]>
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> |
d696c7bd | 33 | #include <linux/nsproxy.h> |
ea781f19 | 34 | #include <linux/rculist_nulls.h> |
9fb9cbb1 | 35 | |
9fb9cbb1 YK |
36 | #include <net/netfilter/nf_conntrack.h> |
37 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
605dcad6 | 38 | #include <net/netfilter/nf_conntrack_l4proto.h> |
77ab9cff | 39 | #include <net/netfilter/nf_conntrack_expect.h> |
9fb9cbb1 YK |
40 | #include <net/netfilter/nf_conntrack_helper.h> |
41 | #include <net/netfilter/nf_conntrack_core.h> | |
ecfab2c9 | 42 | #include <net/netfilter/nf_conntrack_extend.h> |
58401572 | 43 | #include <net/netfilter/nf_conntrack_acct.h> |
a0891aa6 | 44 | #include <net/netfilter/nf_conntrack_ecache.h> |
5d0aa2cc | 45 | #include <net/netfilter/nf_conntrack_zones.h> |
e6a7d3c0 | 46 | #include <net/netfilter/nf_nat.h> |
e17b666a | 47 | #include <net/netfilter/nf_nat_core.h> |
9fb9cbb1 | 48 | |
dc808fe2 | 49 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 50 | |
e17b666a PM |
51 | int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, |
52 | enum nf_nat_manip_type manip, | |
39938324 | 53 | const struct nlattr *attr) __read_mostly; |
e6a7d3c0 PNA |
54 | EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); |
55 | ||
f8ba1aff | 56 | DEFINE_SPINLOCK(nf_conntrack_lock); |
13b18339 | 57 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 | 58 | |
e2b7606c | 59 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
60 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
61 | ||
e478075c | 62 | unsigned int nf_conntrack_max __read_mostly; |
a999e683 | 63 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 64 | |
b3c5163f ED |
65 | DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked); |
66 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked); | |
13b18339 | 67 | |
f682cefa | 68 | unsigned int nf_conntrack_hash_rnd __read_mostly; |
9fb9cbb1 | 69 | |
99f07e91 | 70 | static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone) |
9fb9cbb1 | 71 | { |
0794935e | 72 | unsigned int n; |
0794935e PM |
73 | |
74 | /* The direction must be ignored, so we hash everything up to the | |
75 | * destination ports (which is a multiple of 4) and treat the last | |
76 | * three bytes manually. | |
77 | */ | |
78 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
99f07e91 CG |
79 | return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^ |
80 | (((__force __u16)tuple->dst.u.all << 16) | | |
81 | tuple->dst.protonum)); | |
82 | } | |
83 | ||
84 | static u32 __hash_bucket(u32 hash, unsigned int size) | |
85 | { | |
86 | return ((u64)hash * size) >> 32; | |
87 | } | |
88 | ||
89 | static u32 hash_bucket(u32 hash, const struct net *net) | |
90 | { | |
91 | return __hash_bucket(hash, net->ct.htable_size); | |
92 | } | |
0794935e | 93 | |
99f07e91 CG |
94 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, |
95 | u16 zone, unsigned int size) | |
96 | { | |
97 | return __hash_bucket(hash_conntrack_raw(tuple, zone), size); | |
9fb9cbb1 YK |
98 | } |
99 | ||
5d0aa2cc | 100 | static inline u_int32_t hash_conntrack(const struct net *net, u16 zone, |
d696c7bd | 101 | const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 | 102 | { |
99f07e91 | 103 | return __hash_conntrack(tuple, zone, net->ct.htable_size); |
9fb9cbb1 YK |
104 | } |
105 | ||
5f2b4c90 | 106 | bool |
9fb9cbb1 YK |
107 | nf_ct_get_tuple(const struct sk_buff *skb, |
108 | unsigned int nhoff, | |
109 | unsigned int dataoff, | |
110 | u_int16_t l3num, | |
111 | u_int8_t protonum, | |
112 | struct nf_conntrack_tuple *tuple, | |
113 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 114 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 115 | { |
443a70d5 | 116 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
117 | |
118 | tuple->src.l3num = l3num; | |
119 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 120 | return false; |
9fb9cbb1 YK |
121 | |
122 | tuple->dst.protonum = protonum; | |
123 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
124 | ||
605dcad6 | 125 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 126 | } |
13b18339 | 127 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 128 | |
5f2b4c90 JE |
129 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
130 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
131 | { |
132 | struct nf_conntrack_l3proto *l3proto; | |
133 | struct nf_conntrack_l4proto *l4proto; | |
134 | unsigned int protoff; | |
135 | u_int8_t protonum; | |
136 | int ret; | |
137 | ||
138 | rcu_read_lock(); | |
139 | ||
140 | l3proto = __nf_ct_l3proto_find(l3num); | |
141 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
142 | if (ret != NF_ACCEPT) { | |
143 | rcu_read_unlock(); | |
5f2b4c90 | 144 | return false; |
e2a3123f YK |
145 | } |
146 | ||
147 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
148 | ||
149 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
150 | l3proto, l4proto); | |
151 | ||
152 | rcu_read_unlock(); | |
153 | return ret; | |
154 | } | |
155 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
156 | ||
5f2b4c90 | 157 | bool |
9fb9cbb1 YK |
158 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
159 | const struct nf_conntrack_tuple *orig, | |
160 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 161 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 162 | { |
443a70d5 | 163 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
164 | |
165 | inverse->src.l3num = orig->src.l3num; | |
166 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 167 | return false; |
9fb9cbb1 YK |
168 | |
169 | inverse->dst.dir = !orig->dst.dir; | |
170 | ||
171 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 172 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 173 | } |
13b18339 | 174 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 175 | |
9fb9cbb1 YK |
176 | static void |
177 | clean_from_lists(struct nf_conn *ct) | |
178 | { | |
0d53778e | 179 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
180 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
181 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
182 | |
183 | /* Destroy all pending expectations */ | |
c1d10adb | 184 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
185 | } |
186 | ||
187 | static void | |
188 | destroy_conntrack(struct nf_conntrack *nfct) | |
189 | { | |
190 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
0d55af87 | 191 | struct net *net = nf_ct_net(ct); |
605dcad6 | 192 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 193 | |
0d53778e | 194 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
195 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
196 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
197 | ||
9fb9cbb1 YK |
198 | /* To make sure we don't get any weird locking issues here: |
199 | * destroy_conntrack() MUST NOT be called with a write lock | |
200 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 201 | rcu_read_lock(); |
5e8fbe2a | 202 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
203 | if (l4proto && l4proto->destroy) |
204 | l4proto->destroy(ct); | |
9fb9cbb1 | 205 | |
982d9a9c | 206 | rcu_read_unlock(); |
9fb9cbb1 | 207 | |
f8ba1aff | 208 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
209 | /* Expectations will have been removed in clean_from_lists, |
210 | * except TFTP can create an expectation on the first packet, | |
