | ct = resolve_normal_ct(net, skb, dataoff, pf, protonum, |
714 | l3proto, l4proto, &set_reply, &ctinfo); | |
9fb9cbb1 YK |
715 | if (!ct) { |
716 | /* Not valid part of a connection */ | |
0d55af87 | 717 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
9fb9cbb1 YK |
718 | return NF_ACCEPT; |
719 | } | |
720 | ||
721 | if (IS_ERR(ct)) { | |
722 | /* Too stressed to deal. */ | |
0d55af87 | 723 | NF_CT_STAT_INC_ATOMIC(net, drop); |
9fb9cbb1 YK |
724 | return NF_DROP; |
725 | } | |
726 | ||
3db05fea | 727 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 728 | |
3db05fea | 729 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
730 | if (ret < 0) { |
731 | /* Invalid: inverse of the return code tells | |
732 | * the netfilter core what to do */ | |
0d53778e | 733 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
734 | nf_conntrack_put(skb->nfct); |
735 | skb->nfct = NULL; | |
0d55af87 | 736 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
9fb9cbb1 YK |
737 | return -ret; |
738 | } | |
739 | ||
740 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
a71996fc | 741 | nf_conntrack_event_cache(IPCT_STATUS, ct); |
9fb9cbb1 YK |
742 | |
743 | return ret; | |
744 | } | |
13b18339 | 745 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 746 | |
5f2b4c90 JE |
747 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
748 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 749 | { |
5f2b4c90 | 750 | bool ret; |
923f4902 PM |
751 | |
752 | rcu_read_lock(); | |
753 | ret = nf_ct_invert_tuple(inverse, orig, | |
754 | __nf_ct_l3proto_find(orig->src.l3num), | |
755 | __nf_ct_l4proto_find(orig->src.l3num, | |
756 | orig->dst.protonum)); | |
757 | rcu_read_unlock(); | |
758 | return ret; | |
9fb9cbb1 | 759 | } |
13b18339 | 760 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 761 | |
5b1158e9 JK |
762 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
763 | implicitly racy: see __nf_conntrack_confirm */ | |
764 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
765 | const struct nf_conntrack_tuple *newreply) | |
766 | { | |
767 | struct nf_conn_help *help = nfct_help(ct); | |
ceceae1b | 768 | struct nf_conntrack_helper *helper; |
5b1158e9 | 769 | |
5b1158e9 JK |
770 | /* Should be unconfirmed, so not in hash table yet */ |
771 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
772 | ||
0d53778e | 773 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 774 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
775 | |
776 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 777 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 778 | return; |
ceceae1b | 779 | |
c52fbb41 | 780 | rcu_read_lock(); |
ceceae1b YK |
781 | helper = __nf_ct_helper_find(newreply); |
782 | if (helper == NULL) { | |
783 | if (help) | |
784 | rcu_assign_pointer(help->helper, NULL); | |
785 | goto out; | |
5d78a849 | 786 | } |
ceceae1b YK |
787 | |
788 | if (help == NULL) { | |
b560580a PM |
789 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
790 | if (help == NULL) | |
ceceae1b | 791 | goto out; |
ceceae1b YK |
792 | } else { |
793 | memset(&help->help, 0, sizeof(help->help)); | |
794 | } | |
795 | ||
796 | rcu_assign_pointer(help->helper, helper); | |
797 | out: | |
c52fbb41 | 798 | rcu_read_unlock(); |
5b1158e9 | 799 | } |
13b18339 | 800 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 801 | |
9fb9cbb1 YK |
802 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
803 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
804 | enum ip_conntrack_info ctinfo, | |
805 | const struct sk_buff *skb, | |
806 | unsigned long extra_jiffies, | |
807 | int do_acct) | |
808 | { | |
809 | int event = 0; | |
810 | ||
811 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
812 | NF_CT_ASSERT(skb); | |
813 | ||
f8ba1aff | 814 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 815 | |
997ae831 | 816 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
817 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
818 | goto acct; | |
997ae831 | 819 | |
9fb9cbb1 YK |
820 | /* If not in hash table, timer will not be active yet */ |
