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 | ||
9fb9cbb1 YK |
66 | static int nf_conntrack_hash_rnd_initted; |
67 | static unsigned int nf_conntrack_hash_rnd; | |
68 | ||
69 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
70 | unsigned int size, unsigned int rnd) | |
71 | { | |
0794935e PM |
72 | unsigned int n; |
73 | u_int32_t h; | |
74 | ||
75 | /* The direction must be ignored, so we hash everything up to the | |
76 | * destination ports (which is a multiple of 4) and treat the last | |
77 | * three bytes manually. | |
78 | */ | |
79 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
80 | h = jhash2((u32 *)tuple, n, | |
81 | rnd ^ (((__force __u16)tuple->dst.u.all << 16) | | |
82 | tuple->dst.protonum)); | |
83 | ||
84 | return ((u64)h * size) >> 32; | |
9fb9cbb1 YK |
85 | } |
86 | ||
87 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
88 | { | |
89 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
90 | nf_conntrack_hash_rnd); | |
91 | } | |
92 | ||
5f2b4c90 | 93 | bool |
9fb9cbb1 YK |
94 | nf_ct_get_tuple(const struct sk_buff *skb, |
95 | unsigned int nhoff, | |
96 | unsigned int dataoff, | |
97 | u_int16_t l3num, | |
98 | u_int8_t protonum, | |
99 | struct nf_conntrack_tuple *tuple, | |
100 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 101 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 102 | { |
443a70d5 | 103 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
104 | |
105 | tuple->src.l3num = l3num; | |
106 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 107 | return false; |
9fb9cbb1 YK |
108 | |
109 | tuple->dst.protonum = protonum; | |
110 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
111 | ||
605dcad6 | 112 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 113 | } |
13b18339 | 114 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 115 | |
5f2b4c90 JE |
116 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
117 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
118 | { |
119 | struct nf_conntrack_l3proto *l3proto; | |
120 | struct nf_conntrack_l4proto *l4proto; | |
121 | unsigned int protoff; | |
122 | u_int8_t protonum; | |
123 | int ret; | |
124 | ||
125 | rcu_read_lock(); | |
126 | ||
127 | l3proto = __nf_ct_l3proto_find(l3num); | |
128 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
129 | if (ret != NF_ACCEPT) { | |
130 | rcu_read_unlock(); | |
5f2b4c90 | 131 | return false; |
e2a3123f YK |
132 | } |
133 | ||
134 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
135 | ||
136 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
137 | l3proto, l4proto); | |
138 | ||
139 | rcu_read_unlock(); | |
140 | return ret; | |
141 | } | |
142 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
143 | ||
5f2b4c90 | 144 | bool |
9fb9cbb1 YK |
145 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
146 | const struct nf_conntrack_tuple *orig, | |
147 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 148 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 149 | { |
443a70d5 | 150 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
151 | |
152 | inverse->src.l3num = orig->src.l3num; | |
153 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 154 | return false; |
9fb9cbb1 YK |
155 | |
156 | inverse->dst.dir = !orig->dst.dir; | |
157 | ||
158 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 159 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 160 | } |
13b18339 | 161 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 162 | |
9fb9cbb1 YK |
163 | static void |
164 | clean_from_lists(struct nf_conn *ct) | |
165 | { | |
0d53778e | 166 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
167 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
168 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
169 | |
170 | /* Destroy all pending expectations */ | |
c1d10adb | 171 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
172 | } |
173 | ||
174 | static void | |
175 | destroy_conntrack(struct nf_conntrack *nfct) | |
176 | { | |
177 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
0d55af87 | 178 | struct net *net = nf_ct_net(ct); |
605dcad6 | 179 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 180 | |
0d53778e | 181 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
182 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
183 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
184 | ||
9fb9cbb1 YK |
185 | /* To make sure we don't get any weird locking issues here: |
186 | * destroy_conntrack() MUST NOT be called with a write lock | |
187 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 188 | rcu_read_lock(); |
5e8fbe2a | 189 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
190 | if (l4proto && l4proto->destroy) |
191 | l4proto->destroy(ct); | |
9fb9cbb1 | 192 | |
982d9a9c | 193 | rcu_read_unlock(); |
9fb9cbb1 | 194 | |
f8ba1aff | 195 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
196 | /* Expectations will have been removed in clean_from_lists, |
197 | * except TFTP can create an expectation on the first packet, | |
198 | * before connection is in the list, so we need to clean here, | |
199 | * too. */ | |
c1d10adb | 200 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
201 | |
202 | /* We overload first tuple to link into unconfirmed list. */ | |
203 | if (!nf_ct_is_confirmed(ct)) { | |
ea781f19 ED |
204 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); |
205 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
9fb9cbb1 YK |
206 | } |
207 | ||
0d55af87 | 208 | NF_CT_STAT_INC(net, delete); |
f8ba1aff | 209 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
210 | |
211 | if (ct->master) | |
212 | nf_ct_put(ct->master); | |
213 | ||
0d53778e | 214 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
215 | nf_conntrack_free(ct); |
216 | } | |
217 | ||
dd7669a9 | 218 | void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 219 | { |
0d55af87 | 220 | struct net *net = nf_ct_net(ct); |
9fb9cbb1 | 221 | |
9858a3ae | 222 | nf_ct_helper_destroy(ct); |
f8ba1aff | 223 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
224 | /* Inside lock so preempt is disabled on module removal path. |
225 | * Otherwise we can get spurious warnings. */ | |
0d55af87 | 226 | NF_CT_STAT_INC(net, delete_list); |
9fb9cbb1 | 227 | clean_from_lists(ct); |
f8ba1aff | 228 | spin_unlock_bh(&nf_conntrack_lock); |
dd7669a9 PNA |
229 | } |
230 | EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists); | |
231 | ||
232 | static void death_by_event(unsigned long ul_conntrack) | |
233 | { | |
234 | struct nf_conn *ct = (void *)ul_conntrack; | |