211 | * before connection is in the list, so we need to clean here, | |
212 | * too. */ | |
c1d10adb | 213 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
214 | |
215 | /* We overload first tuple to link into unconfirmed list. */ | |
216 | if (!nf_ct_is_confirmed(ct)) { | |
ea781f19 ED |
217 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); |
218 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
9fb9cbb1 YK |
219 | } |
220 | ||
0d55af87 | 221 | NF_CT_STAT_INC(net, delete); |
f8ba1aff | 222 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
223 | |
224 | if (ct->master) | |
225 | nf_ct_put(ct->master); | |
226 | ||
0d53778e | 227 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
228 | nf_conntrack_free(ct); |
229 | } | |
230 | ||
dd7669a9 | 231 | void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 232 | { |
0d55af87 | 233 | struct net *net = nf_ct_net(ct); |
9fb9cbb1 | 234 | |
9858a3ae | 235 | nf_ct_helper_destroy(ct); |
f8ba1aff | 236 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
237 | /* Inside lock so preempt is disabled on module removal path. |
238 | * Otherwise we can get spurious warnings. */ | |
0d55af87 | 239 | NF_CT_STAT_INC(net, delete_list); |
9fb9cbb1 | 240 | clean_from_lists(ct); |
f8ba1aff | 241 | spin_unlock_bh(&nf_conntrack_lock); |
dd7669a9 PNA |
242 | } |
243 | EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists); | |
244 | ||
245 | static void death_by_event(unsigned long ul_conntrack) | |
246 | { | |
247 | struct nf_conn *ct = (void *)ul_conntrack; | |
248 | struct net *net = nf_ct_net(ct); | |
249 | ||
250 | if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) { | |
251 | /* bad luck, let's retry again */ | |
252 | ct->timeout.expires = jiffies + | |
253 | (random32() % net->ct.sysctl_events_retry_timeout); | |
254 | add_timer(&ct->timeout); | |
255 | return; | |
256 | } | |
257 | /* we've got the event delivered, now it's dying */ | |
258 | set_bit(IPS_DYING_BIT, &ct->status); | |
259 | spin_lock(&nf_conntrack_lock); | |
260 | hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
261 | spin_unlock(&nf_conntrack_lock); | |
262 | nf_ct_put(ct); | |
263 | } | |
264 | ||
265 | void nf_ct_insert_dying_list(struct nf_conn *ct) | |
266 | { | |
267 | struct net *net = nf_ct_net(ct); | |
268 | ||
269 | /* add this conntrack to the dying list */ | |
270 | spin_lock_bh(&nf_conntrack_lock); | |
271 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
272 | &net->ct.dying); | |
273 | spin_unlock_bh(&nf_conntrack_lock); | |
274 | /* set a new timer to retry event delivery */ | |
275 | setup_timer(&ct->timeout, death_by_event, (unsigned long)ct); | |
276 | ct->timeout.expires = jiffies + | |
277 | (random32() % net->ct.sysctl_events_retry_timeout); | |
278 | add_timer(&ct->timeout); | |
279 | } | |
280 | EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list); | |
281 | ||
282 | static void death_by_timeout(unsigned long ul_conntrack) | |
283 | { | |
284 | struct nf_conn *ct = (void *)ul_conntrack; | |
285 | ||
286 | if (!test_bit(IPS_DYING_BIT, &ct->status) && | |
287 | unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) { | |
288 | /* destroy event was not delivered */ | |
289 | nf_ct_delete_from_lists(ct); | |
290 | nf_ct_insert_dying_list(ct); | |
291 | return; | |
292 | } | |
293 | set_bit(IPS_DYING_BIT, &ct->status); | |
294 | nf_ct_delete_from_lists(ct); | |
9fb9cbb1 YK |
295 | nf_ct_put(ct); |
296 | } | |
297 | ||
ea781f19 ED |
298 | /* |
299 | * Warning : | |
300 | * - Caller must take a reference on returned object | |
301 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
302 | * OR | |
303 | * - Caller must lock nf_conntrack_lock before calling this function | |
304 | */ | |
99f07e91 CG |
305 | static struct nf_conntrack_tuple_hash * |
306 | ____nf_conntrack_find(struct net *net, u16 zone, | |
307 | const struct nf_conntrack_tuple *tuple, u32 hash) | |
9fb9cbb1 YK |
308 | { |
309 | struct nf_conntrack_tuple_hash *h; | |
ea781f19 | 310 | struct hlist_nulls_node *n; |
99f07e91 | 311 | unsigned int bucket = hash_bucket(hash, net); |
9fb9cbb1 | 312 | |
4e29e9ec PM |
313 | /* Disable BHs the entire time since we normally need to disable them |
314 | * at least once for the stats anyway. | |
315 | */ | |
316 | local_bh_disable(); | |
ea781f19 | 317 | begin: |
99f07e91 | 318 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) { |
5d0aa2cc PM |
319 | if (nf_ct_tuple_equal(tuple, &h->tuple) && |
320 | nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) { | |
0d55af87 | 321 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 322 | local_bh_enable(); |
9fb9cbb1 YK |
323 | return h; |
324 | } | |
0d55af87 | 325 | NF_CT_STAT_INC(net, searched); |
9fb9cbb1 | 326 | } |
ea781f19 ED |
327 | /* |
328 | * if the nulls value we got at the end of this lookup is | |
329 | * not the expected one, we must restart lookup. | |
330 | * We probably met an item that was moved to another chain. | |
331 | */ | |
99f07e91 | 332 | if (get_nulls_value(n) != bucket) { |
af740b2c | 333 | NF_CT_STAT_INC(net, search_restart); |
ea781f19 | 334 | goto begin; |
af740b2c | 335 | } |
4e29e9ec | 336 | local_bh_enable(); |
9fb9cbb1 YK |
337 | |
338 | return NULL; | |
339 | } | |
99f07e91 CG |
340 | |
341 | struct nf_conntrack_tuple_hash * | |
342 | __nf_conntrack_find(struct net *net, u16 zone, | |
343 | const struct nf_conntrack_tuple *tuple) | |
344 | { | |
345 | return ____nf_conntrack_find(net, zone, tuple, | |
346 | hash_conntrack_raw(tuple, zone)); | |
347 | } | |
13b18339 | 348 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
349 | |
350 | /* Find a connection corresponding to a tuple. */ | |
99f07e91 CG |
351 | static struct nf_conntrack_tuple_hash * |
352 | __nf_conntrack_find_get(struct net *net, u16 zone, | |
353 | const struct nf_conntrack_tuple *tuple, u32 hash) | |
9fb9cbb1 YK |
354 | { |
355 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 356 | struct nf_conn *ct; |
9fb9cbb1 | 357 | |
76507f69 | 358 | rcu_read_lock(); |
ea781f19 | 359 | begin: |
99f07e91 | 360 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
361 | if (h) { |
362 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
363 | if (unlikely(nf_ct_is_dying(ct) || |
364 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 365 | h = NULL; |
ea781f19 | 366 | else { |
5d0aa2cc PM |
367 | if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) || |
368 | nf_ct_zone(ct) != zone)) { | |
ea781f19 ED |
369 | nf_ct_put(ct); |
370 | goto begin; | |
371 | } | |
372 | } | |
76507f69 PM |
373 | } |
374 | rcu_read_unlock(); | |
9fb9cbb1 YK |
375 | |
376 | return h; | |
377 | } | |
99f07e91 CG |
378 | |
379 | struct nf_conntrack_tuple_hash * | |
380 | nf_conntrack_find_get(struct net *net, u16 zone, | |
381 | const struct nf_conntrack_tuple *tuple) | |
382 | { | |
383 | return __nf_conntrack_find_get(net, zone, tuple, | |
384 | hash_conntrack_raw(tuple, zone)); | |
385 | } | |
13b18339 | 386 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 387 | |
c1d10adb PNA |
388 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
389 | unsigned int hash, | |
601e68e1 | 390 | unsigned int repl_hash) |
c1d10adb | 391 | { |
400dad39 AD |
392 | struct net *net = nf_ct_net(ct); |
393 | ||
ea781f19 | 394 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
400dad39 | 395 | &net->ct.hash[hash]); |
ea781f19 | 396 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
400dad39 | 397 | &net->ct.hash[repl_hash]); |