821 | if (!nf_ct_is_confirmed(ct)) { | |
822 | ct->timeout.expires = extra_jiffies; | |
823 | event = IPCT_REFRESH; | |
824 | } else { | |
be00c8e4 MJ |
825 | unsigned long newtime = jiffies + extra_jiffies; |
826 | ||
827 | /* Only update the timeout if the new timeout is at least | |
828 | HZ jiffies from the old timeout. Need del_timer for race | |
829 | avoidance (may already be dying). */ | |
830 | if (newtime - ct->timeout.expires >= HZ | |
831 | && del_timer(&ct->timeout)) { | |
832 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
833 | add_timer(&ct->timeout); |
834 | event = IPCT_REFRESH; | |
835 | } | |
836 | } | |
837 | ||
47d95045 | 838 | acct: |
9fb9cbb1 | 839 | if (do_acct) { |
58401572 | 840 | struct nf_conn_counter *acct; |
3ffd5eeb | 841 | |
58401572 KPO |
842 | acct = nf_conn_acct_find(ct); |
843 | if (acct) { | |
844 | acct[CTINFO2DIR(ctinfo)].packets++; | |
845 | acct[CTINFO2DIR(ctinfo)].bytes += | |
846 | skb->len - skb_network_offset(skb); | |
847 | } | |
9fb9cbb1 | 848 | } |
9fb9cbb1 | 849 | |
f8ba1aff | 850 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
851 | |
852 | /* must be unlocked when calling event cache */ | |
853 | if (event) | |
a71996fc | 854 | nf_conntrack_event_cache(event, ct); |
9fb9cbb1 | 855 | } |
13b18339 | 856 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 857 | |
4c889498 DM |
858 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
859 | enum ip_conntrack_info ctinfo, | |
860 | const struct sk_buff *skb, | |
861 | int do_acct) | |
51091764 | 862 | { |
718d4ad9 | 863 | if (do_acct) { |
58401572 KPO |
864 | struct nf_conn_counter *acct; |
865 | ||
718d4ad9 | 866 | spin_lock_bh(&nf_conntrack_lock); |
58401572 KPO |
867 | acct = nf_conn_acct_find(ct); |
868 | if (acct) { | |
869 | acct[CTINFO2DIR(ctinfo)].packets++; | |
870 | acct[CTINFO2DIR(ctinfo)].bytes += | |
871 | skb->len - skb_network_offset(skb); | |
872 | } | |
718d4ad9 FH |
873 | spin_unlock_bh(&nf_conntrack_lock); |
874 | } | |
58401572 | 875 | |
4c889498 | 876 | if (del_timer(&ct->timeout)) { |
51091764 | 877 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
878 | return true; |
879 | } | |
880 | return false; | |
51091764 | 881 | } |
718d4ad9 | 882 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 883 | |
e281db5c | 884 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
885 | |
886 | #include <linux/netfilter/nfnetlink.h> | |
887 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
888 | #include <linux/mutex.h> |
889 | ||
c1d10adb PNA |
890 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
891 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
892 | * or depend on ctnetlink */ | |
fdf70832 | 893 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
894 | const struct nf_conntrack_tuple *tuple) |
895 | { | |
77236b6e PM |
896 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
897 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
898 | return 0; |
899 | ||
df6fb868 | 900 | nla_put_failure: |
c1d10adb PNA |
901 | return -1; |
902 | } | |
fdf70832 | 903 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 904 | |
f73e924c PM |
905 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
906 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
907 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 908 | }; |
f73e924c | 909 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 910 | |
fdf70832 | 911 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
912 | struct nf_conntrack_tuple *t) |
913 | { | |
df6fb868 | 914 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
915 | return -EINVAL; |
916 | ||
77236b6e PM |
917 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
918 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
919 | |
920 | return 0; | |
921 | } | |
fdf70832 | 922 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
c1d10adb PNA |
923 | #endif |
924 | ||
9fb9cbb1 | 925 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 926 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
927 | { |
928 | struct nf_conn *ct; | |
929 | enum ip_conntrack_info ctinfo; | |