235 | struct net *net = nf_ct_net(ct); | |
236 | ||
237 | if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) { | |
238 | /* bad luck, let's retry again */ | |
239 | ct->timeout.expires = jiffies + | |
240 | (random32() % net->ct.sysctl_events_retry_timeout); | |
241 | add_timer(&ct->timeout); | |
242 | return; | |
243 | } | |
244 | /* we've got the event delivered, now it's dying */ | |
245 | set_bit(IPS_DYING_BIT, &ct->status); | |
246 | spin_lock(&nf_conntrack_lock); | |
247 | hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
248 | spin_unlock(&nf_conntrack_lock); | |
249 | nf_ct_put(ct); | |
250 | } | |
251 | ||
252 | void nf_ct_insert_dying_list(struct nf_conn *ct) | |
253 | { | |
254 | struct net *net = nf_ct_net(ct); | |
255 | ||
256 | /* add this conntrack to the dying list */ | |
257 | spin_lock_bh(&nf_conntrack_lock); | |
258 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
259 | &net->ct.dying); | |
260 | spin_unlock_bh(&nf_conntrack_lock); | |
261 | /* set a new timer to retry event delivery */ | |
262 | setup_timer(&ct->timeout, death_by_event, (unsigned long)ct); | |
263 | ct->timeout.expires = jiffies + | |
264 | (random32() % net->ct.sysctl_events_retry_timeout); | |
265 | add_timer(&ct->timeout); | |
266 | } | |
267 | EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list); | |
268 | ||
269 | static void death_by_timeout(unsigned long ul_conntrack) | |
270 | { | |
271 | struct nf_conn *ct = (void *)ul_conntrack; | |
272 | ||
273 | if (!test_bit(IPS_DYING_BIT, &ct->status) && | |
274 | unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) { | |
275 | /* destroy event was not delivered */ | |
276 | nf_ct_delete_from_lists(ct); | |
277 | nf_ct_insert_dying_list(ct); | |
278 | return; | |
279 | } | |
280 | set_bit(IPS_DYING_BIT, &ct->status); | |
281 | nf_ct_delete_from_lists(ct); | |
9fb9cbb1 YK |
282 | nf_ct_put(ct); |
283 | } | |
284 | ||
ea781f19 ED |
285 | /* |
286 | * Warning : | |
287 | * - Caller must take a reference on returned object | |
288 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
289 | * OR | |
290 | * - Caller must lock nf_conntrack_lock before calling this function | |
291 | */ | |
c1d10adb | 292 | struct nf_conntrack_tuple_hash * |
400dad39 | 293 | __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
294 | { |
295 | struct nf_conntrack_tuple_hash *h; | |
ea781f19 | 296 | struct hlist_nulls_node *n; |
9fb9cbb1 YK |
297 | unsigned int hash = hash_conntrack(tuple); |
298 | ||
4e29e9ec PM |
299 | /* Disable BHs the entire time since we normally need to disable them |
300 | * at least once for the stats anyway. | |
301 | */ | |
302 | local_bh_disable(); | |
ea781f19 ED |
303 | begin: |
304 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { | |
ba419aff | 305 | if (nf_ct_tuple_equal(tuple, &h->tuple)) { |
0d55af87 | 306 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 307 | local_bh_enable(); |
9fb9cbb1 YK |
308 | return h; |
309 | } | |
0d55af87 | 310 | NF_CT_STAT_INC(net, searched); |
9fb9cbb1 | 311 | } |
ea781f19 ED |
312 | /* |
313 | * if the nulls value we got at the end of this lookup is | |
314 | * not the expected one, we must restart lookup. | |
315 | * We probably met an item that was moved to another chain. | |
316 | */ | |
317 | if (get_nulls_value(n) != hash) | |
318 | goto begin; | |
4e29e9ec | 319 | local_bh_enable(); |
9fb9cbb1 YK |
320 | |
321 | return NULL; | |
322 | } | |
13b18339 | 323 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
324 | |
325 | /* Find a connection corresponding to a tuple. */ | |
326 | struct nf_conntrack_tuple_hash * | |
400dad39 | 327 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
328 | { |
329 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 330 | struct nf_conn *ct; |
9fb9cbb1 | 331 | |
76507f69 | 332 | rcu_read_lock(); |
ea781f19 | 333 | begin: |
400dad39 | 334 | h = __nf_conntrack_find(net, tuple); |
76507f69 PM |
335 | if (h) { |
336 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
337 | if (unlikely(nf_ct_is_dying(ct) || |
338 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 339 | h = NULL; |
ea781f19 ED |
340 | else { |
341 | if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) { | |
342 | nf_ct_put(ct); | |
343 | goto begin; | |
344 | } | |
345 | } | |
76507f69 PM |
346 | } |
347 | rcu_read_unlock(); | |
9fb9cbb1 YK |
348 | |
349 | return h; | |
350 | } | |
13b18339 | 351 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 352 | |
c1d10adb PNA |
353 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
354 | unsigned int hash, | |
601e68e1 | 355 | unsigned int repl_hash) |
c1d10adb | 356 | { |
400dad39 AD |
357 | struct net *net = nf_ct_net(ct); |
358 | ||
ea781f19 | 359 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
400dad39 | 360 | &net->ct.hash[hash]); |
ea781f19 | 361 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
400dad39 | 362 | &net->ct.hash[repl_hash]); |
c1d10adb PNA |
363 | } |
364 | ||
365 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
366 | { | |
367 | unsigned int hash, repl_hash; | |
368 | ||
369 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
370 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
371 | ||
c1d10adb | 372 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
c1d10adb | 373 | } |
13b18339 | 374 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 375 | |
9fb9cbb1 YK |
376 | /* Confirm a connection given skb; places it in hash table */ |
377 | int | |
3db05fea | 378 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
379 | { |
380 | unsigned int hash, repl_hash; | |
df0933dc | 381 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 382 | struct nf_conn *ct; |
df0933dc | 383 | struct nf_conn_help *help; |
ea781f19 | 384 | struct hlist_nulls_node *n; |
9fb9cbb1 | 385 | enum ip_conntrack_info ctinfo; |
400dad39 | 386 | struct net *net; |
9fb9cbb1 | 387 | |
3db05fea | 388 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 389 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
390 | |
391 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
392 | ICMP/TCP RST packets in other direction. Actual packet | |
393 | which created connection will be IP_CT_NEW or for an | |
394 | expected connection, IP_CT_RELATED. */ | |
395 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
396 | return NF_ACCEPT; | |
397 | ||
398 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
399 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
400 | ||
401 | /* We're not in hash table, and we refuse to set up related | |
