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> | |
d43c36dc | 17 | #include <linux/sched.h> |
9fb9cbb1 YK |
18 | #include <linux/skbuff.h> |
19 | #include <linux/proc_fs.h> | |
20 | #include <linux/vmalloc.h> | |
21 | #include <linux/stddef.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/random.h> | |
24 | #include <linux/jhash.h> | |
25 | #include <linux/err.h> | |
26 | #include <linux/percpu.h> | |
27 | #include <linux/moduleparam.h> | |
28 | #include <linux/notifier.h> | |
29 | #include <linux/kernel.h> | |
30 | #include <linux/netdevice.h> | |
31 | #include <linux/socket.h> | |
d7fe0f24 | 32 | #include <linux/mm.h> |
ea781f19 | 33 | #include <linux/rculist_nulls.h> |
9fb9cbb1 | 34 | |
9fb9cbb1 YK |
35 | #include <net/netfilter/nf_conntrack.h> |
36 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
605dcad6 | 37 | #include <net/netfilter/nf_conntrack_l4proto.h> |
77ab9cff | 38 | #include <net/netfilter/nf_conntrack_expect.h> |
9fb9cbb1 YK |
39 | #include <net/netfilter/nf_conntrack_helper.h> |
40 | #include <net/netfilter/nf_conntrack_core.h> | |
ecfab2c9 | 41 | #include <net/netfilter/nf_conntrack_extend.h> |
58401572 | 42 | #include <net/netfilter/nf_conntrack_acct.h> |
a0891aa6 | 43 | #include <net/netfilter/nf_conntrack_ecache.h> |
e6a7d3c0 | 44 | #include <net/netfilter/nf_nat.h> |
e17b666a | 45 | #include <net/netfilter/nf_nat_core.h> |
9fb9cbb1 | 46 | |
dc808fe2 | 47 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 48 | |
e17b666a PM |
49 | int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, |
50 | enum nf_nat_manip_type manip, | |
39938324 | 51 | const struct nlattr *attr) __read_mostly; |
e6a7d3c0 PNA |
52 | EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); |
53 | ||
f8ba1aff | 54 | DEFINE_SPINLOCK(nf_conntrack_lock); |
13b18339 | 55 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 | 56 | |
e2b7606c | 57 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
58 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
59 | ||
e478075c | 60 | unsigned int nf_conntrack_max __read_mostly; |
a999e683 | 61 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 62 | |
e2b7606c | 63 | struct nf_conn nf_conntrack_untracked __read_mostly; |
13b18339 PM |
64 | EXPORT_SYMBOL_GPL(nf_conntrack_untracked); |
65 | ||
dacd2a1a | 66 | static struct kmem_cache *nf_conntrack_cachep __read_mostly; |
77ab9cff | 67 | |
9fb9cbb1 YK |
68 | static int nf_conntrack_hash_rnd_initted; |
69 | static unsigned int nf_conntrack_hash_rnd; | |
70 | ||
71 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
72 | unsigned int size, unsigned int rnd) | |
73 | { | |
0794935e PM |
74 | unsigned int n; |
75 | u_int32_t h; | |
76 | ||
77 | /* The direction must be ignored, so we hash everything up to the | |
78 | * destination ports (which is a multiple of 4) and treat the last | |
79 | * three bytes manually. | |
80 | */ | |
81 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
82 | h = jhash2((u32 *)tuple, n, | |
83 | rnd ^ (((__force __u16)tuple->dst.u.all << 16) | | |
84 | tuple->dst.protonum)); | |
85 | ||
86 | return ((u64)h * size) >> 32; | |
9fb9cbb1 YK |
87 | } |
88 | ||
89 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
90 | { | |
91 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
92 | nf_conntrack_hash_rnd); | |
93 | } | |
94 | ||
5f2b4c90 | 95 | bool |
9fb9cbb1 YK |
96 | nf_ct_get_tuple(const struct sk_buff *skb, |
97 | unsigned int nhoff, | |
98 | unsigned int dataoff, | |
99 | u_int16_t l3num, | |
100 | u_int8_t protonum, | |
101 | struct nf_conntrack_tuple *tuple, | |
102 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 103 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 104 | { |
443a70d5 | 105 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
106 | |
107 | tuple->src.l3num = l3num; | |
108 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 109 | return false; |
9fb9cbb1 YK |
110 | |
111 | tuple->dst.protonum = protonum; | |
112 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
113 | ||
605dcad6 | 114 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 115 | } |
13b18339 | 116 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 117 | |
5f2b4c90 JE |
118 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
119 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
120 | { |
121 | struct nf_conntrack_l3proto *l3proto; | |
122 | struct nf_conntrack_l4proto *l4proto; | |
123 | unsigned int protoff; | |
124 | u_int8_t protonum; | |
125 | int ret; | |
126 | ||
127 | rcu_read_lock(); | |
128 | ||
129 | l3proto = __nf_ct_l3proto_find(l3num); | |
130 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
131 | if (ret != NF_ACCEPT) { | |
132 | rcu_read_unlock(); | |
5f2b4c90 | 133 | return false; |
e2a3123f YK |
134 | } |
135 | ||
136 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
137 | ||
138 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
139 | l3proto, l4proto); | |
140 | ||
141 | rcu_read_unlock(); | |
142 | return ret; | |
143 | } | |
144 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
145 | ||
5f2b4c90 | 146 | bool |
9fb9cbb1 YK |
147 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
148 | const struct nf_conntrack_tuple *orig, | |
149 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 150 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 151 | { |
443a70d5 | 152 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
153 | |
154 | inverse->src.l3num = orig->src.l3num; | |
155 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 156 | return false; |
9fb9cbb1 YK |
157 | |
158 | inverse->dst.dir = !orig->dst.dir; | |
159 | ||
160 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 161 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 162 | } |
13b18339 | 163 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 164 | |
9fb9cbb1 YK |
165 | static void |
166 | clean_from_lists(struct nf_conn *ct) | |
167 | { | |
0d53778e | 168 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
169 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
170 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
171 | |
172 | /* Destroy all pending expectations */ | |
c1d10adb | 173 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
174 | } |
175 | ||
176 | static void | |
177 | destroy_conntrack(struct nf_conntrack *nfct) | |
178 | { | |
179 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
0d55af87 | 180 | struct net *net = nf_ct_net(ct); |
605dcad6 | 181 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 182 | |
0d53778e | 183 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
184 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
185 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
186 | ||
9fb9cbb1 YK |
187 | /* To make sure we don't get any weird locking issues here: |