c1d10adb PNA |
398 | } |
399 | ||
400 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
401 | { | |
d696c7bd | 402 | struct net *net = nf_ct_net(ct); |
c1d10adb | 403 | unsigned int hash, repl_hash; |
5d0aa2cc | 404 | u16 zone; |
c1d10adb | 405 | |
5d0aa2cc PM |
406 | zone = nf_ct_zone(ct); |
407 | hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
408 | repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
c1d10adb | 409 | |
c1d10adb | 410 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
c1d10adb | 411 | } |
13b18339 | 412 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 413 | |
9fb9cbb1 YK |
414 | /* Confirm a connection given skb; places it in hash table */ |
415 | int | |
3db05fea | 416 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
417 | { |
418 | unsigned int hash, repl_hash; | |
df0933dc | 419 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 420 | struct nf_conn *ct; |
df0933dc | 421 | struct nf_conn_help *help; |
ea781f19 | 422 | struct hlist_nulls_node *n; |
9fb9cbb1 | 423 | enum ip_conntrack_info ctinfo; |
400dad39 | 424 | struct net *net; |
5d0aa2cc | 425 | u16 zone; |
9fb9cbb1 | 426 | |
3db05fea | 427 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 428 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
429 | |
430 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
431 | ICMP/TCP RST packets in other direction. Actual packet | |
432 | which created connection will be IP_CT_NEW or for an | |
433 | expected connection, IP_CT_RELATED. */ | |
434 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
435 | return NF_ACCEPT; | |
436 | ||
5d0aa2cc | 437 | zone = nf_ct_zone(ct); |
99f07e91 CG |
438 | /* reuse the hash saved before */ |
439 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
440 | hash = hash_bucket(hash, net); | |
441 | repl_hash = hash_conntrack(net, zone, | |
442 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
9fb9cbb1 YK |
443 | |
444 | /* We're not in hash table, and we refuse to set up related | |
445 | connections for unconfirmed conns. But packet copies and | |
446 | REJECT will give spurious warnings here. */ | |
447 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
448 | ||
449 | /* No external references means noone else could have | |
450 | confirmed us. */ | |
451 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 452 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 453 | |
f8ba1aff | 454 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 455 | |
fc350777 JM |
456 | /* We have to check the DYING flag inside the lock to prevent |
457 | a race against nf_ct_get_next_corpse() possibly called from | |
458 | user context, else we insert an already 'dead' hash, blocking | |
459 | further use of that particular connection -JM */ | |
460 | ||
461 | if (unlikely(nf_ct_is_dying(ct))) { | |
462 | spin_unlock_bh(&nf_conntrack_lock); | |
463 | return NF_ACCEPT; | |
464 | } | |
465 | ||
9fb9cbb1 YK |
466 | /* See if there's one in the list already, including reverse: |
467 | NAT could have grabbed it without realizing, since we're | |
468 | not in the hash. If there is, we lost race. */ | |
ea781f19 | 469 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) |
df0933dc | 470 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
5d0aa2cc PM |
471 | &h->tuple) && |
472 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
df0933dc | 473 | goto out; |
ea781f19 | 474 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode) |
df0933dc | 475 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
5d0aa2cc PM |
476 | &h->tuple) && |
477 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
df0933dc | 478 | goto out; |
9fb9cbb1 | 479 | |
df0933dc | 480 | /* Remove from unconfirmed list */ |
ea781f19 | 481 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
df0933dc | 482 | |
df0933dc PM |
483 | /* Timer relative to confirmation time, not original |
484 | setting time, otherwise we'd get timer wrap in | |
485 | weird delay cases. */ | |
486 | ct->timeout.expires += jiffies; | |
487 | add_timer(&ct->timeout); | |
488 | atomic_inc(&ct->ct_general.use); | |
489 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
5c8ec910 PM |
490 | |
491 | /* Since the lookup is lockless, hash insertion must be done after | |
492 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
493 | * guarantee that no other CPU can find the conntrack before the above | |
494 | * stores are visible. | |
495 | */ | |
496 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
0d55af87 | 497 | NF_CT_STAT_INC(net, insert); |
f8ba1aff | 498 | spin_unlock_bh(&nf_conntrack_lock); |
5c8ec910 | 499 | |
df0933dc PM |
500 | help = nfct_help(ct); |
501 | if (help && help->helper) | |
a71996fc | 502 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 503 | |
df0933dc | 504 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 505 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 506 | return NF_ACCEPT; |
9fb9cbb1 | 507 | |
df0933dc | 508 | out: |
0d55af87 | 509 | NF_CT_STAT_INC(net, insert_failed); |
f8ba1aff | 510 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
511 | return NF_DROP; |
512 | } | |
13b18339 | 513 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
514 | |
515 | /* Returns true if a connection correspondings to the tuple (required | |
516 | for NAT). */ | |
517 | int | |
518 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
519 | const struct nf_conn *ignored_conntrack) | |
520 | { | |
400dad39 | 521 | struct net *net = nf_ct_net(ignored_conntrack); |
9fb9cbb1 | 522 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 523 | struct hlist_nulls_node *n; |
5d0aa2cc PM |
524 | struct nf_conn *ct; |
525 | u16 zone = nf_ct_zone(ignored_conntrack); | |
526 | unsigned int hash = hash_conntrack(net, zone, tuple); | |
9fb9cbb1 | 527 | |
4e29e9ec PM |
528 | /* Disable BHs the entire time since we need to disable them at |
529 | * least once for the stats anyway. | |
530 | */ | |
531 | rcu_read_lock_bh(); | |
ea781f19 | 532 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { |
5d0aa2cc PM |
533 | ct = nf_ct_tuplehash_to_ctrack(h); |
534 | if (ct != ignored_conntrack && | |
535 | nf_ct_tuple_equal(tuple, &h->tuple) && | |
536 | nf_ct_zone(ct) == zone) { | |
0d55af87 | 537 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 538 | rcu_read_unlock_bh(); |
ba419aff PM |
539 | return 1; |
540 | } | |
0d55af87 | 541 | NF_CT_STAT_INC(net, searched); |
ba419aff | 542 | } |
4e29e9ec | 543 | rcu_read_unlock_bh(); |
9fb9cbb1 | 544 | |
ba419aff | 545 | return 0; |
9fb9cbb1 | 546 | } |
13b18339 | 547 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 548 | |
7ae7730f PM |
549 | #define NF_CT_EVICTION_RANGE 8 |
550 | ||
9fb9cbb1 YK |
551 | /* There's a small race here where we may free a just-assured |
552 | connection. Too bad: we're in trouble anyway. */ | |
400dad39 | 553 | static noinline int early_drop(struct net *net, unsigned int hash) |
9fb9cbb1 | 554 | { |
f205c5e0 | 555 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 556 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 557 | struct nf_conn *ct = NULL, *tmp; |
ea781f19 | 558 | struct hlist_nulls_node *n; |
7ae7730f | 559 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
560 | int dropped = 0; |
561 | ||
76507f69 | 562 | rcu_read_lock(); |
d696c7bd | 563 | for (i = 0; i < net->ct.htable_size; i++) { |
ea781f19 ED |