402 | connections for unconfirmed conns. But packet copies and | |
403 | REJECT will give spurious warnings here. */ | |
404 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
405 | ||
406 | /* No external references means noone else could have | |
407 | confirmed us. */ | |
408 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 409 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 410 | |
f8ba1aff | 411 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
412 | |
413 | /* See if there's one in the list already, including reverse: | |
414 | NAT could have grabbed it without realizing, since we're | |
415 | not in the hash. If there is, we lost race. */ | |
ea781f19 | 416 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) |
df0933dc PM |
417 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
418 | &h->tuple)) | |
419 | goto out; | |
ea781f19 | 420 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode) |
df0933dc PM |
421 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
422 | &h->tuple)) | |
423 | goto out; | |
9fb9cbb1 | 424 | |
df0933dc | 425 | /* Remove from unconfirmed list */ |
ea781f19 | 426 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
df0933dc | 427 | |
df0933dc PM |
428 | /* Timer relative to confirmation time, not original |
429 | setting time, otherwise we'd get timer wrap in | |
430 | weird delay cases. */ | |
431 | ct->timeout.expires += jiffies; | |
432 | add_timer(&ct->timeout); | |
433 | atomic_inc(&ct->ct_general.use); | |
434 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
5c8ec910 PM |
435 | |
436 | /* Since the lookup is lockless, hash insertion must be done after | |
437 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
438 | * guarantee that no other CPU can find the conntrack before the above | |
439 | * stores are visible. | |
440 | */ | |
441 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
0d55af87 | 442 | NF_CT_STAT_INC(net, insert); |
f8ba1aff | 443 | spin_unlock_bh(&nf_conntrack_lock); |
5c8ec910 | 444 | |
df0933dc PM |
445 | help = nfct_help(ct); |
446 | if (help && help->helper) | |
a71996fc | 447 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 448 | |
df0933dc | 449 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 450 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 451 | return NF_ACCEPT; |
9fb9cbb1 | 452 | |
df0933dc | 453 | out: |
0d55af87 | 454 | NF_CT_STAT_INC(net, insert_failed); |
f8ba1aff | 455 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
456 | return NF_DROP; |
457 | } | |
13b18339 | 458 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
459 | |
460 | /* Returns true if a connection correspondings to the tuple (required | |
461 | for NAT). */ | |
462 | int | |
463 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
464 | const struct nf_conn *ignored_conntrack) | |
465 | { | |
400dad39 | 466 | struct net *net = nf_ct_net(ignored_conntrack); |
9fb9cbb1 | 467 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 468 | struct hlist_nulls_node *n; |
ba419aff | 469 | unsigned int hash = hash_conntrack(tuple); |
9fb9cbb1 | 470 | |
4e29e9ec PM |
471 | /* Disable BHs the entire time since we need to disable them at |
472 | * least once for the stats anyway. | |
473 | */ | |
474 | rcu_read_lock_bh(); | |
ea781f19 | 475 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { |
ba419aff PM |
476 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
477 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
0d55af87 | 478 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 479 | rcu_read_unlock_bh(); |
ba419aff PM |
480 | return 1; |
481 | } | |
0d55af87 | 482 | NF_CT_STAT_INC(net, searched); |
ba419aff | 483 | } |
4e29e9ec | 484 | rcu_read_unlock_bh(); |
9fb9cbb1 | 485 | |
ba419aff | 486 | return 0; |
9fb9cbb1 | 487 | } |
13b18339 | 488 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 489 | |
7ae7730f PM |
490 | #define NF_CT_EVICTION_RANGE 8 |
491 | ||
9fb9cbb1 YK |
492 | /* There's a small race here where we may free a just-assured |
493 | connection. Too bad: we're in trouble anyway. */ | |
400dad39 | 494 | static noinline int early_drop(struct net *net, unsigned int hash) |
9fb9cbb1 | 495 | { |
f205c5e0 | 496 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 497 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 498 | struct nf_conn *ct = NULL, *tmp; |
ea781f19 | 499 | struct hlist_nulls_node *n; |
7ae7730f | 500 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
501 | int dropped = 0; |
502 | ||
76507f69 | 503 | rcu_read_lock(); |
7ae7730f | 504 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
ea781f19 ED |
505 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], |
506 | hnnode) { | |
7ae7730f PM |
507 | tmp = nf_ct_tuplehash_to_ctrack(h); |
508 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
509 | ct = tmp; | |
510 | cnt++; | |
511 | } | |
76507f69 | 512 | |
5ae27aa2 CG |
513 | if (ct != NULL) { |
514 | if (likely(!nf_ct_is_dying(ct) && | |
515 | atomic_inc_not_zero(&ct->ct_general.use))) | |
516 | break; | |
517 | else | |
518 | ct = NULL; | |
519 | } | |
520 | ||
521 | if (cnt >= NF_CT_EVICTION_RANGE) | |
7ae7730f | 522 | break; |
5ae27aa2 | 523 | |
7ae7730f | 524 | hash = (hash + 1) % nf_conntrack_htable_size; |
9fb9cbb1 | 525 | } |
76507f69 | 526 | rcu_read_unlock(); |
9fb9cbb1 YK |
527 | |
528 | if (!ct) | |
529 | return dropped; | |
530 | ||
531 | if (del_timer(&ct->timeout)) { | |
532 | death_by_timeout((unsigned long)ct); | |
533 | dropped = 1; | |
0d55af87 | 534 | NF_CT_STAT_INC_ATOMIC(net, early_drop); |
9fb9cbb1 YK |
535 | } |
536 | nf_ct_put(ct); | |
537 | return dropped; | |
538 | } | |
539 | ||
5a1fb391 AD |
540 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
541 | const struct nf_conntrack_tuple *orig, | |
b891c5a8 PNA |
542 | const struct nf_conntrack_tuple *repl, |
543 | gfp_t gfp) | |
9fb9cbb1 | 544 | { |
cd7fcbf1 | 545 | struct nf_conn *ct; |
9fb9cbb1 | 546 | |
dc808fe2 | 547 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
af07d241 HPP |
548 | get_random_bytes(&nf_conntrack_hash_rnd, |
549 | sizeof(nf_conntrack_hash_rnd)); | |
9fb9cbb1 YK |
550 | nf_conntrack_hash_rnd_initted = 1; |
551 | } | |
552 | ||
5251e2d2 | 553 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 554 | atomic_inc(&net->ct.count); |
5251e2d2 | 555 | |
76eb9460 | 556 | if (nf_conntrack_max && |
49ac8713 | 557 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
9fb9cbb1 | 558 | unsigned int hash = hash_conntrack(orig); |
400dad39 | 559 | if (!early_drop(net, hash)) { |
49ac8713 | 560 | atomic_dec(&net->ct.count); |
9fb9cbb1 YK |