188 | * destroy_conntrack() MUST NOT be called with a write lock | |
189 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 190 | rcu_read_lock(); |
5e8fbe2a | 191 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
192 | if (l4proto && l4proto->destroy) |
193 | l4proto->destroy(ct); | |
9fb9cbb1 | 194 | |
982d9a9c | 195 | rcu_read_unlock(); |
9fb9cbb1 | 196 | |
f8ba1aff | 197 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
198 | /* Expectations will have been removed in clean_from_lists, |
199 | * except TFTP can create an expectation on the first packet, | |
200 | * before connection is in the list, so we need to clean here, | |
201 | * too. */ | |
c1d10adb | 202 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
203 | |
204 | /* We overload first tuple to link into unconfirmed list. */ | |
205 | if (!nf_ct_is_confirmed(ct)) { | |
ea781f19 ED |
206 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); |
207 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
9fb9cbb1 YK |
208 | } |
209 | ||
0d55af87 | 210 | NF_CT_STAT_INC(net, delete); |
f8ba1aff | 211 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
212 | |
213 | if (ct->master) | |
214 | nf_ct_put(ct->master); | |
215 | ||
0d53778e | 216 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
217 | nf_conntrack_free(ct); |
218 | } | |
219 | ||
dd7669a9 | 220 | void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 221 | { |
0d55af87 | 222 | struct net *net = nf_ct_net(ct); |
9fb9cbb1 | 223 | |
9858a3ae | 224 | nf_ct_helper_destroy(ct); |
f8ba1aff | 225 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
226 | /* Inside lock so preempt is disabled on module removal path. |
227 | * Otherwise we can get spurious warnings. */ | |
0d55af87 | 228 | NF_CT_STAT_INC(net, delete_list); |
9fb9cbb1 | 229 | clean_from_lists(ct); |
f8ba1aff | 230 | spin_unlock_bh(&nf_conntrack_lock); |
dd7669a9 PNA |
231 | } |
232 | EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists); | |
233 | ||
234 | static void death_by_event(unsigned long ul_conntrack) | |
235 | { | |
236 | struct nf_conn *ct = (void *)ul_conntrack; | |
237 | struct net *net = nf_ct_net(ct); | |
238 | ||
239 | if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) { | |
240 | /* bad luck, let's retry again */ | |
241 | ct->timeout.expires = jiffies + | |
242 | (random32() % net->ct.sysctl_events_retry_timeout); | |
243 | add_timer(&ct->timeout); | |
244 | return; | |
245 | } | |
246 | /* we've got the event delivered, now it's dying */ | |
247 | set_bit(IPS_DYING_BIT, &ct->status); | |
248 | spin_lock(&nf_conntrack_lock); | |
249 | hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
250 | spin_unlock(&nf_conntrack_lock); | |
251 | nf_ct_put(ct); | |
252 | } | |
253 | ||
254 | void nf_ct_insert_dying_list(struct nf_conn *ct) | |
255 | { | |
256 | struct net *net = nf_ct_net(ct); | |
257 | ||
258 | /* add this conntrack to the dying list */ | |
259 | spin_lock_bh(&nf_conntrack_lock); | |
260 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
261 | &net->ct.dying); | |
262 | spin_unlock_bh(&nf_conntrack_lock); | |
263 | /* set a new timer to retry event delivery */ | |
264 | setup_timer(&ct->timeout, death_by_event, (unsigned long)ct); | |
265 | ct->timeout.expires = jiffies + | |
266 | (random32() % net->ct.sysctl_events_retry_timeout); | |
267 | add_timer(&ct->timeout); | |
268 | } | |
269 | EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list); | |
270 | ||
271 | static void death_by_timeout(unsigned long ul_conntrack) | |
272 | { | |
273 | struct nf_conn *ct = (void *)ul_conntrack; | |
274 | ||
275 | if (!test_bit(IPS_DYING_BIT, &ct->status) && | |
276 | unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) { | |
277 | /* destroy event was not delivered */ | |
278 | nf_ct_delete_from_lists(ct); | |
279 | nf_ct_insert_dying_list(ct); | |
280 | return; | |
281 | } | |
282 | set_bit(IPS_DYING_BIT, &ct->status); | |
283 | nf_ct_delete_from_lists(ct); | |
9fb9cbb1 YK |
284 | nf_ct_put(ct); |
285 | } | |
286 | ||
ea781f19 ED |
287 | /* |
288 | * Warning : | |
289 | * - Caller must take a reference on returned object | |
290 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
291 | * OR | |
292 | * - Caller must lock nf_conntrack_lock before calling this function | |
293 | */ | |
c1d10adb | 294 | struct nf_conntrack_tuple_hash * |
400dad39 | 295 | __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
296 | { |
297 | struct nf_conntrack_tuple_hash *h; | |
ea781f19 | 298 | struct hlist_nulls_node *n; |
9fb9cbb1 YK |
299 | unsigned int hash = hash_conntrack(tuple); |
300 | ||
4e29e9ec PM |
301 | /* Disable BHs the entire time since we normally need to disable them |
302 | * at least once for the stats anyway. | |
303 | */ | |
304 | local_bh_disable(); | |
ea781f19 ED |
305 | begin: |
306 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { | |
ba419aff | 307 | if (nf_ct_tuple_equal(tuple, &h->tuple)) { |
0d55af87 | 308 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 309 | local_bh_enable(); |
9fb9cbb1 YK |
310 | return h; |
311 | } | |
0d55af87 | 312 | NF_CT_STAT_INC(net, searched); |
9fb9cbb1 | 313 | } |
ea781f19 ED |
314 | /* |
315 | * if the nulls value we got at the end of this lookup is | |
316 | * not the expected one, we must restart lookup. | |
317 | * We probably met an item that was moved to another chain. | |
318 | */ | |
319 | if (get_nulls_value(n) != hash) | |
320 | goto begin; | |
4e29e9ec | 321 | local_bh_enable(); |
9fb9cbb1 YK |
322 | |
323 | return NULL; | |
324 | } | |
13b18339 | 325 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
326 | |
327 | /* Find a connection corresponding to a tuple. */ | |
328 | struct nf_conntrack_tuple_hash * | |
400dad39 | 329 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
330 | { |
331 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 332 | struct nf_conn *ct; |
9fb9cbb1 | 333 | |
76507f69 | 334 | rcu_read_lock(); |
ea781f19 | 335 | begin: |
400dad39 | 336 | h = __nf_conntrack_find(net, tuple); |
76507f69 PM |
337 | if (h) { |
338 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
339 | if (unlikely(nf_ct_is_dying(ct) || |
340 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 341 | h = NULL; |
ea781f19 ED |
342 | else { |
343 | if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) { | |
344 | nf_ct_put(ct); | |
345 | goto begin; | |
346 | } | |
347 | } | |
76507f69 PM |
348 | } |
349 | rcu_read_unlock(); | |
9fb9cbb1 YK |
350 | |
351 | return h; | |
352 | } | |
13b18339 | 353 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 354 | |
c1d10adb PNA |
355 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
356 | unsigned int hash, | |
601e68e1 | 357 | unsigned int repl_hash) |
c1d10adb | 358 | { |