564 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], |
565 | hnnode) { | |
7ae7730f PM |
566 | tmp = nf_ct_tuplehash_to_ctrack(h); |
567 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
568 | ct = tmp; | |
569 | cnt++; | |
570 | } | |
76507f69 | 571 | |
5ae27aa2 CG |
572 | if (ct != NULL) { |
573 | if (likely(!nf_ct_is_dying(ct) && | |
574 | atomic_inc_not_zero(&ct->ct_general.use))) | |
575 | break; | |
576 | else | |
577 | ct = NULL; | |
578 | } | |
579 | ||
580 | if (cnt >= NF_CT_EVICTION_RANGE) | |
7ae7730f | 581 | break; |
5ae27aa2 | 582 | |
d696c7bd | 583 | hash = (hash + 1) % net->ct.htable_size; |
9fb9cbb1 | 584 | } |
76507f69 | 585 | rcu_read_unlock(); |
9fb9cbb1 YK |
586 | |
587 | if (!ct) | |
588 | return dropped; | |
589 | ||
590 | if (del_timer(&ct->timeout)) { | |
591 | death_by_timeout((unsigned long)ct); | |
592 | dropped = 1; | |
0d55af87 | 593 | NF_CT_STAT_INC_ATOMIC(net, early_drop); |
9fb9cbb1 YK |
594 | } |
595 | nf_ct_put(ct); | |
596 | return dropped; | |
597 | } | |
598 | ||
f682cefa CG |
599 | void init_nf_conntrack_hash_rnd(void) |
600 | { | |
601 | unsigned int rand; | |
602 | ||
603 | /* | |
604 | * Why not initialize nf_conntrack_rnd in a "init()" function ? | |
605 | * Because there isn't enough entropy when system initializing, | |
606 | * and we initialize it as late as possible. | |
607 | */ | |
608 | do { | |
609 | get_random_bytes(&rand, sizeof(rand)); | |
610 | } while (!rand); | |
611 | cmpxchg(&nf_conntrack_hash_rnd, 0, rand); | |
612 | } | |
613 | ||
99f07e91 CG |
614 | static struct nf_conn * |
615 | __nf_conntrack_alloc(struct net *net, u16 zone, | |
616 | const struct nf_conntrack_tuple *orig, | |
617 | const struct nf_conntrack_tuple *repl, | |
618 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 619 | { |
cd7fcbf1 | 620 | struct nf_conn *ct; |
9fb9cbb1 | 621 | |
b2390969 | 622 | if (unlikely(!nf_conntrack_hash_rnd)) { |
f682cefa | 623 | init_nf_conntrack_hash_rnd(); |
99f07e91 CG |
624 | /* recompute the hash as nf_conntrack_hash_rnd is initialized */ |
625 | hash = hash_conntrack_raw(orig, zone); | |
9fb9cbb1 YK |
626 | } |
627 | ||
5251e2d2 | 628 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 629 | atomic_inc(&net->ct.count); |
5251e2d2 | 630 | |
76eb9460 | 631 | if (nf_conntrack_max && |
49ac8713 | 632 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
99f07e91 | 633 | if (!early_drop(net, hash_bucket(hash, net))) { |
49ac8713 | 634 | atomic_dec(&net->ct.count); |
9fb9cbb1 YK |
635 | if (net_ratelimit()) |
636 | printk(KERN_WARNING | |
637 | "nf_conntrack: table full, dropping" | |
638 | " packet.\n"); | |
639 | return ERR_PTR(-ENOMEM); | |
640 | } | |
641 | } | |
642 | ||
941297f4 ED |
643 | /* |
644 | * Do not use kmem_cache_zalloc(), as this cache uses | |
645 | * SLAB_DESTROY_BY_RCU. | |
646 | */ | |
5b3501fa | 647 | ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp); |
c88130bc | 648 | if (ct == NULL) { |
0d53778e | 649 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
49ac8713 | 650 | atomic_dec(&net->ct.count); |
dacd2a1a | 651 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 652 | } |
941297f4 ED |
653 | /* |
654 | * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next | |
655 | * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged. | |
656 | */ | |
657 | memset(&ct->tuplehash[IP_CT_DIR_MAX], 0, | |
e5fc9e7a CG |
658 | offsetof(struct nf_conn, proto) - |
659 | offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX])); | |
440f0d58 | 660 | spin_lock_init(&ct->lock); |
c88130bc | 661 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 662 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 663 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
664 | /* save hash for reusing when confirming */ |
665 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
9fb9cbb1 | 666 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 667 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
c2d9ba9b | 668 | write_pnet(&ct->ct_net, net); |
5d0aa2cc PM |
669 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
670 | if (zone) { | |
671 | struct nf_conntrack_zone *nf_ct_zone; | |
672 | ||
673 | nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC); | |
674 | if (!nf_ct_zone) | |
675 | goto out_free; | |
676 | nf_ct_zone->id = zone; | |
677 | } | |
678 | #endif | |
941297f4 ED |
679 | /* |
680 | * changes to lookup keys must be done before setting refcnt to 1 | |
681 | */ | |
682 | smp_wmb(); | |
683 | atomic_set(&ct->ct_general.use, 1); | |
c88130bc | 684 | return ct; |
5d0aa2cc PM |
685 | |
686 | #ifdef CONFIG_NF_CONNTRACK_ZONES | |
687 | out_free: | |
688 | kmem_cache_free(net->ct.nf_conntrack_cachep, ct); | |
689 | return ERR_PTR(-ENOMEM); | |
690 | #endif | |
9fb9cbb1 | 691 | } |
99f07e91 CG |
692 | |
693 | struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone, | |
694 | const struct nf_conntrack_tuple *orig, | |
695 | const struct nf_conntrack_tuple *repl, | |
696 | gfp_t gfp) | |
697 | { | |
698 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
699 | } | |
13b18339 | 700 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 701 | |
c88130bc | 702 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 703 | { |
1d45209d ED |
704 | struct net *net = nf_ct_net(ct); |
705 | ||
ceeff754 | 706 | nf_ct_ext_destroy(ct); |
1d45209d | 707 | atomic_dec(&net->ct.count); |
ea781f19 | 708 | nf_ct_ext_free(ct); |
5b3501fa | 709 | kmem_cache_free(net->ct.nf_conntrack_cachep, ct); |
76507f69 | 710 | } |
13b18339 | 711 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
712 | |
713 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
714 | failed due to stress. Otherwise it really is unclassifiable. */ | |
715 | static struct nf_conntrack_tuple_hash * | |
b2a15a60 | 716 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 717 | const struct nf_conntrack_tuple *tuple, |
9fb9cbb1 | 718 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 719 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 720 | struct sk_buff *skb, |
99f07e91 | 721 | unsigned int dataoff, u32 hash) |
9fb9cbb1 | 722 | { |
c88130bc | 723 | struct nf_conn *ct; |
3c158f7f | 724 | struct nf_conn_help *help; |
9fb9cbb1 | 725 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 726 | struct nf_conntrack_ecache *ecache; |
9fb9cbb1 | 727 | struct nf_conntrack_expect *exp; |
5d0aa2cc | 728 | u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; |
9fb9cbb1 | 729 | |
605dcad6 | 730 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 731 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
732 | return NULL; |
733 | } | |
734 | ||
99f07e91 CG |
735 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
736 | hash); | |
cd7fcbf1 | 737 | if (IS_ERR(ct)) { |
0d53778e | 738 | pr_debug("Can't allocate conntrack.\n"); |
c88130bc | 739 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 YK |
740 | } |
741 | ||
c88130bc PM |
742 | if (!l4proto->new(ct, skb, dataoff)) { |
743 | nf_conntrack_free(ct); | |
0d53778e | 744 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
745 | return NULL; |
746 | } | |
747 | ||
58401572 | 748 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
b2a15a60 PM |
749 | |
750 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
751 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
752 | ecache ? ecache->expmask : 0, | |
753 | GFP_ATOMIC); | |
58401572 | 754 | |
f8ba1aff | 755 | spin_lock_bh(&nf_conntrack_lock); |