930 | ||
931 | /* This ICMP is in reverse direction to the packet which caused it */ | |
932 | ct = nf_ct_get(skb, &ctinfo); | |
933 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
934 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
935 | else | |
936 | ctinfo = IP_CT_RELATED; | |
937 | ||
938 | /* Attach to new skbuff, and increment count */ | |
939 | nskb->nfct = &ct->ct_general; | |
940 | nskb->nfctinfo = ctinfo; | |
941 | nf_conntrack_get(nskb->nfct); | |
942 | } | |
943 | ||
9fb9cbb1 | 944 | /* Bring out ya dead! */ |
df0933dc | 945 | static struct nf_conn * |
400dad39 | 946 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
947 | void *data, unsigned int *bucket) |
948 | { | |
df0933dc PM |
949 | struct nf_conntrack_tuple_hash *h; |
950 | struct nf_conn *ct; | |
f205c5e0 | 951 | struct hlist_node *n; |
9fb9cbb1 | 952 | |
f8ba1aff | 953 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 954 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
400dad39 | 955 | hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) { |
df0933dc PM |
956 | ct = nf_ct_tuplehash_to_ctrack(h); |
957 | if (iter(ct, data)) | |
958 | goto found; | |
959 | } | |
601e68e1 | 960 | } |
63c9a262 | 961 | hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) { |
df0933dc PM |
962 | ct = nf_ct_tuplehash_to_ctrack(h); |
963 | if (iter(ct, data)) | |
ec68e97d | 964 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 965 | } |
f8ba1aff | 966 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
967 | return NULL; |
968 | found: | |
c073e3fa | 969 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 970 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 971 | return ct; |
9fb9cbb1 YK |
972 | } |
973 | ||
400dad39 AD |
974 | void nf_ct_iterate_cleanup(struct net *net, |
975 | int (*iter)(struct nf_conn *i, void *data), | |
976 | void *data) | |
9fb9cbb1 | 977 | { |
df0933dc | 978 | struct nf_conn *ct; |
9fb9cbb1 YK |
979 | unsigned int bucket = 0; |
980 | ||
400dad39 | 981 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
982 | /* Time to push up daises... */ |
983 | if (del_timer(&ct->timeout)) | |
984 | death_by_timeout((unsigned long)ct); | |
985 | /* ... else the timer will get him soon. */ | |
986 | ||
987 | nf_ct_put(ct); | |
988 | } | |
989 | } | |
13b18339 | 990 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 YK |
991 | |
992 | static int kill_all(struct nf_conn *i, void *data) | |
993 | { | |
994 | return 1; | |
995 | } | |
996 | ||
96eb24d7 | 997 | void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
998 | { |
999 | if (vmalloced) | |
1000 | vfree(hash); | |
1001 | else | |
601e68e1 | 1002 | free_pages((unsigned long)hash, |
f205c5e0 | 1003 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1004 | } |
ac565e5f | 1005 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1006 | |
400dad39 | 1007 | void nf_conntrack_flush(struct net *net) |
c1d10adb | 1008 | { |
400dad39 | 1009 | nf_ct_iterate_cleanup(net, kill_all, NULL); |
c1d10adb | 1010 | } |
13b18339 | 1011 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 1012 | |
9fb9cbb1 YK |
1013 | /* Mishearing the voices in his head, our hero wonders how he's |
1014 | supposed to kill the mall. */ | |
dfdb8d79 | 1015 | void nf_conntrack_cleanup(struct net *net) |
9fb9cbb1 | 1016 | { |
c3a47ab3 | 1017 | rcu_assign_pointer(ip_ct_attach, NULL); |
7d3cdc6b | 1018 | |
9fb9cbb1 YK |
1019 | /* This makes sure all current packets have passed through |
1020 | netfilter framework. Roll on, two-stage module | |
1021 | delete... */ | |
1022 | synchronize_net(); | |
1023 | ||
6058fa6b AD |
1024 | nf_ct_event_cache_flush(net); |
1025 | nf_conntrack_ecache_fini(net); | |
9fb9cbb1 | 1026 | i_see_dead_people: |
400dad39 | 1027 | nf_conntrack_flush(net); |
49ac8713 | 1028 | if (atomic_read(&net->ct.count) != 0) { |
9fb9cbb1 YK |
1029 | schedule(); |
1030 | goto i_see_dead_people; | |
1031 | } | |
6636568c PM |
1032 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1033 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1034 | schedule(); | |