561 | if (net_ratelimit()) |
562 | printk(KERN_WARNING | |
563 | "nf_conntrack: table full, dropping" | |
564 | " packet.\n"); | |
565 | return ERR_PTR(-ENOMEM); | |
566 | } | |
567 | } | |
568 | ||
941297f4 ED |
569 | /* |
570 | * Do not use kmem_cache_zalloc(), as this cache uses | |
571 | * SLAB_DESTROY_BY_RCU. | |
572 | */ | |
5b3501fa | 573 | ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp); |
c88130bc | 574 | if (ct == NULL) { |
0d53778e | 575 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
49ac8713 | 576 | atomic_dec(&net->ct.count); |
dacd2a1a | 577 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 578 | } |
941297f4 ED |
579 | /* |
580 | * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next | |
581 | * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged. | |
582 | */ | |
583 | memset(&ct->tuplehash[IP_CT_DIR_MAX], 0, | |
584 | sizeof(*ct) - offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX])); | |
440f0d58 | 585 | spin_lock_init(&ct->lock); |
c88130bc | 586 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 587 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 588 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
941297f4 | 589 | ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev = NULL; |
9fb9cbb1 | 590 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 591 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
5a1fb391 AD |
592 | #ifdef CONFIG_NET_NS |
593 | ct->ct_net = net; | |
594 | #endif | |
9fb9cbb1 | 595 | |
941297f4 ED |
596 | /* |
597 | * changes to lookup keys must be done before setting refcnt to 1 | |
598 | */ | |
599 | smp_wmb(); | |
600 | atomic_set(&ct->ct_general.use, 1); | |
c88130bc | 601 | return ct; |
9fb9cbb1 | 602 | } |
13b18339 | 603 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 604 | |
c88130bc | 605 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 606 | { |
1d45209d ED |
607 | struct net *net = nf_ct_net(ct); |
608 | ||
ceeff754 | 609 | nf_ct_ext_destroy(ct); |
1d45209d | 610 | atomic_dec(&net->ct.count); |
ea781f19 | 611 | nf_ct_ext_free(ct); |
5b3501fa | 612 | kmem_cache_free(net->ct.nf_conntrack_cachep, ct); |
76507f69 | 613 | } |
13b18339 | 614 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
615 | |
616 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
617 | failed due to stress. Otherwise it really is unclassifiable. */ | |
618 | static struct nf_conntrack_tuple_hash * | |
5a1fb391 AD |
619 | init_conntrack(struct net *net, |
620 | const struct nf_conntrack_tuple *tuple, | |
9fb9cbb1 | 621 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 622 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
623 | struct sk_buff *skb, |
624 | unsigned int dataoff) | |
625 | { | |
c88130bc | 626 | struct nf_conn *ct; |
3c158f7f | 627 | struct nf_conn_help *help; |
9fb9cbb1 YK |
628 | struct nf_conntrack_tuple repl_tuple; |
629 | struct nf_conntrack_expect *exp; | |
630 | ||
605dcad6 | 631 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 632 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
633 | return NULL; |
634 | } | |
635 | ||
5a1fb391 | 636 | ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC); |
cd7fcbf1 | 637 | if (IS_ERR(ct)) { |
0d53778e | 638 | pr_debug("Can't allocate conntrack.\n"); |
c88130bc | 639 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 YK |
640 | } |
641 | ||
c88130bc PM |
642 | if (!l4proto->new(ct, skb, dataoff)) { |
643 | nf_conntrack_free(ct); | |
0d53778e | 644 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
645 | return NULL; |
646 | } | |
647 | ||
58401572 | 648 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a0891aa6 | 649 | nf_ct_ecache_ext_add(ct, GFP_ATOMIC); |
58401572 | 650 | |
f8ba1aff | 651 | spin_lock_bh(&nf_conntrack_lock); |
9b03f38d | 652 | exp = nf_ct_find_expectation(net, tuple); |
9fb9cbb1 | 653 | if (exp) { |
0d53778e | 654 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 655 | ct, exp); |
9fb9cbb1 | 656 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
657 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
658 | ct->master = exp->master; | |
ceceae1b | 659 | if (exp->helper) { |
c88130bc | 660 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b YK |
661 | if (help) |
662 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
663 | } |
664 | ||
9fb9cbb1 | 665 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 666 | ct->mark = exp->master->mark; |
7c9728c3 JM |
667 | #endif |
668 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 669 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 670 | #endif |
c88130bc | 671 | nf_conntrack_get(&ct->master->ct_general); |
0d55af87 | 672 | NF_CT_STAT_INC(net, expect_new); |
22e7410b | 673 | } else { |
226c0c0e | 674 | __nf_ct_try_assign_helper(ct, GFP_ATOMIC); |
0d55af87 | 675 | NF_CT_STAT_INC(net, new); |
22e7410b | 676 | } |
9fb9cbb1 YK |
677 | |
678 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
ea781f19 | 679 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
63c9a262 | 680 | &net->ct.unconfirmed); |
9fb9cbb1 | 681 | |
f8ba1aff | 682 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
683 | |
684 | if (exp) { | |
685 | if (exp->expectfn) | |
c88130bc | 686 | exp->expectfn(ct, exp); |
6823645d | 687 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
688 | } |
689 | ||
c88130bc | 690 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
691 | } |
692 | ||
693 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
694 | static inline struct nf_conn * | |
a702a65f AD |
695 | resolve_normal_ct(struct net *net, |
696 | struct sk_buff *skb, | |
9fb9cbb1 YK |
697 | unsigned int dataoff, |
698 | u_int16_t l3num, | |
699 | u_int8_t protonum, | |
700 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 701 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
702 | int *set_reply, |
703 | enum ip_conntrack_info *ctinfo) | |
704 | { | |
705 | struct nf_conntrack_tuple tuple; | |
706 | struct nf_conntrack_tuple_hash *h; | |
707 | struct nf_conn *ct; | |
708 | ||
bbe735e4 | 709 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 710 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 711 | l4proto)) { |
0d53778e | 712 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
713 | return NULL; |
714 | } | |
715 | ||
716 | /* look for tuple match */ | |
a702a65f | 717 | h = nf_conntrack_find_get(net, &tuple); |
9fb9cbb1 | 718 | if (!h) { |
a702a65f | 719 | h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
720 | if (!h) |
721 | return NULL; | |
722 | if (IS_ERR(h)) | |
723 | return (void *)h; | |
724 | } | |