5d0aa2cc | 756 | exp = nf_ct_find_expectation(net, zone, tuple); |
9fb9cbb1 | 757 | if (exp) { |
0d53778e | 758 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 759 | ct, exp); |
9fb9cbb1 | 760 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
761 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
762 | ct->master = exp->master; | |
ceceae1b | 763 | if (exp->helper) { |
c88130bc | 764 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b YK |
765 | if (help) |
766 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
767 | } |
768 | ||
9fb9cbb1 | 769 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 770 | ct->mark = exp->master->mark; |
7c9728c3 JM |
771 | #endif |
772 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 773 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 774 | #endif |
c88130bc | 775 | nf_conntrack_get(&ct->master->ct_general); |
0d55af87 | 776 | NF_CT_STAT_INC(net, expect_new); |
22e7410b | 777 | } else { |
b2a15a60 | 778 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
0d55af87 | 779 | NF_CT_STAT_INC(net, new); |
22e7410b | 780 | } |
9fb9cbb1 YK |
781 | |
782 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
ea781f19 | 783 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
63c9a262 | 784 | &net->ct.unconfirmed); |
9fb9cbb1 | 785 | |
f8ba1aff | 786 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
787 | |
788 | if (exp) { | |
789 | if (exp->expectfn) | |
c88130bc | 790 | exp->expectfn(ct, exp); |
6823645d | 791 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
792 | } |
793 | ||
c88130bc | 794 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
795 | } |
796 | ||
797 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
798 | static inline struct nf_conn * | |
b2a15a60 | 799 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 800 | struct sk_buff *skb, |
9fb9cbb1 YK |
801 | unsigned int dataoff, |
802 | u_int16_t l3num, | |
803 | u_int8_t protonum, | |
804 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 805 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
806 | int *set_reply, |
807 | enum ip_conntrack_info *ctinfo) | |
808 | { | |
809 | struct nf_conntrack_tuple tuple; | |
810 | struct nf_conntrack_tuple_hash *h; | |
811 | struct nf_conn *ct; | |
5d0aa2cc | 812 | u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; |
99f07e91 | 813 | u32 hash; |
9fb9cbb1 | 814 | |
bbe735e4 | 815 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 816 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 817 | l4proto)) { |
0d53778e | 818 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
819 | return NULL; |
820 | } | |
821 | ||
822 | /* look for tuple match */ | |
99f07e91 CG |
823 | hash = hash_conntrack_raw(&tuple, zone); |
824 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); | |
9fb9cbb1 | 825 | if (!h) { |
b2a15a60 | 826 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
99f07e91 | 827 | skb, dataoff, hash); |
9fb9cbb1 YK |
828 | if (!h) |
829 | return NULL; | |
830 | if (IS_ERR(h)) | |
831 | return (void *)h; | |
832 | } | |
833 | ct = nf_ct_tuplehash_to_ctrack(h); | |
834 | ||
835 | /* It exists; we have (non-exclusive) reference. */ | |
836 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
837 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
838 | /* Please set reply bit if this packet OK */ | |
839 | *set_reply = 1; | |
840 | } else { | |
841 | /* Once we've had two way comms, always ESTABLISHED. */ | |
842 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 843 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
844 | *ctinfo = IP_CT_ESTABLISHED; |
845 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
846 | pr_debug("nf_conntrack_in: related packet for %p\n", |
847 | ct); | |
9fb9cbb1 YK |
848 | *ctinfo = IP_CT_RELATED; |
849 | } else { | |
0d53778e | 850 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
851 | *ctinfo = IP_CT_NEW; |
852 | } | |
853 | *set_reply = 0; | |
854 | } | |
855 | skb->nfct = &ct->ct_general; | |
856 | skb->nfctinfo = *ctinfo; | |
857 | return ct; | |
858 | } | |
859 | ||
860 | unsigned int | |
a702a65f AD |
861 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
862 | struct sk_buff *skb) | |
9fb9cbb1 | 863 | { |
b2a15a60 | 864 | struct nf_conn *ct, *tmpl = NULL; |
9fb9cbb1 YK |
865 | enum ip_conntrack_info ctinfo; |
866 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 867 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
868 | unsigned int dataoff; |
869 | u_int8_t protonum; | |
870 | int set_reply = 0; | |
871 | int ret; | |
872 | ||
3db05fea | 873 | if (skb->nfct) { |
b2a15a60 PM |
874 | /* Previously seen (loopback or untracked)? Ignore. */ |
875 | tmpl = (struct nf_conn *)skb->nfct; | |
876 | if (!nf_ct_is_template(tmpl)) { | |
877 | NF_CT_STAT_INC_ATOMIC(net, ignore); | |
878 | return NF_ACCEPT; | |
879 | } | |
880 | skb->nfct = NULL; | |
9fb9cbb1 YK |
881 | } |
882 | ||
923f4902 | 883 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 884 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 885 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
886 | &dataoff, &protonum); |
887 | if (ret <= 0) { | |
0d53778e | 888 | pr_debug("not prepared to track yet or error occured\n"); |
0d55af87 AD |
889 | NF_CT_STAT_INC_ATOMIC(net, error); |
890 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
891 | ret = -ret; |
892 | goto out; | |
9fb9cbb1 YK |
893 | } |
894 | ||
76108cea | 895 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
896 | |
897 | /* It may be an special packet, error, unclean... | |
898 | * inverse of the return code tells to the netfilter | |
899 | * core what to do with the packet. */ | |
74c51a14 | 900 | if (l4proto->error != NULL) { |
8fea97ec PM |
901 | ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo, |
902 | pf, hooknum); | |
74c51a14 | 903 | if (ret <= 0) { |
0d55af87 AD |
904 | NF_CT_STAT_INC_ATOMIC(net, error); |
905 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
906 | ret = -ret; |
907 | goto out; | |
74c51a14 | 908 | } |
9fb9cbb1 YK |
909 | } |
910 | ||
b2a15a60 | 911 | ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
a702a65f | 912 | l3proto, l4proto, &set_reply, &ctinfo); |
9fb9cbb1 YK |
913 | if (!ct) { |
914 | /* Not valid part of a connection */ | |
0d55af87 | 915 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
b2a15a60 PM |
916 | ret = NF_ACCEPT; |
917 | goto out; | |
9fb9cbb1 YK |
918 | } |
919 | ||
920 | if (IS_ERR(ct)) { | |
921 | /* Too stressed to deal. */ | |
0d55af87 | 922 | NF_CT_STAT_INC_ATOMIC(net, drop); |
b2a15a60 PM |
923 | ret = NF_DROP; |
924 | goto out; | |
9fb9cbb1 YK |
925 | } |
926 | ||
3db05fea | 927 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 928 | |
3db05fea | 929 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
ec8d5409 | 930 | if (ret <= 0) { |
9fb9cbb1 YK |
931 | /* Invalid: inverse of the return code tells |
932 | * the netfilter core what to do */ | |
0d53778e | 933 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
934 | nf_conntrack_put(skb->nfct); |
935 | skb->nfct = NULL; | |
0d55af87 | 936 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
937 | if (ret == -NF_DROP) |
938 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
b2a15a60 PM |
939 | ret = -ret; |
940 | goto out; | |
9fb9cbb1 YK |
941 | } |
942 | ||
943 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 944 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 PM |