400dad39 AD |
359 | struct net *net = nf_ct_net(ct); |
360 | ||
ea781f19 | 361 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
400dad39 | 362 | &net->ct.hash[hash]); |
ea781f19 | 363 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
400dad39 | 364 | &net->ct.hash[repl_hash]); |
c1d10adb PNA |
365 | } |
366 | ||
367 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
368 | { | |
369 | unsigned int hash, repl_hash; | |
370 | ||
371 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
372 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
373 | ||
c1d10adb | 374 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
c1d10adb | 375 | } |
13b18339 | 376 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 377 | |
9fb9cbb1 YK |
378 | /* Confirm a connection given skb; places it in hash table */ |
379 | int | |
3db05fea | 380 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
381 | { |
382 | unsigned int hash, repl_hash; | |
df0933dc | 383 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 384 | struct nf_conn *ct; |
df0933dc | 385 | struct nf_conn_help *help; |
ea781f19 | 386 | struct hlist_nulls_node *n; |
9fb9cbb1 | 387 | enum ip_conntrack_info ctinfo; |
400dad39 | 388 | struct net *net; |
9fb9cbb1 | 389 | |
3db05fea | 390 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 391 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
392 | |
393 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
394 | ICMP/TCP RST packets in other direction. Actual packet | |
395 | which created connection will be IP_CT_NEW or for an | |
396 | expected connection, IP_CT_RELATED. */ | |
397 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
398 | return NF_ACCEPT; | |
399 | ||
400 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
401 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
402 | ||
403 | /* We're not in hash table, and we refuse to set up related | |
404 | connections for unconfirmed conns. But packet copies and | |
405 | REJECT will give spurious warnings here. */ | |
406 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
407 | ||
408 | /* No external references means noone else could have | |
409 | confirmed us. */ | |
410 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 411 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 412 | |
f8ba1aff | 413 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
414 | |
415 | /* See if there's one in the list already, including reverse: | |
416 | NAT could have grabbed it without realizing, since we're | |
417 | not in the hash. If there is, we lost race. */ | |
ea781f19 | 418 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) |
df0933dc PM |
419 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
420 | &h->tuple)) | |
421 | goto out; | |
ea781f19 | 422 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode) |
df0933dc PM |
423 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
424 | &h->tuple)) | |
425 | goto out; | |
9fb9cbb1 | 426 | |
df0933dc | 427 | /* Remove from unconfirmed list */ |
ea781f19 | 428 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
df0933dc | 429 | |
df0933dc PM |
430 | /* Timer relative to confirmation time, not original |
431 | setting time, otherwise we'd get timer wrap in | |
432 | weird delay cases. */ | |
433 | ct->timeout.expires += jiffies; | |
434 | add_timer(&ct->timeout); | |
435 | atomic_inc(&ct->ct_general.use); | |
436 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
5c8ec910 PM |
437 | |
438 | /* Since the lookup is lockless, hash insertion must be done after | |
439 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
440 | * guarantee that no other CPU can find the conntrack before the above | |
441 | * stores are visible. | |
442 | */ | |
443 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
0d55af87 | 444 | NF_CT_STAT_INC(net, insert); |
f8ba1aff | 445 | spin_unlock_bh(&nf_conntrack_lock); |
5c8ec910 | 446 | |
df0933dc PM |
447 | help = nfct_help(ct); |
448 | if (help && help->helper) | |
a71996fc | 449 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 450 | |
df0933dc | 451 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 452 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 453 | return NF_ACCEPT; |
9fb9cbb1 | 454 | |
df0933dc | 455 | out: |
0d55af87 | 456 | NF_CT_STAT_INC(net, insert_failed); |
f8ba1aff | 457 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
458 | return NF_DROP; |
459 | } | |
13b18339 | 460 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
461 | |
462 | /* Returns true if a connection correspondings to the tuple (required | |
463 | for NAT). */ | |
464 | int | |
465 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
466 | const struct nf_conn *ignored_conntrack) | |
467 | { | |
400dad39 | 468 | struct net *net = nf_ct_net(ignored_conntrack); |
9fb9cbb1 | 469 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 470 | struct hlist_nulls_node *n; |
ba419aff | 471 | unsigned int hash = hash_conntrack(tuple); |
9fb9cbb1 | 472 | |
4e29e9ec PM |
473 | /* Disable BHs the entire time since we need to disable them at |
474 | * least once for the stats anyway. | |
475 | */ | |
476 | rcu_read_lock_bh(); | |
ea781f19 | 477 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { |
ba419aff PM |
478 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
479 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
0d55af87 | 480 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 481 | rcu_read_unlock_bh(); |
ba419aff PM |
482 | return 1; |
483 | } | |
0d55af87 | 484 | NF_CT_STAT_INC(net, searched); |
ba419aff | 485 | } |
4e29e9ec | 486 | rcu_read_unlock_bh(); |
9fb9cbb1 | 487 | |
ba419aff | 488 | return 0; |
9fb9cbb1 | 489 | } |
13b18339 | 490 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 491 | |
7ae7730f PM |
492 | #define NF_CT_EVICTION_RANGE 8 |
493 | ||
9fb9cbb1 YK |
494 | /* There's a small race here where we may free a just-assured |
495 | connection. Too bad: we're in trouble anyway. */ | |
400dad39 | 496 | static noinline int early_drop(struct net *net, unsigned int hash) |
9fb9cbb1 | 497 | { |
f205c5e0 | 498 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 499 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 500 | struct nf_conn *ct = NULL, *tmp; |
ea781f19 | 501 | struct hlist_nulls_node *n; |
7ae7730f | 502 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
503 | int dropped = 0; |
504 | ||
76507f69 | 505 | rcu_read_lock(); |
7ae7730f | 506 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
ea781f19 ED |
507 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], |
508 | hnnode) { | |
7ae7730f PM |
509 | tmp = nf_ct_tuplehash_to_ctrack(h); |
510 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
511 | ct = tmp; | |
512 | cnt++; | |