725 | ct = nf_ct_tuplehash_to_ctrack(h); | |
726 | ||
727 | /* It exists; we have (non-exclusive) reference. */ | |
728 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
729 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
730 | /* Please set reply bit if this packet OK */ | |
731 | *set_reply = 1; | |
732 | } else { | |
733 | /* Once we've had two way comms, always ESTABLISHED. */ | |
734 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 735 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
736 | *ctinfo = IP_CT_ESTABLISHED; |
737 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
738 | pr_debug("nf_conntrack_in: related packet for %p\n", |
739 | ct); | |
9fb9cbb1 YK |
740 | *ctinfo = IP_CT_RELATED; |
741 | } else { | |
0d53778e | 742 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
743 | *ctinfo = IP_CT_NEW; |
744 | } | |
745 | *set_reply = 0; | |
746 | } | |
747 | skb->nfct = &ct->ct_general; | |
748 | skb->nfctinfo = *ctinfo; | |
749 | return ct; | |
750 | } | |
751 | ||
752 | unsigned int | |
a702a65f AD |
753 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
754 | struct sk_buff *skb) | |
9fb9cbb1 YK |
755 | { |
756 | struct nf_conn *ct; | |
757 | enum ip_conntrack_info ctinfo; | |
758 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 759 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
760 | unsigned int dataoff; |
761 | u_int8_t protonum; | |
762 | int set_reply = 0; | |
763 | int ret; | |
764 | ||
765 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 766 | if (skb->nfct) { |
0d55af87 | 767 | NF_CT_STAT_INC_ATOMIC(net, ignore); |
9fb9cbb1 YK |
768 | return NF_ACCEPT; |
769 | } | |
770 | ||
923f4902 | 771 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 772 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 773 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
774 | &dataoff, &protonum); |
775 | if (ret <= 0) { | |
0d53778e | 776 | pr_debug("not prepared to track yet or error occured\n"); |
0d55af87 AD |
777 | NF_CT_STAT_INC_ATOMIC(net, error); |
778 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
9fb9cbb1 YK |
779 | return -ret; |
780 | } | |
781 | ||
76108cea | 782 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
783 | |
784 | /* It may be an special packet, error, unclean... | |
785 | * inverse of the return code tells to the netfilter | |
786 | * core what to do with the packet. */ | |
74c51a14 AD |
787 | if (l4proto->error != NULL) { |
788 | ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum); | |
789 | if (ret <= 0) { | |
0d55af87 AD |
790 | NF_CT_STAT_INC_ATOMIC(net, error); |
791 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
74c51a14 AD |
792 | return -ret; |
793 | } | |
9fb9cbb1 YK |
794 | } |
795 | ||
a702a65f AD |
796 | ct = resolve_normal_ct(net, skb, dataoff, pf, protonum, |
797 | l3proto, l4proto, &set_reply, &ctinfo); | |
9fb9cbb1 YK |
798 | if (!ct) { |
799 | /* Not valid part of a connection */ | |
0d55af87 | 800 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
9fb9cbb1 YK |
801 | return NF_ACCEPT; |
802 | } | |
803 | ||
804 | if (IS_ERR(ct)) { | |
805 | /* Too stressed to deal. */ | |
0d55af87 | 806 | NF_CT_STAT_INC_ATOMIC(net, drop); |
9fb9cbb1 YK |
807 | return NF_DROP; |
808 | } | |
809 | ||
3db05fea | 810 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 811 | |
3db05fea | 812 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
ec8d5409 | 813 | if (ret <= 0) { |
9fb9cbb1 YK |
814 | /* Invalid: inverse of the return code tells |
815 | * the netfilter core what to do */ | |
0d53778e | 816 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
817 | nf_conntrack_put(skb->nfct); |
818 | skb->nfct = NULL; | |
0d55af87 | 819 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
820 | if (ret == -NF_DROP) |
821 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
9fb9cbb1 YK |
822 | return -ret; |
823 | } | |
824 | ||
825 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
a71996fc | 826 | nf_conntrack_event_cache(IPCT_STATUS, ct); |
9fb9cbb1 YK |
827 | |
828 | return ret; | |
829 | } | |
13b18339 | 830 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 831 | |
5f2b4c90 JE |
832 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
833 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 834 | { |
5f2b4c90 | 835 | bool ret; |
923f4902 PM |
836 | |
837 | rcu_read_lock(); | |
838 | ret = nf_ct_invert_tuple(inverse, orig, | |
839 | __nf_ct_l3proto_find(orig->src.l3num), | |
840 | __nf_ct_l4proto_find(orig->src.l3num, | |
841 | orig->dst.protonum)); | |
842 | rcu_read_unlock(); | |
843 | return ret; | |
9fb9cbb1 | 844 | } |
13b18339 | 845 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 846 | |
5b1158e9 JK |
847 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
848 | implicitly racy: see __nf_conntrack_confirm */ | |
849 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
850 | const struct nf_conntrack_tuple *newreply) | |
851 | { | |
852 | struct nf_conn_help *help = nfct_help(ct); | |
853 | ||
5b1158e9 JK |
854 | /* Should be unconfirmed, so not in hash table yet */ |
855 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
856 | ||
0d53778e | 857 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 858 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
859 | |
860 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 861 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 862 | return; |
ceceae1b | 863 | |
c52fbb41 | 864 | rcu_read_lock(); |
226c0c0e | 865 | __nf_ct_try_assign_helper(ct, GFP_ATOMIC); |
c52fbb41 | 866 | rcu_read_unlock(); |
5b1158e9 | 867 | } |
13b18339 | 868 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 869 | |
9fb9cbb1 YK |
870 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
871 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
872 | enum ip_conntrack_info ctinfo, | |
873 | const struct sk_buff *skb, | |
874 | unsigned long extra_jiffies, | |
875 | int do_acct) | |
876 | { | |
9fb9cbb1 YK |
877 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); |
878 | NF_CT_ASSERT(skb); | |
879 | ||
997ae831 | 880 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
881 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
882 | goto acct; | |
997ae831 | 883 | |
9fb9cbb1 YK |
884 | /* If not in hash table, timer will not be active yet */ |
885 | if (!nf_ct_is_confirmed(ct)) { | |
886 | ct->timeout.expires = extra_jiffies; | |
9fb9cbb1 | 887 | } else { |
be00c8e4 MJ |
888 | unsigned long newtime = jiffies + extra_jiffies; |
889 | ||
890 | /* Only update the timeout if the new timeout is at least | |