945 | out: |
946 | if (tmpl) | |
947 | nf_ct_put(tmpl); | |
9fb9cbb1 YK |
948 | |
949 | return ret; | |
950 | } | |
13b18339 | 951 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 952 | |
5f2b4c90 JE |
953 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
954 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 955 | { |
5f2b4c90 | 956 | bool ret; |
923f4902 PM |
957 | |
958 | rcu_read_lock(); | |
959 | ret = nf_ct_invert_tuple(inverse, orig, | |
960 | __nf_ct_l3proto_find(orig->src.l3num), | |
961 | __nf_ct_l4proto_find(orig->src.l3num, | |
962 | orig->dst.protonum)); | |
963 | rcu_read_unlock(); | |
964 | return ret; | |
9fb9cbb1 | 965 | } |
13b18339 | 966 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 967 | |
5b1158e9 JK |
968 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
969 | implicitly racy: see __nf_conntrack_confirm */ | |
970 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
971 | const struct nf_conntrack_tuple *newreply) | |
972 | { | |
973 | struct nf_conn_help *help = nfct_help(ct); | |
974 | ||
5b1158e9 JK |
975 | /* Should be unconfirmed, so not in hash table yet */ |
976 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
977 | ||
0d53778e | 978 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 979 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
980 | |
981 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 982 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 983 | return; |
ceceae1b | 984 | |
c52fbb41 | 985 | rcu_read_lock(); |
b2a15a60 | 986 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 987 | rcu_read_unlock(); |
5b1158e9 | 988 | } |
13b18339 | 989 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 990 | |
9fb9cbb1 YK |
991 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
992 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
993 | enum ip_conntrack_info ctinfo, | |
994 | const struct sk_buff *skb, | |
995 | unsigned long extra_jiffies, | |
996 | int do_acct) | |
997 | { | |
9fb9cbb1 YK |
998 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); |
999 | NF_CT_ASSERT(skb); | |
1000 | ||
997ae831 | 1001 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1002 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1003 | goto acct; | |
997ae831 | 1004 | |
9fb9cbb1 YK |
1005 | /* If not in hash table, timer will not be active yet */ |
1006 | if (!nf_ct_is_confirmed(ct)) { | |
1007 | ct->timeout.expires = extra_jiffies; | |
9fb9cbb1 | 1008 | } else { |
be00c8e4 MJ |
1009 | unsigned long newtime = jiffies + extra_jiffies; |
1010 | ||
1011 | /* Only update the timeout if the new timeout is at least | |
1012 | HZ jiffies from the old timeout. Need del_timer for race | |
1013 | avoidance (may already be dying). */ | |
65cb9fda PM |
1014 | if (newtime - ct->timeout.expires >= HZ) |
1015 | mod_timer_pending(&ct->timeout, newtime); | |
9fb9cbb1 YK |
1016 | } |
1017 | ||
47d95045 | 1018 | acct: |
9fb9cbb1 | 1019 | if (do_acct) { |
58401572 | 1020 | struct nf_conn_counter *acct; |
3ffd5eeb | 1021 | |
58401572 KPO |
1022 | acct = nf_conn_acct_find(ct); |
1023 | if (acct) { | |
65cb9fda | 1024 | spin_lock_bh(&ct->lock); |
58401572 | 1025 | acct[CTINFO2DIR(ctinfo)].packets++; |
6661481d | 1026 | acct[CTINFO2DIR(ctinfo)].bytes += skb->len; |
65cb9fda | 1027 | spin_unlock_bh(&ct->lock); |
58401572 | 1028 | } |
9fb9cbb1 | 1029 | } |
9fb9cbb1 | 1030 | } |
13b18339 | 1031 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1032 | |
4c889498 DM |
1033 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
1034 | enum ip_conntrack_info ctinfo, | |
1035 | const struct sk_buff *skb, | |
1036 | int do_acct) | |
51091764 | 1037 | { |
718d4ad9 | 1038 | if (do_acct) { |
58401572 KPO |
1039 | struct nf_conn_counter *acct; |
1040 | ||
58401572 KPO |
1041 | acct = nf_conn_acct_find(ct); |
1042 | if (acct) { | |
65cb9fda | 1043 | spin_lock_bh(&ct->lock); |
58401572 KPO |
1044 | acct[CTINFO2DIR(ctinfo)].packets++; |
1045 | acct[CTINFO2DIR(ctinfo)].bytes += | |
1046 | skb->len - skb_network_offset(skb); | |
65cb9fda | 1047 | spin_unlock_bh(&ct->lock); |
58401572 | 1048 | } |
718d4ad9 | 1049 | } |
58401572 | 1050 | |
4c889498 | 1051 | if (del_timer(&ct->timeout)) { |
51091764 | 1052 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
1053 | return true; |
1054 | } | |
1055 | return false; | |
51091764 | 1056 | } |
718d4ad9 | 1057 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 1058 | |
5d0aa2cc PM |
1059 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1060 | static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = { | |
1061 | .len = sizeof(struct nf_conntrack_zone), | |
1062 | .align = __alignof__(struct nf_conntrack_zone), | |
1063 | .id = NF_CT_EXT_ZONE, | |
1064 | }; | |
1065 | #endif | |
1066 | ||
e281db5c | 1067 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
1068 | |
1069 | #include <linux/netfilter/nfnetlink.h> | |
1070 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1071 | #include <linux/mutex.h> |
1072 | ||
c1d10adb PNA |
1073 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1074 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1075 | * or depend on ctnetlink */ | |
fdf70832 | 1076 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1077 | const struct nf_conntrack_tuple *tuple) |
1078 | { | |
77236b6e PM |
1079 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
1080 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
1081 | return 0; |
1082 | ||
df6fb868 | 1083 | nla_put_failure: |
c1d10adb PNA |
1084 | return -1; |
1085 | } | |
fdf70832 | 1086 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1087 | |
f73e924c PM |
1088 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1089 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1090 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1091 | }; |
f73e924c | 1092 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1093 | |
fdf70832 | 1094 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1095 | struct nf_conntrack_tuple *t) |
1096 | { | |
df6fb868 | 1097 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1098 | return -EINVAL; |
1099 | ||
77236b6e PM |
1100 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1101 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1102 | |
1103 | return 0; | |
1104 | } | |
fdf70832 | 1105 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1106 | |
1107 | int nf_ct_port_nlattr_tuple_size(void) | |
1108 | { | |
1109 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1110 | } | |
1111 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1112 | #endif |
1113 | ||
9fb9cbb1 | 1114 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 1115 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
1116 | { |
1117 | struct nf_conn *ct; | |
1118 | enum ip_conntrack_info ctinfo; | |
1119 | ||
1120 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1121 | ct = nf_ct_get(skb, &ctinfo); | |
1122 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
1123 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
1124 | else | |
1125 | ctinfo = IP_CT_RELATED; | |
1126 | ||
1127 | /* Attach to new skbuff, and increment count */ | |
1128 | nskb->nfct = &ct->ct_general; | |
1129 | nskb->nfctinfo = ctinfo; | |
1130 | nf_conntrack_get(nskb->nfct); | |
1131 | } | |
1132 | ||
9fb9cbb1 | 1133 | /* Bring out ya dead! */ |
df0933dc | 1134 | static struct nf_conn * |