513 | } | |
76507f69 | 514 | |
8d8890b7 PM |
515 | if (ct && unlikely(nf_ct_is_dying(ct) || |
516 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 517 | ct = NULL; |
7ae7730f PM |
518 | if (ct || cnt >= NF_CT_EVICTION_RANGE) |
519 | break; | |
520 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 521 | } |
76507f69 | 522 | rcu_read_unlock(); |
9fb9cbb1 YK |
523 | |
524 | if (!ct) | |
525 | return dropped; | |
526 | ||
527 | if (del_timer(&ct->timeout)) { | |
528 | death_by_timeout((unsigned long)ct); | |
529 | dropped = 1; | |
0d55af87 | 530 | NF_CT_STAT_INC_ATOMIC(net, early_drop); |
9fb9cbb1 YK |
531 | } |
532 | nf_ct_put(ct); | |
533 | return dropped; | |
534 | } | |
535 | ||
5a1fb391 AD |
536 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
537 | const struct nf_conntrack_tuple *orig, | |
b891c5a8 PNA |
538 | const struct nf_conntrack_tuple *repl, |
539 | gfp_t gfp) | |
9fb9cbb1 | 540 | { |
cd7fcbf1 | 541 | struct nf_conn *ct; |
9fb9cbb1 | 542 | |
dc808fe2 | 543 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
af07d241 HPP |
544 | get_random_bytes(&nf_conntrack_hash_rnd, |
545 | sizeof(nf_conntrack_hash_rnd)); | |
9fb9cbb1 YK |
546 | nf_conntrack_hash_rnd_initted = 1; |
547 | } | |
548 | ||
5251e2d2 | 549 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 550 | atomic_inc(&net->ct.count); |
5251e2d2 | 551 | |
76eb9460 | 552 | if (nf_conntrack_max && |
49ac8713 | 553 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
9fb9cbb1 | 554 | unsigned int hash = hash_conntrack(orig); |
400dad39 | 555 | if (!early_drop(net, hash)) { |
49ac8713 | 556 | atomic_dec(&net->ct.count); |
9fb9cbb1 YK |
557 | if (net_ratelimit()) |
558 | printk(KERN_WARNING | |
559 | "nf_conntrack: table full, dropping" | |
560 | " packet.\n"); | |
561 | return ERR_PTR(-ENOMEM); | |
562 | } | |
563 | } | |
564 | ||
941297f4 ED |
565 | /* |
566 | * Do not use kmem_cache_zalloc(), as this cache uses | |
567 | * SLAB_DESTROY_BY_RCU. | |
568 | */ | |
569 | ct = kmem_cache_alloc(nf_conntrack_cachep, gfp); | |
c88130bc | 570 | if (ct == NULL) { |
0d53778e | 571 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
49ac8713 | 572 | atomic_dec(&net->ct.count); |
dacd2a1a | 573 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 574 | } |
941297f4 ED |
575 | /* |
576 | * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next | |
577 | * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged. | |
578 | */ | |
579 | memset(&ct->tuplehash[IP_CT_DIR_MAX], 0, | |
580 | sizeof(*ct) - offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX])); | |
440f0d58 | 581 | spin_lock_init(&ct->lock); |
c88130bc | 582 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 583 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 584 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
941297f4 | 585 | ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev = NULL; |
9fb9cbb1 | 586 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 587 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
5a1fb391 AD |
588 | #ifdef CONFIG_NET_NS |
589 | ct->ct_net = net; | |
590 | #endif | |
9fb9cbb1 | 591 | |
941297f4 ED |
592 | /* |
593 | * changes to lookup keys must be done before setting refcnt to 1 | |
594 | */ | |
595 | smp_wmb(); | |
596 | atomic_set(&ct->ct_general.use, 1); | |
c88130bc | 597 | return ct; |
9fb9cbb1 | 598 | } |
13b18339 | 599 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 600 | |
c88130bc | 601 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 602 | { |
1d45209d ED |
603 | struct net *net = nf_ct_net(ct); |
604 | ||
ceeff754 | 605 | nf_ct_ext_destroy(ct); |
1d45209d | 606 | atomic_dec(&net->ct.count); |
ea781f19 ED |
607 | nf_ct_ext_free(ct); |
608 | kmem_cache_free(nf_conntrack_cachep, ct); | |
76507f69 | 609 | } |
13b18339 | 610 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
611 | |
612 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
613 | failed due to stress. Otherwise it really is unclassifiable. */ | |
614 | static struct nf_conntrack_tuple_hash * | |
5a1fb391 AD |
615 | init_conntrack(struct net *net, |
616 | const struct nf_conntrack_tuple *tuple, | |
9fb9cbb1 | 617 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 618 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
619 | struct sk_buff *skb, |
620 | unsigned int dataoff) | |
621 | { | |
c88130bc | 622 | struct nf_conn *ct; |
3c158f7f | 623 | struct nf_conn_help *help; |
9fb9cbb1 YK |
624 | struct nf_conntrack_tuple repl_tuple; |
625 | struct nf_conntrack_expect *exp; | |
626 | ||
605dcad6 | 627 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 628 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
629 | return NULL; |
630 | } | |
631 | ||
5a1fb391 | 632 | ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC); |
cd7fcbf1 | 633 | if (IS_ERR(ct)) { |
0d53778e | 634 | pr_debug("Can't allocate conntrack.\n"); |
c88130bc | 635 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 YK |
636 | } |
637 | ||
c88130bc PM |
638 | if (!l4proto->new(ct, skb, dataoff)) { |
639 | nf_conntrack_free(ct); | |
0d53778e | 640 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
641 | return NULL; |
642 | } | |
643 | ||
58401572 | 644 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a0891aa6 | 645 | nf_ct_ecache_ext_add(ct, GFP_ATOMIC); |
58401572 | 646 | |
f8ba1aff | 647 | spin_lock_bh(&nf_conntrack_lock); |
9b03f38d | 648 | exp = nf_ct_find_expectation(net, tuple); |
9fb9cbb1 | 649 | if (exp) { |
0d53778e | 650 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 651 | ct, exp); |
9fb9cbb1 | 652 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
653 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
654 | ct->master = exp->master; | |
ceceae1b | 655 | if (exp->helper) { |
c88130bc | 656 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b YK |
657 | if (help) |
658 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
659 | } |
660 | ||
9fb9cbb1 | 661 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 662 | ct->mark = exp->master->mark; |
7c9728c3 JM |
663 | #endif |
664 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 665 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 666 | #endif |
c88130bc | 667 | nf_conntrack_get(&ct->master->ct_general); |
0d55af87 | 668 | NF_CT_STAT_INC(net, expect_new); |
22e7410b | 669 | } else { |
226c0c0e | 670 | __nf_ct_try_assign_helper(ct, GFP_ATOMIC); |
0d55af87 | 671 | NF_CT_STAT_INC(net, new); |
22e7410b | 672 | } |
9fb9cbb1 YK |
673 | |
674 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
ea781f19 | 675 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
63c9a262 | 676 | &net->ct.unconfirmed); |
9fb9cbb1 | 677 | |
f8ba1aff | 678 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
679 | |
680 | if (exp) { | |