891 | HZ jiffies from the old timeout. Need del_timer for race | |
892 | avoidance (may already be dying). */ | |
65cb9fda PM |
893 | if (newtime - ct->timeout.expires >= HZ) |
894 | mod_timer_pending(&ct->timeout, newtime); | |
9fb9cbb1 YK |
895 | } |
896 | ||
47d95045 | 897 | acct: |
9fb9cbb1 | 898 | if (do_acct) { |
58401572 | 899 | struct nf_conn_counter *acct; |
3ffd5eeb | 900 | |
58401572 KPO |
901 | acct = nf_conn_acct_find(ct); |
902 | if (acct) { | |
65cb9fda | 903 | spin_lock_bh(&ct->lock); |
58401572 KPO |
904 | acct[CTINFO2DIR(ctinfo)].packets++; |
905 | acct[CTINFO2DIR(ctinfo)].bytes += | |
906 | skb->len - skb_network_offset(skb); | |
65cb9fda | 907 | spin_unlock_bh(&ct->lock); |
58401572 | 908 | } |
9fb9cbb1 | 909 | } |
9fb9cbb1 | 910 | } |
13b18339 | 911 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 912 | |
4c889498 DM |
913 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
914 | enum ip_conntrack_info ctinfo, | |
915 | const struct sk_buff *skb, | |
916 | int do_acct) | |
51091764 | 917 | { |
718d4ad9 | 918 | if (do_acct) { |
58401572 KPO |
919 | struct nf_conn_counter *acct; |
920 | ||
58401572 KPO |
921 | acct = nf_conn_acct_find(ct); |
922 | if (acct) { | |
65cb9fda | 923 | spin_lock_bh(&ct->lock); |
58401572 KPO |
924 | acct[CTINFO2DIR(ctinfo)].packets++; |
925 | acct[CTINFO2DIR(ctinfo)].bytes += | |
926 | skb->len - skb_network_offset(skb); | |
65cb9fda | 927 | spin_unlock_bh(&ct->lock); |
58401572 | 928 | } |
718d4ad9 | 929 | } |
58401572 | 930 | |
4c889498 | 931 | if (del_timer(&ct->timeout)) { |
51091764 | 932 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
933 | return true; |
934 | } | |
935 | return false; | |
51091764 | 936 | } |
718d4ad9 | 937 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 938 | |
e281db5c | 939 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
940 | |
941 | #include <linux/netfilter/nfnetlink.h> | |
942 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
943 | #include <linux/mutex.h> |
944 | ||
c1d10adb PNA |
945 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
946 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
947 | * or depend on ctnetlink */ | |
fdf70832 | 948 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
949 | const struct nf_conntrack_tuple *tuple) |
950 | { | |
77236b6e PM |
951 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
952 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
953 | return 0; |
954 | ||
df6fb868 | 955 | nla_put_failure: |
c1d10adb PNA |
956 | return -1; |
957 | } | |
fdf70832 | 958 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 959 | |
f73e924c PM |
960 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
961 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
962 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 963 | }; |
f73e924c | 964 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 965 | |
fdf70832 | 966 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
967 | struct nf_conntrack_tuple *t) |
968 | { | |
df6fb868 | 969 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
970 | return -EINVAL; |
971 | ||
77236b6e PM |
972 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
973 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
974 | |
975 | return 0; | |
976 | } | |
fdf70832 | 977 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
978 | |
979 | int nf_ct_port_nlattr_tuple_size(void) | |
980 | { | |
981 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
982 | } | |
983 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
984 | #endif |
985 | ||
9fb9cbb1 | 986 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 987 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
988 | { |
989 | struct nf_conn *ct; | |
990 | enum ip_conntrack_info ctinfo; | |
991 | ||
992 | /* This ICMP is in reverse direction to the packet which caused it */ | |
993 | ct = nf_ct_get(skb, &ctinfo); | |
994 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
995 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
996 | else | |
997 | ctinfo = IP_CT_RELATED; | |
998 | ||
999 | /* Attach to new skbuff, and increment count */ | |
1000 | nskb->nfct = &ct->ct_general; | |
1001 | nskb->nfctinfo = ctinfo; | |
1002 | nf_conntrack_get(nskb->nfct); | |
1003 | } | |
1004 | ||
9fb9cbb1 | 1005 | /* Bring out ya dead! */ |
df0933dc | 1006 | static struct nf_conn * |
400dad39 | 1007 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1008 | void *data, unsigned int *bucket) |
1009 | { | |
df0933dc PM |
1010 | struct nf_conntrack_tuple_hash *h; |
1011 | struct nf_conn *ct; | |
ea781f19 | 1012 | struct hlist_nulls_node *n; |
9fb9cbb1 | 1013 | |
f8ba1aff | 1014 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1015 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
ea781f19 | 1016 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) { |
df0933dc PM |
1017 | ct = nf_ct_tuplehash_to_ctrack(h); |
1018 | if (iter(ct, data)) | |
1019 | goto found; | |
1020 | } | |
601e68e1 | 1021 | } |
ea781f19 | 1022 | hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) { |
df0933dc PM |
1023 | ct = nf_ct_tuplehash_to_ctrack(h); |
1024 | if (iter(ct, data)) | |
ec68e97d | 1025 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 1026 | } |
f8ba1aff | 1027 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
1028 | return NULL; |
1029 | found: | |
c073e3fa | 1030 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 1031 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 1032 | return ct; |
9fb9cbb1 YK |
1033 | } |
1034 | ||
400dad39 AD |
1035 | void nf_ct_iterate_cleanup(struct net *net, |
1036 | int (*iter)(struct nf_conn *i, void *data), | |
1037 | void *data) | |
9fb9cbb1 | 1038 | { |
df0933dc | 1039 | struct nf_conn *ct; |
9fb9cbb1 YK |
1040 | unsigned int bucket = 0; |
1041 | ||
400dad39 | 1042 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
1043 | /* Time to push up daises... */ |
1044 | if (del_timer(&ct->timeout)) | |
1045 | death_by_timeout((unsigned long)ct); | |
1046 | /* ... else the timer will get him soon. */ | |
1047 | ||
1048 | nf_ct_put(ct); | |
1049 | } | |
1050 | } | |
13b18339 | 1051 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1052 | |
19abb7b0 PNA |
1053 | struct __nf_ct_flush_report { |
1054 | u32 pid; | |
1055 | int report; | |
1056 | }; | |
1057 | ||
274d383b | 1058 | static int kill_report(struct nf_conn *i, void *data) |
9fb9cbb1 | 1059 | { |
19abb7b0 PNA |
1060 | struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data; |
1061 | ||
dd7669a9 PNA |
1062 | /* If we fail to deliver the event, death_by_timeout() will retry */ |