400dad39 | 1135 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1136 | void *data, unsigned int *bucket) |
1137 | { | |
df0933dc PM |
1138 | struct nf_conntrack_tuple_hash *h; |
1139 | struct nf_conn *ct; | |
ea781f19 | 1140 | struct hlist_nulls_node *n; |
9fb9cbb1 | 1141 | |
f8ba1aff | 1142 | spin_lock_bh(&nf_conntrack_lock); |
d696c7bd | 1143 | for (; *bucket < net->ct.htable_size; (*bucket)++) { |
ea781f19 | 1144 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) { |
df0933dc PM |
1145 | ct = nf_ct_tuplehash_to_ctrack(h); |
1146 | if (iter(ct, data)) | |
1147 | goto found; | |
1148 | } | |
601e68e1 | 1149 | } |
ea781f19 | 1150 | hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) { |
df0933dc PM |
1151 | ct = nf_ct_tuplehash_to_ctrack(h); |
1152 | if (iter(ct, data)) | |
ec68e97d | 1153 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 1154 | } |
f8ba1aff | 1155 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
1156 | return NULL; |
1157 | found: | |
c073e3fa | 1158 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 1159 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 1160 | return ct; |
9fb9cbb1 YK |
1161 | } |
1162 | ||
400dad39 AD |
1163 | void nf_ct_iterate_cleanup(struct net *net, |
1164 | int (*iter)(struct nf_conn *i, void *data), | |
1165 | void *data) | |
9fb9cbb1 | 1166 | { |
df0933dc | 1167 | struct nf_conn *ct; |
9fb9cbb1 YK |
1168 | unsigned int bucket = 0; |
1169 | ||
400dad39 | 1170 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
1171 | /* Time to push up daises... */ |
1172 | if (del_timer(&ct->timeout)) | |
1173 | death_by_timeout((unsigned long)ct); | |
1174 | /* ... else the timer will get him soon. */ | |
1175 | ||
1176 | nf_ct_put(ct); | |
1177 | } | |
1178 | } | |
13b18339 | 1179 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1180 | |
19abb7b0 PNA |
1181 | struct __nf_ct_flush_report { |
1182 | u32 pid; | |
1183 | int report; | |
1184 | }; | |
1185 | ||
274d383b | 1186 | static int kill_report(struct nf_conn *i, void *data) |
9fb9cbb1 | 1187 | { |
19abb7b0 PNA |
1188 | struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data; |
1189 | ||
dd7669a9 PNA |
1190 | /* If we fail to deliver the event, death_by_timeout() will retry */ |
1191 | if (nf_conntrack_event_report(IPCT_DESTROY, i, | |
1192 | fr->pid, fr->report) < 0) | |
1193 | return 1; | |
1194 | ||
1195 | /* Avoid the delivery of the destroy event in death_by_timeout(). */ | |
1196 | set_bit(IPS_DYING_BIT, &i->status); | |
9fb9cbb1 YK |
1197 | return 1; |
1198 | } | |
1199 | ||
274d383b PNA |
1200 | static int kill_all(struct nf_conn *i, void *data) |
1201 | { | |
1202 | return 1; | |
1203 | } | |
1204 | ||
ea781f19 | 1205 | void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
1206 | { |
1207 | if (vmalloced) | |
1208 | vfree(hash); | |
1209 | else | |
601e68e1 | 1210 | free_pages((unsigned long)hash, |
f205c5e0 | 1211 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1212 | } |
ac565e5f | 1213 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1214 | |
274d383b | 1215 | void nf_conntrack_flush_report(struct net *net, u32 pid, int report) |
c1d10adb | 1216 | { |
19abb7b0 PNA |
1217 | struct __nf_ct_flush_report fr = { |
1218 | .pid = pid, | |
1219 | .report = report, | |
1220 | }; | |
274d383b | 1221 | nf_ct_iterate_cleanup(net, kill_report, &fr); |
c1d10adb | 1222 | } |
274d383b | 1223 | EXPORT_SYMBOL_GPL(nf_conntrack_flush_report); |
c1d10adb | 1224 | |
ee254fa4 | 1225 | static void nf_ct_release_dying_list(struct net *net) |
dd7669a9 PNA |
1226 | { |
1227 | struct nf_conntrack_tuple_hash *h; | |
1228 | struct nf_conn *ct; | |
1229 | struct hlist_nulls_node *n; | |
1230 | ||
1231 | spin_lock_bh(&nf_conntrack_lock); | |
ee254fa4 | 1232 | hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) { |
dd7669a9 PNA |
1233 | ct = nf_ct_tuplehash_to_ctrack(h); |
1234 | /* never fails to remove them, no listeners at this point */ | |
1235 | nf_ct_kill(ct); | |
1236 | } | |
1237 | spin_unlock_bh(&nf_conntrack_lock); | |
1238 | } | |
1239 | ||
b3c5163f ED |
1240 | static int untrack_refs(void) |
1241 | { | |
1242 | int cnt = 0, cpu; | |
1243 | ||
1244 | for_each_possible_cpu(cpu) { | |
1245 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
1246 | ||
1247 | cnt += atomic_read(&ct->ct_general.use) - 1; | |
1248 | } | |
1249 | return cnt; | |
1250 | } | |
1251 | ||
08f6547d | 1252 | static void nf_conntrack_cleanup_init_net(void) |
9fb9cbb1 | 1253 | { |
b3c5163f | 1254 | while (untrack_refs() > 0) |
9edd7ca0 PM |
1255 | schedule(); |
1256 | ||
08f6547d AD |
1257 | nf_conntrack_helper_fini(); |
1258 | nf_conntrack_proto_fini(); | |
5d0aa2cc PM |
1259 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1260 | nf_ct_extend_unregister(&nf_ct_zone_extend); | |
1261 | #endif | |
08f6547d | 1262 | } |
9fb9cbb1 | 1263 | |
08f6547d AD |
1264 | static void nf_conntrack_cleanup_net(struct net *net) |
1265 | { | |
9fb9cbb1 | 1266 | i_see_dead_people: |
274d383b | 1267 | nf_ct_iterate_cleanup(net, kill_all, NULL); |
ee254fa4 | 1268 | nf_ct_release_dying_list(net); |
49ac8713 | 1269 | if (atomic_read(&net->ct.count) != 0) { |
9fb9cbb1 YK |
1270 | schedule(); |
1271 | goto i_see_dead_people; | |
1272 | } | |
1273 | ||
400dad39 | 1274 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
d696c7bd | 1275 | net->ct.htable_size); |
a0891aa6 | 1276 | nf_conntrack_ecache_fini(net); |
d716a4df | 1277 | nf_conntrack_acct_fini(net); |
9b03f38d | 1278 | nf_conntrack_expect_fini(net); |
5b3501fa ED |
1279 | kmem_cache_destroy(net->ct.nf_conntrack_cachep); |
1280 | kfree(net->ct.slabname); | |
0d55af87 | 1281 | free_percpu(net->ct.stat); |
08f6547d AD |
1282 | } |
1283 | ||
1284 | /* Mishearing the voices in his head, our hero wonders how he's | |
1285 | supposed to kill the mall. */ | |
1286 | void nf_conntrack_cleanup(struct net *net) | |
1287 | { | |
1288 | if (net_eq(net, &init_net)) | |
1289 | rcu_assign_pointer(ip_ct_attach, NULL); | |
1290 | ||
1291 | /* This makes sure all current packets have passed through | |
1292 | netfilter framework. Roll on, two-stage module | |
1293 | delete... */ | |
1294 | synchronize_net(); | |
1295 | ||
1296 | nf_conntrack_cleanup_net(net); | |
1297 | ||
1298 | if (net_eq(net, &init_net)) { | |
1299 | rcu_assign_pointer(nf_ct_destroy, NULL); | |
1300 | nf_conntrack_cleanup_init_net(); | |
1301 | } | |
9fb9cbb1 YK |
1302 | } |
1303 | ||
ea781f19 | 1304 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls) |
9fb9cbb1 | 1305 | { |
ea781f19 ED |
1306 | struct hlist_nulls_head *hash; |
1307 | unsigned int nr_slots, i; | |
1308 | size_t sz; | |
9fb9cbb1 | 1309 | |
601e68e1 | 1310 | *vmalloced = 0; |
8e5105a0 | 1311 | |
ea781f19 ED |
1312 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1313 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
1314 | sz = nr_slots * sizeof(struct hlist_nulls_head); | |
1315 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1316 | get_order(sz)); | |
601e68e1 | 1317 | if (!hash) { |
9fb9cbb1 YK |
1318 | *vmalloced = 1; |
1319 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
6b1686a7 ED |
1320 | hash = __vmalloc(sz, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
1321 | PAGE_KERNEL); | |
9fb9cbb1 YK |
1322 | } |
1323 | ||
ea781f19 ED |
1324 | if (hash && nulls) |
1325 | for (i = 0; i < nr_slots; i++) | |
1326 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1327 | |
1328 | return hash; | |
1329 | } | |