9fb9cbb1 | 1035 | |
de6e05c4 YK |
1036 | rcu_assign_pointer(nf_ct_destroy, NULL); |
1037 | ||
dacd2a1a | 1038 | kmem_cache_destroy(nf_conntrack_cachep); |
400dad39 | 1039 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1040 | nf_conntrack_htable_size); |
5a6f294e | 1041 | |
58401572 | 1042 | nf_conntrack_acct_fini(); |
9b03f38d | 1043 | nf_conntrack_expect_fini(net); |
0d55af87 | 1044 | free_percpu(net->ct.stat); |
9714be7d KPO |
1045 | nf_conntrack_helper_fini(); |
1046 | nf_conntrack_proto_fini(); | |
9fb9cbb1 YK |
1047 | } |
1048 | ||
96eb24d7 | 1049 | struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced) |
9fb9cbb1 | 1050 | { |
f205c5e0 | 1051 | struct hlist_head *hash; |
8e5105a0 | 1052 | unsigned int size, i; |
9fb9cbb1 | 1053 | |
601e68e1 | 1054 | *vmalloced = 0; |
8e5105a0 | 1055 | |
f205c5e0 | 1056 | size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); |
29b67497 | 1057 | hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN, |
f205c5e0 | 1058 | get_order(sizeof(struct hlist_head) |
9fb9cbb1 | 1059 | * size)); |
601e68e1 | 1060 | if (!hash) { |
9fb9cbb1 YK |
1061 | *vmalloced = 1; |
1062 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
f205c5e0 | 1063 | hash = vmalloc(sizeof(struct hlist_head) * size); |
9fb9cbb1 YK |
1064 | } |
1065 | ||
1066 | if (hash) | |
601e68e1 | 1067 | for (i = 0; i < size; i++) |
f205c5e0 | 1068 | INIT_HLIST_HEAD(&hash[i]); |
9fb9cbb1 YK |
1069 | |
1070 | return hash; | |
1071 | } | |
ac565e5f | 1072 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1073 | |
fae718dd | 1074 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1075 | { |
96eb24d7 SH |
1076 | int i, bucket, vmalloced, old_vmalloced; |
1077 | unsigned int hashsize, old_size; | |
9fb9cbb1 | 1078 | int rnd; |
f205c5e0 | 1079 | struct hlist_head *hash, *old_hash; |
9fb9cbb1 YK |
1080 | struct nf_conntrack_tuple_hash *h; |
1081 | ||
1082 | /* On boot, we can set this without any fancy locking. */ | |
1083 | if (!nf_conntrack_htable_size) | |
1084 | return param_set_uint(val, kp); | |
1085 | ||
96eb24d7 | 1086 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1087 | if (!hashsize) |
1088 | return -EINVAL; | |
1089 | ||
ac565e5f | 1090 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); |
9fb9cbb1 YK |
1091 | if (!hash) |
1092 | return -ENOMEM; | |
1093 | ||
1094 | /* We have to rehahs for the new table anyway, so we also can | |
1095 | * use a newrandom seed */ | |
1096 | get_random_bytes(&rnd, 4); | |
1097 | ||
76507f69 PM |
1098 | /* Lookups in the old hash might happen in parallel, which means we |
1099 | * might get false negatives during connection lookup. New connections | |
1100 | * created because of a false negative won't make it into the hash | |
1101 | * though since that required taking the lock. | |
1102 | */ | |
f8ba1aff | 1103 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1104 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
400dad39 AD |
1105 | while (!hlist_empty(&init_net.ct.hash[i])) { |
1106 | h = hlist_entry(init_net.ct.hash[i].first, | |
f205c5e0 | 1107 | struct nf_conntrack_tuple_hash, hnode); |
76507f69 | 1108 | hlist_del_rcu(&h->hnode); |
9fb9cbb1 | 1109 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
f205c5e0 | 1110 | hlist_add_head(&h->hnode, &hash[bucket]); |
9fb9cbb1 YK |
1111 | } |
1112 | } | |
1113 | old_size = nf_conntrack_htable_size; | |
400dad39 AD |
1114 | old_vmalloced = init_net.ct.hash_vmalloc; |
1115 | old_hash = init_net.ct.hash; | |
9fb9cbb1 YK |
1116 | |
1117 | nf_conntrack_htable_size = hashsize; | |
400dad39 AD |
1118 | init_net.ct.hash_vmalloc = vmalloced; |
1119 | init_net.ct.hash = hash; | |
9fb9cbb1 | 1120 | nf_conntrack_hash_rnd = rnd; |
f8ba1aff | 1121 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1122 | |
ac565e5f | 1123 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1124 | return 0; |
1125 | } | |
fae718dd | 1126 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1127 | |
fae718dd | 1128 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1129 | &nf_conntrack_htable_size, 0600); |
1130 | ||