1063 | if (nf_conntrack_event_report(IPCT_DESTROY, i, | |
1064 | fr->pid, fr->report) < 0) | |
1065 | return 1; | |
1066 | ||
1067 | /* Avoid the delivery of the destroy event in death_by_timeout(). */ | |
1068 | set_bit(IPS_DYING_BIT, &i->status); | |
9fb9cbb1 YK |
1069 | return 1; |
1070 | } | |
1071 | ||
274d383b PNA |
1072 | static int kill_all(struct nf_conn *i, void *data) |
1073 | { | |
1074 | return 1; | |
1075 | } | |
1076 | ||
ea781f19 | 1077 | void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
1078 | { |
1079 | if (vmalloced) | |
1080 | vfree(hash); | |
1081 | else | |
601e68e1 | 1082 | free_pages((unsigned long)hash, |
f205c5e0 | 1083 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1084 | } |
ac565e5f | 1085 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1086 | |
274d383b | 1087 | void nf_conntrack_flush_report(struct net *net, u32 pid, int report) |
c1d10adb | 1088 | { |
19abb7b0 PNA |
1089 | struct __nf_ct_flush_report fr = { |
1090 | .pid = pid, | |
1091 | .report = report, | |
1092 | }; | |
274d383b | 1093 | nf_ct_iterate_cleanup(net, kill_report, &fr); |
c1d10adb | 1094 | } |
274d383b | 1095 | EXPORT_SYMBOL_GPL(nf_conntrack_flush_report); |
c1d10adb | 1096 | |
ee254fa4 | 1097 | static void nf_ct_release_dying_list(struct net *net) |
dd7669a9 PNA |
1098 | { |
1099 | struct nf_conntrack_tuple_hash *h; | |
1100 | struct nf_conn *ct; | |
1101 | struct hlist_nulls_node *n; | |
1102 | ||
1103 | spin_lock_bh(&nf_conntrack_lock); | |
ee254fa4 | 1104 | hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) { |
dd7669a9 PNA |
1105 | ct = nf_ct_tuplehash_to_ctrack(h); |
1106 | /* never fails to remove them, no listeners at this point */ | |
1107 | nf_ct_kill(ct); | |
1108 | } | |
1109 | spin_unlock_bh(&nf_conntrack_lock); | |
1110 | } | |
1111 | ||
08f6547d | 1112 | static void nf_conntrack_cleanup_init_net(void) |
9fb9cbb1 | 1113 | { |
9edd7ca0 PM |
1114 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1115 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1116 | schedule(); | |
1117 | ||
08f6547d AD |
1118 | nf_conntrack_helper_fini(); |
1119 | nf_conntrack_proto_fini(); | |
08f6547d | 1120 | } |
9fb9cbb1 | 1121 | |
08f6547d AD |
1122 | static void nf_conntrack_cleanup_net(struct net *net) |
1123 | { | |
9fb9cbb1 | 1124 | i_see_dead_people: |
274d383b | 1125 | nf_ct_iterate_cleanup(net, kill_all, NULL); |
ee254fa4 | 1126 | nf_ct_release_dying_list(net); |
49ac8713 | 1127 | if (atomic_read(&net->ct.count) != 0) { |
9fb9cbb1 YK |
1128 | schedule(); |
1129 | goto i_see_dead_people; | |
1130 | } | |
1131 | ||
400dad39 | 1132 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1133 | nf_conntrack_htable_size); |
a0891aa6 | 1134 | nf_conntrack_ecache_fini(net); |
d716a4df | 1135 | nf_conntrack_acct_fini(net); |
9b03f38d | 1136 | nf_conntrack_expect_fini(net); |
5b3501fa ED |
1137 | kmem_cache_destroy(net->ct.nf_conntrack_cachep); |
1138 | kfree(net->ct.slabname); | |
0d55af87 | 1139 | free_percpu(net->ct.stat); |
08f6547d AD |
1140 | } |
1141 | ||
1142 | /* Mishearing the voices in his head, our hero wonders how he's | |
1143 | supposed to kill the mall. */ | |
1144 | void nf_conntrack_cleanup(struct net *net) | |
1145 | { | |
1146 | if (net_eq(net, &init_net)) | |
1147 | rcu_assign_pointer(ip_ct_attach, NULL); | |
1148 | ||
1149 | /* This makes sure all current packets have passed through | |
1150 | netfilter framework. Roll on, two-stage module | |
1151 | delete... */ | |
1152 | synchronize_net(); | |
1153 | ||
1154 | nf_conntrack_cleanup_net(net); | |
1155 | ||
1156 | if (net_eq(net, &init_net)) { | |
1157 | rcu_assign_pointer(nf_ct_destroy, NULL); | |
1158 | nf_conntrack_cleanup_init_net(); | |
1159 | } | |
9fb9cbb1 YK |
1160 | } |
1161 | ||
ea781f19 | 1162 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls) |
9fb9cbb1 | 1163 | { |
ea781f19 ED |
1164 | struct hlist_nulls_head *hash; |
1165 | unsigned int nr_slots, i; | |
1166 | size_t sz; | |
9fb9cbb1 | 1167 | |
601e68e1 | 1168 | *vmalloced = 0; |
8e5105a0 | 1169 | |
ea781f19 ED |
1170 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1171 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
1172 | sz = nr_slots * sizeof(struct hlist_nulls_head); | |
1173 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1174 | get_order(sz)); | |
601e68e1 | 1175 | if (!hash) { |
9fb9cbb1 YK |
1176 | *vmalloced = 1; |
1177 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
ea781f19 | 1178 | hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL); |
9fb9cbb1 YK |
1179 | } |
1180 | ||
ea781f19 ED |
1181 | if (hash && nulls) |
1182 | for (i = 0; i < nr_slots; i++) | |
1183 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1184 | |
1185 | return hash; | |
1186 | } | |
ac565e5f | 1187 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1188 | |
fae718dd | 1189 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1190 | { |
96eb24d7 SH |
1191 | int i, bucket, vmalloced, old_vmalloced; |
1192 | unsigned int hashsize, old_size; | |
9fb9cbb1 | 1193 | int rnd; |
ea781f19 | 1194 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 YK |
1195 | struct nf_conntrack_tuple_hash *h; |
1196 | ||
1197 | /* On boot, we can set this without any fancy locking. */ | |
1198 | if (!nf_conntrack_htable_size) | |
1199 | return param_set_uint(val, kp); | |
1200 | ||
96eb24d7 | 1201 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1202 | if (!hashsize) |
1203 | return -EINVAL; | |
1204 | ||
ea781f19 | 1205 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1); |
9fb9cbb1 YK |
1206 | if (!hash) |
1207 | return -ENOMEM; | |
1208 | ||
1209 | /* We have to rehahs for the new table anyway, so we also can | |
1210 | * use a newrandom seed */ | |
af07d241 | 1211 | get_random_bytes(&rnd, sizeof(rnd)); |
9fb9cbb1 | 1212 | |
76507f69 PM |
1213 | /* Lookups in the old hash might happen in parallel, which means we |
1214 | * might get false negatives during connection lookup. New connections | |
1215 | * created because of a false negative won't make it into the hash | |
1216 | * though since that required taking the lock. | |
1217 | */ | |
f8ba1aff | 1218 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1219 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
ea781f19 ED |
1220 | while (!hlist_nulls_empty(&init_net.ct.hash[i])) { |
1221 | h = hlist_nulls_entry(init_net.ct.hash[i].first, | |
1222 | struct nf_conntrack_tuple_hash, hnnode); | |
1223 | hlist_nulls_del_rcu(&h->hnnode); | |
9fb9cbb1 | 1224 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