681 | if (exp->expectfn) | |
c88130bc | 682 | exp->expectfn(ct, exp); |
6823645d | 683 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
684 | } |
685 | ||
c88130bc | 686 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
687 | } |
688 | ||
689 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
690 | static inline struct nf_conn * | |
a702a65f AD |
691 | resolve_normal_ct(struct net *net, |
692 | struct sk_buff *skb, | |
9fb9cbb1 YK |
693 | unsigned int dataoff, |
694 | u_int16_t l3num, | |
695 | u_int8_t protonum, | |
696 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 697 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
698 | int *set_reply, |
699 | enum ip_conntrack_info *ctinfo) | |
700 | { | |
701 | struct nf_conntrack_tuple tuple; | |
702 | struct nf_conntrack_tuple_hash *h; | |
703 | struct nf_conn *ct; | |
704 | ||
bbe735e4 | 705 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 706 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 707 | l4proto)) { |
0d53778e | 708 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
709 | return NULL; |
710 | } | |
711 | ||
712 | /* look for tuple match */ | |
a702a65f | 713 | h = nf_conntrack_find_get(net, &tuple); |
9fb9cbb1 | 714 | if (!h) { |
a702a65f | 715 | h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
716 | if (!h) |
717 | return NULL; | |
718 | if (IS_ERR(h)) | |
719 | return (void *)h; | |
720 | } | |
721 | ct = nf_ct_tuplehash_to_ctrack(h); | |
722 | ||
723 | /* It exists; we have (non-exclusive) reference. */ | |
724 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
725 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
726 | /* Please set reply bit if this packet OK */ | |
727 | *set_reply = 1; | |
728 | } else { | |
729 | /* Once we've had two way comms, always ESTABLISHED. */ | |
730 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 731 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
732 | *ctinfo = IP_CT_ESTABLISHED; |
733 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
734 | pr_debug("nf_conntrack_in: related packet for %p\n", |
735 | ct); | |
9fb9cbb1 YK |
736 | *ctinfo = IP_CT_RELATED; |
737 | } else { | |
0d53778e | 738 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
739 | *ctinfo = IP_CT_NEW; |
740 | } | |
741 | *set_reply = 0; | |
742 | } | |
743 | skb->nfct = &ct->ct_general; | |
744 | skb->nfctinfo = *ctinfo; | |
745 | return ct; | |
746 | } | |
747 | ||
748 | unsigned int | |
a702a65f AD |
749 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
750 | struct sk_buff *skb) | |
9fb9cbb1 YK |
751 | { |
752 | struct nf_conn *ct; | |
753 | enum ip_conntrack_info ctinfo; | |
754 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 755 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
756 | unsigned int dataoff; |
757 | u_int8_t protonum; | |
758 | int set_reply = 0; | |
759 | int ret; | |
760 | ||
761 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 762 | if (skb->nfct) { |
0d55af87 | 763 | NF_CT_STAT_INC_ATOMIC(net, ignore); |
9fb9cbb1 YK |
764 | return NF_ACCEPT; |
765 | } | |
766 | ||
923f4902 | 767 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 768 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 769 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
770 | &dataoff, &protonum); |
771 | if (ret <= 0) { | |
0d53778e | 772 | pr_debug("not prepared to track yet or error occured\n"); |
0d55af87 AD |
773 | NF_CT_STAT_INC_ATOMIC(net, error); |
774 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
9fb9cbb1 YK |
775 | return -ret; |
776 | } | |
777 | ||
76108cea | 778 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
779 | |
780 | /* It may be an special packet, error, unclean... | |
781 | * inverse of the return code tells to the netfilter | |
782 | * core what to do with the packet. */ | |
74c51a14 AD |
783 | if (l4proto->error != NULL) { |
784 | ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum); | |
785 | if (ret <= 0) { | |
0d55af87 AD |
786 | NF_CT_STAT_INC_ATOMIC(net, error); |
787 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
74c51a14 AD |
788 | return -ret; |
789 | } | |
9fb9cbb1 YK |
790 | } |
791 | ||
a702a65f AD |
792 | ct = resolve_normal_ct(net, skb, dataoff, pf, protonum, |
793 | l3proto, l4proto, &set_reply, &ctinfo); | |
9fb9cbb1 YK |
794 | if (!ct) { |
795 | /* Not valid part of a connection */ | |
0d55af87 | 796 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
9fb9cbb1 YK |
797 | return NF_ACCEPT; |
798 | } | |
799 | ||
800 | if (IS_ERR(ct)) { | |
801 | /* Too stressed to deal. */ | |
0d55af87 | 802 | NF_CT_STAT_INC_ATOMIC(net, drop); |
9fb9cbb1 YK |
803 | return NF_DROP; |
804 | } | |
805 | ||
3db05fea | 806 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 807 | |
3db05fea | 808 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
ec8d5409 | 809 | if (ret <= 0) { |
9fb9cbb1 YK |
810 | /* Invalid: inverse of the return code tells |
811 | * the netfilter core what to do */ | |
0d53778e | 812 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
813 | nf_conntrack_put(skb->nfct); |
814 | skb->nfct = NULL; | |
0d55af87 | 815 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
816 | if (ret == -NF_DROP) |
817 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
9fb9cbb1 YK |
818 | return -ret; |
819 | } | |
820 | ||
821 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
a71996fc | 822 | nf_conntrack_event_cache(IPCT_STATUS, ct); |
9fb9cbb1 YK |
823 | |
824 | return ret; | |
825 | } | |
13b18339 | 826 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 827 | |
5f2b4c90 JE |
828 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
829 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 830 | { |
5f2b4c90 | 831 | bool ret; |
923f4902 PM |
832 | |
833 | rcu_read_lock(); | |
834 | ret = nf_ct_invert_tuple(inverse, orig, | |
835 | __nf_ct_l3proto_find(orig->src.l3num), | |
836 | __nf_ct_l4proto_find(orig->src.l3num, | |
837 | orig->dst.protonum)); | |
838 | rcu_read_unlock(); | |
839 | return ret; | |
9fb9cbb1 | 840 | } |
13b18339 | 841 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 842 | |
5b1158e9 JK |
843 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
844 | implicitly racy: see __nf_conntrack_confirm */ | |
845 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
846 | const struct nf_conntrack_tuple *newreply) | |
847 | { | |
848 | struct nf_conn_help *help = nfct_help(ct); | |
849 | ||
5b1158e9 JK |
850 | /* Should be unconfirmed, so not in hash table yet */ |
851 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
852 | ||
0d53778e | 853 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 854 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
855 | |
856 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 857 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 858 | return; |