dfdb8d79 | 1131 | int nf_conntrack_init(struct net *net) |
9fb9cbb1 | 1132 | { |
f205c5e0 | 1133 | int max_factor = 8; |
9fb9cbb1 YK |
1134 | int ret; |
1135 | ||
1136 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1137 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1138 | if (!nf_conntrack_htable_size) { |
1139 | nf_conntrack_htable_size | |
1140 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
f205c5e0 | 1141 | / sizeof(struct hlist_head)); |
9fb9cbb1 | 1142 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1143 | nf_conntrack_htable_size = 16384; |
1144 | if (nf_conntrack_htable_size < 32) | |
1145 | nf_conntrack_htable_size = 32; | |
1146 | ||
1147 | /* Use a max. factor of four by default to get the same max as | |
1148 | * with the old struct list_heads. When a table size is given | |
1149 | * we use the old value of 8 to avoid reducing the max. | |
1150 | * entries. */ | |
1151 | max_factor = 4; | |
9fb9cbb1 | 1152 | } |
49ac8713 | 1153 | atomic_set(&net->ct.count, 0); |
0d55af87 AD |
1154 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1155 | if (!net->ct.stat) | |
1156 | goto err_stat; | |
6058fa6b AD |
1157 | ret = nf_conntrack_ecache_init(net); |
1158 | if (ret < 0) | |
1159 | goto err_ecache; | |
400dad39 AD |
1160 | net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
1161 | &net->ct.hash_vmalloc); | |
1162 | if (!net->ct.hash) { | |
9fb9cbb1 | 1163 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); |
6058fa6b | 1164 | goto err_hash; |
9fb9cbb1 | 1165 | } |
63c9a262 | 1166 | INIT_HLIST_HEAD(&net->ct.unconfirmed); |
9fb9cbb1 | 1167 | |
f205c5e0 | 1168 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1169 | |
1170 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1171 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1172 | nf_conntrack_max); | |
1173 | ||
dacd2a1a YK |
1174 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1175 | sizeof(struct nf_conn), | |
20c2df83 | 1176 | 0, 0, NULL); |
dacd2a1a | 1177 | if (!nf_conntrack_cachep) { |
9fb9cbb1 YK |
1178 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
1179 | goto err_free_hash; | |
1180 | } | |
1181 | ||
e9c1b084 PM |
1182 | ret = nf_conntrack_proto_init(); |
1183 | if (ret < 0) | |
9fb9cbb1 | 1184 | goto err_free_conntrack_slab; |
9fb9cbb1 | 1185 | |
9b03f38d | 1186 | ret = nf_conntrack_expect_init(net); |
933a41e7 | 1187 | if (ret < 0) |
e9c1b084 | 1188 | goto out_fini_proto; |
933a41e7 | 1189 | |
ceceae1b YK |
1190 | ret = nf_conntrack_helper_init(); |
1191 | if (ret < 0) | |
e9c1b084 | 1192 | goto out_fini_expect; |
ceceae1b | 1193 | |
58401572 KPO |
1194 | ret = nf_conntrack_acct_init(); |
1195 | if (ret < 0) | |
1196 | goto out_fini_helper; | |
1197 | ||
7d3cdc6b | 1198 | /* For use by REJECT target */ |
b334aadc | 1199 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); |
de6e05c4 | 1200 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); |
7d3cdc6b | 1201 | |
9fb9cbb1 YK |
1202 | /* Set up fake conntrack: |
1203 | - to never be deleted, not in any hashes */ | |
5a1fb391 AD |
1204 | #ifdef CONFIG_NET_NS |
1205 | nf_conntrack_untracked.ct_net = &init_net; | |
1206 | #endif | |
9fb9cbb1 YK |
1207 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); |
1208 | /* - and look it like as a confirmed connection */ | |
1209 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1210 | ||
1211 | return ret; | |
1212 | ||
58401572 KPO |
1213 | out_fini_helper: |
1214 | nf_conntrack_helper_fini(); | |
e9c1b084 | 1215 | out_fini_expect: |
9b03f38d | 1216 | nf_conntrack_expect_fini(net); |
ceceae1b YK |
1217 | out_fini_proto: |
1218 | nf_conntrack_proto_fini(); | |
9fb9cbb1 | 1219 | err_free_conntrack_slab: |
dacd2a1a | 1220 | kmem_cache_destroy(nf_conntrack_cachep); |
9fb9cbb1 | 1221 | err_free_hash: |
400dad39 | 1222 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1223 | nf_conntrack_htable_size); |
6058fa6b AD |
1224 | err_hash: |
1225 | nf_conntrack_ecache_fini(net); | |
1226 | err_ecache: | |
0d55af87 AD |
1227 | free_percpu(net->ct.stat); |
1228 | err_stat: | |
9fb9cbb1 YK |
1229 | return -ENOMEM; |
1230 | } |