ea781f19 | 1225 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1226 | } |
1227 | } | |
1228 | old_size = nf_conntrack_htable_size; | |
400dad39 AD |
1229 | old_vmalloced = init_net.ct.hash_vmalloc; |
1230 | old_hash = init_net.ct.hash; | |
9fb9cbb1 YK |
1231 | |
1232 | nf_conntrack_htable_size = hashsize; | |
400dad39 AD |
1233 | init_net.ct.hash_vmalloc = vmalloced; |
1234 | init_net.ct.hash = hash; | |
9fb9cbb1 | 1235 | nf_conntrack_hash_rnd = rnd; |
f8ba1aff | 1236 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1237 | |
ac565e5f | 1238 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1239 | return 0; |
1240 | } | |
fae718dd | 1241 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1242 | |
fae718dd | 1243 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1244 | &nf_conntrack_htable_size, 0600); |
1245 | ||
08f6547d | 1246 | static int nf_conntrack_init_init_net(void) |
9fb9cbb1 | 1247 | { |
f205c5e0 | 1248 | int max_factor = 8; |
9fb9cbb1 YK |
1249 | int ret; |
1250 | ||
1251 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1252 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1253 | if (!nf_conntrack_htable_size) { |
1254 | nf_conntrack_htable_size | |
4481374c | 1255 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1256 | / sizeof(struct hlist_head)); |
4481374c | 1257 | if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1258 | nf_conntrack_htable_size = 16384; |
1259 | if (nf_conntrack_htable_size < 32) | |
1260 | nf_conntrack_htable_size = 32; | |
1261 | ||
1262 | /* Use a max. factor of four by default to get the same max as | |
1263 | * with the old struct list_heads. When a table size is given | |
1264 | * we use the old value of 8 to avoid reducing the max. | |
1265 | * entries. */ | |
1266 | max_factor = 4; | |
9fb9cbb1 | 1267 | } |
f205c5e0 | 1268 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1269 | |
1270 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1271 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1272 | nf_conntrack_max); | |
1273 | ||
e9c1b084 PM |
1274 | ret = nf_conntrack_proto_init(); |
1275 | if (ret < 0) | |
08f6547d | 1276 | goto err_proto; |
933a41e7 | 1277 | |
ceceae1b YK |
1278 | ret = nf_conntrack_helper_init(); |
1279 | if (ret < 0) | |
08f6547d AD |
1280 | goto err_helper; |
1281 | ||
9edd7ca0 PM |
1282 | /* Set up fake conntrack: to never be deleted, not in any hashes */ |
1283 | #ifdef CONFIG_NET_NS | |
1284 | nf_conntrack_untracked.ct_net = &init_net; | |
1285 | #endif | |
1286 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); | |
1287 | /* - and look it like as a confirmed connection */ | |
1288 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1289 | ||
08f6547d AD |
1290 | return 0; |
1291 | ||
1292 | err_helper: | |
1293 | nf_conntrack_proto_fini(); | |
1294 | err_proto: | |
08f6547d AD |
1295 | return ret; |
1296 | } | |
1297 | ||
8cc20198 ED |
1298 | /* |
1299 | * We need to use special "null" values, not used in hash table | |
1300 | */ | |
1301 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1302 | #define DYING_NULLS_VAL ((1<<30)+1) | |
1303 | ||
08f6547d AD |
1304 | static int nf_conntrack_init_net(struct net *net) |
1305 | { | |
1306 | int ret; | |
ceceae1b | 1307 | |
08f6547d | 1308 | atomic_set(&net->ct.count, 0); |
8cc20198 ED |
1309 | INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL); |
1310 | INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL); | |
08f6547d AD |
1311 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1312 | if (!net->ct.stat) { | |
1313 | ret = -ENOMEM; | |
1314 | goto err_stat; | |
1315 | } | |
5b3501fa ED |
1316 | |
1317 | net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net); | |
1318 | if (!net->ct.slabname) { | |
1319 | ret = -ENOMEM; | |
1320 | goto err_slabname; | |
1321 | } | |
1322 | ||
1323 | net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname, | |
1324 | sizeof(struct nf_conn), 0, | |
1325 | SLAB_DESTROY_BY_RCU, NULL); | |
1326 | if (!net->ct.nf_conntrack_cachep) { | |
1327 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); | |
1328 | ret = -ENOMEM; | |
1329 | goto err_cache; | |
1330 | } | |
08f6547d | 1331 | net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
ea781f19 | 1332 | &net->ct.hash_vmalloc, 1); |
08f6547d AD |
1333 | if (!net->ct.hash) { |
1334 | ret = -ENOMEM; | |
1335 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1336 | goto err_hash; | |
1337 | } | |
1338 | ret = nf_conntrack_expect_init(net); | |
1339 | if (ret < 0) | |
1340 | goto err_expect; | |
d716a4df | 1341 | ret = nf_conntrack_acct_init(net); |
58401572 | 1342 | if (ret < 0) |
08f6547d | 1343 | goto err_acct; |
a0891aa6 PNA |
1344 | ret = nf_conntrack_ecache_init(net); |
1345 | if (ret < 0) | |
1346 | goto err_ecache; | |
7d3cdc6b | 1347 | |
08f6547d | 1348 | return 0; |
9fb9cbb1 | 1349 | |
a0891aa6 PNA |
1350 | err_ecache: |
1351 | nf_conntrack_acct_fini(net); | |
08f6547d | 1352 | err_acct: |
9b03f38d | 1353 | nf_conntrack_expect_fini(net); |
08f6547d | 1354 | err_expect: |
400dad39 | 1355 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
ac565e5f | 1356 | nf_conntrack_htable_size); |
6058fa6b | 1357 | err_hash: |
5b3501fa ED |
1358 | kmem_cache_destroy(net->ct.nf_conntrack_cachep); |
1359 | err_cache: | |
1360 | kfree(net->ct.slabname); | |
1361 | err_slabname: | |
0d55af87 AD |
1362 | free_percpu(net->ct.stat); |
1363 | err_stat: | |
08f6547d AD |
1364 | return ret; |
1365 | } | |
1366 | ||
f9dd09c7 JK |
1367 | s16 (*nf_ct_nat_offset)(const struct nf_conn *ct, |
1368 | enum ip_conntrack_dir dir, | |
1369 | u32 seq); | |
1370 | EXPORT_SYMBOL_GPL(nf_ct_nat_offset); | |
1371 | ||
08f6547d AD |
1372 | int nf_conntrack_init(struct net *net) |
1373 | { | |
1374 | int ret; | |
1375 | ||
1376 | if (net_eq(net, &init_net)) { | |
1377 | ret = nf_conntrack_init_init_net(); | |
1378 | if (ret < 0) | |
1379 | goto out_init_net; | |
1380 | } | |
1381 | ret = nf_conntrack_init_net(net); | |
1382 | if (ret < 0) | |
1383 | goto out_net; | |
1384 | ||
1385 | if (net_eq(net, &init_net)) { | |
1386 | /* For use by REJECT target */ | |
1387 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); | |
1388 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); | |
f9dd09c7 JK |
1389 | |
1390 | /* Howto get NAT offsets */ | |
1391 | rcu_assign_pointer(nf_ct_nat_offset, NULL); | |
08f6547d AD |
1392 | } |
1393 | return 0; | |
1394 | ||
1395 | out_net: | |
1396 | if (net_eq(net, &init_net)) | |
1397 | nf_conntrack_cleanup_init_net(); | |
1398 | out_init_net: | |
1399 | return ret; | |
9fb9cbb1 | 1400 | } |