ceceae1b | 859 | |
c52fbb41 | 860 | rcu_read_lock(); |
226c0c0e | 861 | __nf_ct_try_assign_helper(ct, GFP_ATOMIC); |
c52fbb41 | 862 | rcu_read_unlock(); |
5b1158e9 | 863 | } |
13b18339 | 864 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 865 | |
9fb9cbb1 YK |
866 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
867 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
868 | enum ip_conntrack_info ctinfo, | |
869 | const struct sk_buff *skb, | |
870 | unsigned long extra_jiffies, | |
871 | int do_acct) | |
872 | { | |
9fb9cbb1 YK |
873 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); |
874 | NF_CT_ASSERT(skb); | |
875 | ||
997ae831 | 876 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
877 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
878 | goto acct; | |
997ae831 | 879 | |
9fb9cbb1 YK |
880 | /* If not in hash table, timer will not be active yet */ |
881 | if (!nf_ct_is_confirmed(ct)) { | |
882 | ct->timeout.expires = extra_jiffies; | |
9fb9cbb1 | 883 | } else { |
be00c8e4 MJ |
884 | unsigned long newtime = jiffies + extra_jiffies; |
885 | ||
886 | /* Only update the timeout if the new timeout is at least | |
887 | HZ jiffies from the old timeout. Need del_timer for race | |
888 | avoidance (may already be dying). */ | |
65cb9fda PM |
889 | if (newtime - ct->timeout.expires >= HZ) |
890 | mod_timer_pending(&ct->timeout, newtime); | |
9fb9cbb1 YK |
891 | } |
892 | ||
47d95045 | 893 | acct: |
9fb9cbb1 | 894 | if (do_acct) { |
58401572 | 895 | struct nf_conn_counter *acct; |
3ffd5eeb | 896 | |
58401572 KPO |
897 | acct = nf_conn_acct_find(ct); |
898 | if (acct) { | |
65cb9fda | 899 | spin_lock_bh(&ct->lock); |
58401572 KPO |
900 | acct[CTINFO2DIR(ctinfo)].packets++; |
901 | acct[CTINFO2DIR(ctinfo)].bytes += | |
902 | skb->len - skb_network_offset(skb); | |
65cb9fda | 903 | spin_unlock_bh(&ct->lock); |
58401572 | 904 | } |
9fb9cbb1 | 905 | } |
9fb9cbb1 | 906 | } |
13b18339 | 907 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 908 | |
4c889498 DM |
909 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
910 | enum ip_conntrack_info ctinfo, | |
911 | const struct sk_buff *skb, | |
912 | int do_acct) | |
51091764 | 913 | { |
718d4ad9 | 914 | if (do_acct) { |
58401572 KPO |
915 | struct nf_conn_counter *acct; |
916 | ||
58401572 KPO |
917 | acct = nf_conn_acct_find(ct); |
918 | if (acct) { | |
65cb9fda | 919 | spin_lock_bh(&ct->lock); |
58401572 KPO |
920 | acct[CTINFO2DIR(ctinfo)].packets++; |
921 | acct[CTINFO2DIR(ctinfo)].bytes += | |
922 | skb->len - skb_network_offset(skb); | |
65cb9fda | 923 | spin_unlock_bh(&ct->lock); |
58401572 | 924 | } |
718d4ad9 | 925 | } |
58401572 | 926 | |
4c889498 | 927 | if (del_timer(&ct->timeout)) { |
51091764 | 928 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
929 | return true; |
930 | } | |
931 | return false; | |
51091764 | 932 | } |
718d4ad9 | 933 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 934 | |
e281db5c | 935 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
936 | |
937 | #include <linux/netfilter/nfnetlink.h> | |
938 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
939 | #include <linux/mutex.h> |
940 | ||
c1d10adb PNA |
941 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
942 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
943 | * or depend on ctnetlink */ | |
fdf70832 | 944 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
945 | const struct nf_conntrack_tuple *tuple) |
946 | { | |
77236b6e PM |
947 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
948 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
949 | return 0; |
950 | ||
df6fb868 | 951 | nla_put_failure: |
c1d10adb PNA |
952 | return -1; |
953 | } | |
fdf70832 | 954 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 955 | |
f73e924c PM |
956 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
957 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
958 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 959 | }; |
f73e924c | 960 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 961 | |
fdf70832 | 962 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
963 | struct nf_conntrack_tuple *t) |
964 | { | |
df6fb868 | 965 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
966 | return -EINVAL; |
967 | ||
77236b6e PM |
968 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
969 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
970 | |
971 | return 0; | |
972 | } | |
fdf70832 | 973 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
974 | |
975 | int nf_ct_port_nlattr_tuple_size(void) | |
976 | { | |
977 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
978 | } | |
979 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
980 | #endif |
981 | ||
9fb9cbb1 | 982 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 983 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
984 | { |
985 | struct nf_conn *ct; | |
986 | enum ip_conntrack_info ctinfo; | |
987 | ||
988 | /* This ICMP is in reverse direction to the packet which caused it */ | |
989 | ct = nf_ct_get(skb, &ctinfo); | |
990 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
991 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
992 | else | |
993 | ctinfo = IP_CT_RELATED; | |
994 | ||
995 | /* Attach to new skbuff, and increment count */ | |
996 | nskb->nfct = &ct->ct_general; | |
997 | nskb->nfctinfo = ctinfo; | |
998 | nf_conntrack_get(nskb->nfct); | |
999 | } | |
1000 | ||
9fb9cbb1 | 1001 | /* Bring out ya dead! */ |
df0933dc | 1002 | static struct nf_conn * |
400dad39 | 1003 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1004 | void *data, unsigned int *bucket) |
1005 | { | |
df0933dc PM |
1006 | struct nf_conntrack_tuple_hash *h; |
1007 | struct nf_conn *ct; | |
ea781f19 | 1008 | struct hlist_nulls_node *n; |
9fb9cbb1 | 1009 | |
f8ba1aff | 1010 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1011 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
ea781f19 | 1012 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) { |
df0933dc PM |
1013 | ct = nf_ct_tuplehash_to_ctrack(h); |
1014 | if (iter(ct, data)) | |
1015 | goto found; | |
1016 | } | |
601e68e1 | 1017 | } |
ea781f19 | 1018 | hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) { |
df0933dc PM |
1019 | ct = nf_ct_tuplehash_to_ctrack(h); |
1020 | if (iter(ct, data)) | |
ec68e97d | 1021 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 1022 | } |
f8ba1aff | 1023 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
1024 | return NULL; |
1025 | found: | |
c073e3fa | 1026 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 1027 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 1028 | return ct; |
9fb9cbb1 YK |
1029 | } |
1030 | ||
400dad39 AD |
1031 | void nf_ct_iterate_cleanup(struct net *net, |
1032 | int (*iter)(struct nf_conn *i, void *data), | |
1033 | void *data) | |
9fb9cbb1 | 1034 | { |
df0933dc | 1035 | struct nf_conn *ct; |
9fb9cbb1 YK |
1036 | unsigned int bucket = 0; |
1037 | ||
400dad39 | 1038 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
1039 | /* Time to push up daises... */ |
1040 | if (del_timer(&ct->timeout)) | |
1041 | death_by_timeout((unsigned long)ct); | |
1042 | /* ... else the timer will get him soon. */ | |
1043 | ||
1044 | nf_ct_put(ct); | |
1045 | } | |
1046 | } | |
13b18339 | 1047 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1048 | |
19abb7b0 PNA |
1049 | struct __nf_ct_flush_report { |
1050 | u32 pid; | |
1051 | int report; | |
1052 | }; | |
1053 | ||
274d383b | 1054 | static int kill_report(struct nf_conn *i, void *data) |
9fb9cbb1 | 1055 | { |
19abb7b0 PNA |
1056 | struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data; |
1057 | ||
dd7669a9 PNA |
1058 | /* If we fail to deliver the event, death_by_timeout() will retry */ |
1059 | if (nf_conntrack_event_report(IPCT_DESTROY, i, | |
1060 | fr->pid, fr->report) < 0) | |
1061 | return 1; | |
1062 | ||
1063 | /* Avoid the delivery of the destroy event in death_by_timeout(). */ | |
1064 | set_bit(IPS_DYING_BIT, &i->status); | |
9fb9cbb1 YK |
1065 | return 1; |
1066 | } | |
1067 | ||
274d383b PNA |
1068 | static int kill_all(struct nf_conn *i, void *data) |
1069 | { | |
1070 | return 1; | |
1071 | } | |
1072 | ||
ea781f19 | 1073 | void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
1074 | { |
1075 | if (vmalloced) | |
1076 | vfree(hash); | |
1077 | else | |
601e68e1 | 1078 | free_pages((unsigned long)hash, |
f205c5e0 | 1079 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1080 | } |
ac565e5f | 1081 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1082 | |
274d383b | 1083 | void nf_conntrack_flush_report(struct net *net, u32 pid, int report) |
c1d10adb | 1084 | { |
19abb7b0 PNA |
1085 | struct __nf_ct_flush_report fr = { |
1086 | .pid = pid, | |
1087 | .report = report, | |
1088 | }; | |
274d383b | 1089 | nf_ct_iterate_cleanup(net, kill_report, &fr); |
c1d10adb | 1090 | } |
274d383b | 1091 | EXPORT_SYMBOL_GPL(nf_conntrack_flush_report); |
c1d10adb | 1092 | |
ee254fa4 | 1093 | static void nf_ct_release_dying_list(struct net *net) |
dd7669a9 PNA |
1094 | { |
1095 | struct nf_conntrack_tuple_hash *h; | |
1096 | struct nf_conn *ct; | |
1097 | struct hlist_nulls_node *n; | |
1098 | ||
1099 | spin_lock_bh(&nf_conntrack_lock); | |
ee254fa4 | 1100 | hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) { |
dd7669a9 PNA |
1101 | ct = nf_ct_tuplehash_to_ctrack(h); |
1102 | /* never fails to remove them, no listeners at this point */ | |
1103 | nf_ct_kill(ct); | |
1104 | } | |
1105 | spin_unlock_bh(&nf_conntrack_lock); | |
1106 | } | |
1107 | ||
08f6547d | 1108 | static void nf_conntrack_cleanup_init_net(void) |
9fb9cbb1 | 1109 | { |
08f6547d AD |
1110 | nf_conntrack_helper_fini(); |
1111 | nf_conntrack_proto_fini(); | |
1112 | kmem_cache_destroy(nf_conntrack_cachep); | |
1113 | } | |
9fb9cbb1 | 1114 | |
08f6547d AD |
1115 | static void nf_conntrack_cleanup_net(struct net *net) |
1116 | { | |
9fb9cbb1 | 1117 | i_see_dead_people: |
274d383b | 1118 | nf_ct_iterate_cleanup(net, kill_all, NULL); |
ee254fa4 | 1119 | nf_ct_release_dying_list(net); |
49ac8713 | 1120 | if (atomic_read(&net->ct.count) != 0) { |
9fb9cbb1 YK |
1121 | schedule(); |
1122 | goto i_see_dead_people; | |
1123 | } | |
6636568c PM |
1124 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1125 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1126 | schedule(); | |
9fb9cbb1 | 1127 | |
400dad39 | 1128 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1129 | nf_conntrack_htable_size); |
a0891aa6 | 1130 | nf_conntrack_ecache_fini(net); |
d716a4df | 1131 | nf_conntrack_acct_fini(net); |
9b03f38d | 1132 | nf_conntrack_expect_fini(net); |
0d55af87 | 1133 | free_percpu(net->ct.stat); |
08f6547d AD |
1134 | } |
1135 | ||
1136 | /* Mishearing the voices in his head, our hero wonders how he's | |
1137 | supposed to kill the mall. */ | |
1138 | void nf_conntrack_cleanup(struct net *net) | |
1139 | { | |
1140 | if (net_eq(net, &init_net)) | |
1141 | rcu_assign_pointer(ip_ct_attach, NULL); | |
1142 | ||
1143 | /* This makes sure all current packets have passed through | |
1144 | netfilter framework. Roll on, two-stage module | |
1145 | delete... */ | |
1146 | synchronize_net(); | |
1147 | ||
1148 | nf_conntrack_cleanup_net(net); | |
1149 | ||
1150 | if (net_eq(net, &init_net)) { | |
1151 | rcu_assign_pointer(nf_ct_destroy, NULL); | |
1152 | nf_conntrack_cleanup_init_net(); | |
1153 | } | |
9fb9cbb1 YK |
1154 | } |
1155 | ||
ea781f19 | 1156 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls) |
9fb9cbb1 | 1157 | { |
ea781f19 ED |
1158 | struct hlist_nulls_head *hash; |
1159 | unsigned int nr_slots, i; | |
1160 | size_t sz; | |
9fb9cbb1 | 1161 | |
601e68e1 | 1162 | *vmalloced = 0; |
8e5105a0 | 1163 | |
ea781f19 ED |
1164 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1165 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
1166 | sz = nr_slots * sizeof(struct hlist_nulls_head); | |
1167 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1168 | get_order(sz)); | |
601e68e1 | 1169 | if (!hash) { |
9fb9cbb1 YK |
1170 | *vmalloced = 1; |
1171 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
ea781f19 | 1172 | hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL); |
9fb9cbb1 YK |
1173 | } |
1174 | ||
ea781f19 ED |
1175 | if (hash && nulls) |
1176 | for (i = 0; i < nr_slots; i++) | |
1177 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1178 | |
1179 | return hash; | |
1180 | } | |
ac565e5f | 1181 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1182 | |
fae718dd | 1183 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1184 | { |
96eb24d7 SH |
1185 | int i, bucket, vmalloced, old_vmalloced; |
1186 | unsigned int hashsize, old_size; | |
9fb9cbb1 | 1187 | int rnd; |
ea781f19 | 1188 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 YK |
1189 | struct nf_conntrack_tuple_hash *h; |
1190 | ||
1191 | /* On boot, we can set this without any fancy locking. */ | |
1192 | if (!nf_conntrack_htable_size) | |
1193 | return param_set_uint(val, kp); | |
1194 | ||
96eb24d7 | 1195 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1196 | if (!hashsize) |
1197 | return -EINVAL; | |
1198 | ||
ea781f19 | 1199 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1); |
9fb9cbb1 YK |
1200 | if (!hash) |
1201 | return -ENOMEM; | |
1202 | ||
1203 | /* We have to rehahs for the new table anyway, so we also can | |
1204 | * use a newrandom seed */ | |
af07d241 | 1205 | get_random_bytes(&rnd, sizeof(rnd)); |
9fb9cbb1 | 1206 | |
76507f69 PM |
1207 | /* Lookups in the old hash might happen in parallel, which means we |
1208 | * might get false negatives during connection lookup. New connections | |
1209 | * created because of a false negative won't make it into the hash | |
1210 | * though since that required taking the lock. | |
1211 | */ | |
f8ba1aff | 1212 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1213 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
ea781f19 ED |
1214 | while (!hlist_nulls_empty(&init_net.ct.hash[i])) { |
1215 | h = hlist_nulls_entry(init_net.ct.hash[i].first, | |
1216 | struct nf_conntrack_tuple_hash, hnnode); | |
1217 | hlist_nulls_del_rcu(&h->hnnode); | |
9fb9cbb1 | 1218 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
ea781f19 | 1219 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1220 | } |
1221 | } | |
1222 | old_size = nf_conntrack_htable_size; | |
400dad39 AD |
1223 | old_vmalloced = init_net.ct.hash_vmalloc; |
1224 | old_hash = init_net.ct.hash; | |
9fb9cbb1 YK |
1225 | |
1226 | nf_conntrack_htable_size = hashsize; | |
400dad39 AD |
1227 | init_net.ct.hash_vmalloc = vmalloced; |
1228 | init_net.ct.hash = hash; | |
9fb9cbb1 | 1229 | nf_conntrack_hash_rnd = rnd; |
f8ba1aff | 1230 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1231 | |
ac565e5f | 1232 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1233 | return 0; |
1234 | } | |
fae718dd | 1235 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1236 | |
fae718dd | 1237 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1238 | &nf_conntrack_htable_size, 0600); |
1239 | ||
08f6547d | 1240 | static int nf_conntrack_init_init_net(void) |
9fb9cbb1 | 1241 | { |
f205c5e0 | 1242 | int max_factor = 8; |
9fb9cbb1 YK |
1243 | int ret; |
1244 | ||
1245 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1246 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1247 | if (!nf_conntrack_htable_size) { |
1248 | nf_conntrack_htable_size | |
4481374c | 1249 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1250 | / sizeof(struct hlist_head)); |
4481374c | 1251 | if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1252 | nf_conntrack_htable_size = 16384; |
1253 | if (nf_conntrack_htable_size < 32) | |
1254 | nf_conntrack_htable_size = 32; | |
1255 | ||
1256 | /* Use a max. factor of four by default to get the same max as | |
1257 | * with the old struct list_heads. When a table size is given | |
1258 | * we use the old value of 8 to avoid reducing the max. | |
1259 | * entries. */ | |
1260 | max_factor = 4; | |
9fb9cbb1 | 1261 | } |
f205c5e0 | 1262 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1263 | |
1264 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1265 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1266 | nf_conntrack_max); | |
1267 | ||
dacd2a1a YK |
1268 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1269 | sizeof(struct nf_conn), | |
ea781f19 | 1270 | 0, SLAB_DESTROY_BY_RCU, NULL); |
dacd2a1a | 1271 | if (!nf_conntrack_cachep) { |
9fb9cbb1 | 1272 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
08f6547d AD |
1273 | ret = -ENOMEM; |
1274 | goto err_cache; | |
9fb9cbb1 YK |
1275 | } |
1276 | ||
e9c1b084 PM |
1277 | ret = nf_conntrack_proto_init(); |
1278 | if (ret < 0) | |
08f6547d | 1279 | goto err_proto; |
933a41e7 | 1280 | |
ceceae1b YK |
1281 | ret = nf_conntrack_helper_init(); |
1282 | if (ret < 0) | |
08f6547d AD |
1283 | goto err_helper; |
1284 | ||
1285 | return 0; | |
1286 | ||
1287 | err_helper: | |
1288 | nf_conntrack_proto_fini(); | |
1289 | err_proto: | |
1290 | kmem_cache_destroy(nf_conntrack_cachep); | |
1291 | err_cache: | |
1292 | return ret; | |
1293 | } | |
1294 | ||
8cc20198 ED |
1295 | /* |
1296 | * We need to use special "null" values, not used in hash table | |
1297 | */ | |
1298 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1299 | #define DYING_NULLS_VAL ((1<<30)+1) | |
1300 | ||
08f6547d AD |
1301 | static int nf_conntrack_init_net(struct net *net) |
1302 | { | |
1303 | int ret; | |
ceceae1b | 1304 | |
08f6547d | 1305 | atomic_set(&net->ct.count, 0); |
8cc20198 ED |
1306 | INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL); |
1307 | INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL); | |
08f6547d AD |
1308 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1309 | if (!net->ct.stat) { | |
1310 | ret = -ENOMEM; | |
1311 | goto err_stat; | |
1312 | } | |
08f6547d | 1313 | net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
ea781f19 | 1314 | &net->ct.hash_vmalloc, 1); |
08f6547d AD |
1315 | if (!net->ct.hash) { |
1316 | ret = -ENOMEM; | |
1317 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1318 | goto err_hash; | |
1319 | } | |
1320 | ret = nf_conntrack_expect_init(net); | |
1321 | if (ret < 0) | |
1322 | goto err_expect; | |
d716a4df | 1323 | ret = nf_conntrack_acct_init(net); |
58401572 | 1324 | if (ret < 0) |
08f6547d | 1325 | goto err_acct; |
a0891aa6 PNA |
1326 | ret = nf_conntrack_ecache_init(net); |
1327 | if (ret < 0) | |
1328 | goto err_ecache; | |
7d3cdc6b | 1329 | |
9fb9cbb1 YK |
1330 | /* Set up fake conntrack: |
1331 | - to never be deleted, not in any hashes */ | |
5a1fb391 AD |
1332 | #ifdef CONFIG_NET_NS |
1333 | nf_conntrack_untracked.ct_net = &init_net; | |
1334 | #endif | |
9fb9cbb1 YK |
1335 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); |
1336 | /* - and look it like as a confirmed connection */ | |
1337 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1338 | ||
08f6547d | 1339 | return 0; |
9fb9cbb1 | 1340 | |
a0891aa6 PNA |
1341 | err_ecache: |
1342 | nf_conntrack_acct_fini(net); | |
08f6547d | 1343 | err_acct: |
9b03f38d | 1344 | nf_conntrack_expect_fini(net); |
08f6547d | 1345 | err_expect: |
400dad39 | 1346 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1347 | nf_conntrack_htable_size); |
6058fa6b | 1348 | err_hash: |
0d55af87 AD |
1349 | free_percpu(net->ct.stat); |
1350 | err_stat: | |
08f6547d AD |
1351 | return ret; |
1352 | } | |
1353 | ||
1354 | int nf_conntrack_init(struct net *net) | |
1355 | { | |
1356 | int ret; | |
1357 | ||
1358 | if (net_eq(net, &init_net)) { | |
1359 | ret = nf_conntrack_init_init_net(); | |
1360 | if (ret < 0) | |
1361 | goto out_init_net; | |
1362 | } | |
1363 | ret = nf_conntrack_init_net(net); | |
1364 | if (ret < 0) | |
1365 | goto out_net; | |
1366 | ||
1367 | if (net_eq(net, &init_net)) { | |
1368 | /* For use by REJECT target */ | |
1369 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); | |
1370 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); | |
1371 | } | |
1372 | return 0; | |
1373 | ||
1374 | out_net: | |
1375 | if (net_eq(net, &init_net)) | |
1376 | nf_conntrack_cleanup_init_net(); | |
1377 | out_init_net: | |
1378 | return ret; | |
9fb9cbb1 | 1379 | } |