ac565e5f | 1330 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1331 | |
fae718dd | 1332 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1333 | { |
96eb24d7 SH |
1334 | int i, bucket, vmalloced, old_vmalloced; |
1335 | unsigned int hashsize, old_size; | |
ea781f19 | 1336 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1337 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1338 | struct nf_conn *ct; |
9fb9cbb1 | 1339 | |
d696c7bd PM |
1340 | if (current->nsproxy->net_ns != &init_net) |
1341 | return -EOPNOTSUPP; | |
1342 | ||
9fb9cbb1 YK |
1343 | /* On boot, we can set this without any fancy locking. */ |
1344 | if (!nf_conntrack_htable_size) | |
1345 | return param_set_uint(val, kp); | |
1346 | ||
96eb24d7 | 1347 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1348 | if (!hashsize) |
1349 | return -EINVAL; | |
1350 | ||
ea781f19 | 1351 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1); |
9fb9cbb1 YK |
1352 | if (!hash) |
1353 | return -ENOMEM; | |
1354 | ||
76507f69 PM |
1355 | /* Lookups in the old hash might happen in parallel, which means we |
1356 | * might get false negatives during connection lookup. New connections | |
1357 | * created because of a false negative won't make it into the hash | |
1358 | * though since that required taking the lock. | |
1359 | */ | |
f8ba1aff | 1360 | spin_lock_bh(&nf_conntrack_lock); |
d696c7bd | 1361 | for (i = 0; i < init_net.ct.htable_size; i++) { |
ea781f19 ED |
1362 | while (!hlist_nulls_empty(&init_net.ct.hash[i])) { |
1363 | h = hlist_nulls_entry(init_net.ct.hash[i].first, | |
1364 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1365 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1366 | hlist_nulls_del_rcu(&h->hnnode); |
5d0aa2cc | 1367 | bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct), |
99f07e91 | 1368 | hashsize); |
ea781f19 | 1369 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1370 | } |
1371 | } | |
d696c7bd | 1372 | old_size = init_net.ct.htable_size; |
400dad39 AD |
1373 | old_vmalloced = init_net.ct.hash_vmalloc; |
1374 | old_hash = init_net.ct.hash; | |
9fb9cbb1 | 1375 | |
d696c7bd | 1376 | init_net.ct.htable_size = nf_conntrack_htable_size = hashsize; |
400dad39 AD |
1377 | init_net.ct.hash_vmalloc = vmalloced; |
1378 | init_net.ct.hash = hash; | |
f8ba1aff | 1379 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1380 | |
ac565e5f | 1381 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1382 | return 0; |
1383 | } | |
fae718dd | 1384 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1385 | |
fae718dd | 1386 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1387 | &nf_conntrack_htable_size, 0600); |
1388 | ||
5bfddbd4 ED |
1389 | void nf_ct_untracked_status_or(unsigned long bits) |
1390 | { | |
b3c5163f ED |
1391 | int cpu; |
1392 | ||
1393 | for_each_possible_cpu(cpu) | |
1394 | per_cpu(nf_conntrack_untracked, cpu).status |= bits; | |
5bfddbd4 ED |
1395 | } |
1396 | EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); | |
1397 | ||
08f6547d | 1398 | static int nf_conntrack_init_init_net(void) |
9fb9cbb1 | 1399 | { |
f205c5e0 | 1400 | int max_factor = 8; |
b3c5163f | 1401 | int ret, cpu; |
9fb9cbb1 YK |
1402 | |
1403 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1404 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1405 | if (!nf_conntrack_htable_size) { |
1406 | nf_conntrack_htable_size | |
4481374c | 1407 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1408 | / sizeof(struct hlist_head)); |
4481374c | 1409 | if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1410 | nf_conntrack_htable_size = 16384; |
1411 | if (nf_conntrack_htable_size < 32) | |
1412 | nf_conntrack_htable_size = 32; | |
1413 | ||
1414 | /* Use a max. factor of four by default to get the same max as | |
1415 | * with the old struct list_heads. When a table size is given | |
1416 | * we use the old value of 8 to avoid reducing the max. | |
1417 | * entries. */ | |
1418 | max_factor = 4; | |
9fb9cbb1 | 1419 | } |
f205c5e0 | 1420 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 1421 | |
654d0fbd | 1422 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
1423 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
1424 | nf_conntrack_max); | |
1425 | ||
e9c1b084 PM |
1426 | ret = nf_conntrack_proto_init(); |
1427 | if (ret < 0) | |
08f6547d | 1428 | goto err_proto; |
933a41e7 | 1429 | |
ceceae1b YK |
1430 | ret = nf_conntrack_helper_init(); |
1431 | if (ret < 0) | |
08f6547d AD |
1432 | goto err_helper; |
1433 | ||
5d0aa2cc PM |
1434 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1435 | ret = nf_ct_extend_register(&nf_ct_zone_extend); | |
1436 | if (ret < 0) | |
1437 | goto err_extend; | |
1438 | #endif | |
9edd7ca0 | 1439 | /* Set up fake conntrack: to never be deleted, not in any hashes */ |
b3c5163f ED |
1440 | for_each_possible_cpu(cpu) { |
1441 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
b3c5163f ED |
1442 | write_pnet(&ct->ct_net, &init_net); |
1443 | atomic_set(&ct->ct_general.use, 1); | |
1444 | } | |
9edd7ca0 | 1445 | /* - and look it like as a confirmed connection */ |
5bfddbd4 | 1446 | nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); |
08f6547d AD |
1447 | return 0; |
1448 | ||
5d0aa2cc PM |
1449 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1450 | err_extend: | |
1451 | nf_conntrack_helper_fini(); | |
1452 | #endif | |
08f6547d AD |
1453 | err_helper: |
1454 | nf_conntrack_proto_fini(); | |
1455 | err_proto: | |
08f6547d AD |
1456 | return ret; |
1457 | } | |
1458 | ||
8cc20198 ED |
1459 | /* |
1460 | * We need to use special "null" values, not used in hash table | |
1461 | */ | |
1462 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1463 | #define DYING_NULLS_VAL ((1<<30)+1) | |
1464 | ||
08f6547d AD |
1465 | static int nf_conntrack_init_net(struct net *net) |
1466 | { | |
1467 | int ret; | |
ceceae1b | 1468 | |
08f6547d | 1469 | atomic_set(&net->ct.count, 0); |
8cc20198 ED |
1470 | INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL); |
1471 | INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL); | |
08f6547d AD |
1472 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1473 | if (!net->ct.stat) { | |
1474 | ret = -ENOMEM; | |
1475 | goto err_stat; | |
1476 | } | |
5b3501fa ED |
1477 | |
1478 | net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net); | |
1479 | if (!net->ct.slabname) { | |
1480 | ret = -ENOMEM; | |
1481 | goto err_slabname; | |
1482 | } | |
1483 | ||
1484 | net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname, | |
1485 | sizeof(struct nf_conn), 0, | |
1486 | SLAB_DESTROY_BY_RCU, NULL); | |
1487 | if (!net->ct.nf_conntrack_cachep) { | |
1488 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); | |
1489 | ret = -ENOMEM; | |
1490 | goto err_cache; | |
1491 | } | |
d696c7bd PM |
1492 | |
1493 | net->ct.htable_size = nf_conntrack_htable_size; | |
1494 | net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, | |
ea781f19 | 1495 | &net->ct.hash_vmalloc, 1); |
08f6547d AD |
1496 | if (!net->ct.hash) { |
1497 | ret = -ENOMEM; | |
1498 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1499 | goto err_hash; | |
1500 | } | |
1501 | ret = nf_conntrack_expect_init(net); | |
1502 | if (ret < 0) | |
1503 | goto err_expect; | |
d716a4df | 1504 | ret = nf_conntrack_acct_init(net); |
58401572 | 1505 | if (ret < 0) |
08f6547d | 1506 | goto err_acct; |
a0891aa6 PNA |
1507 | ret = nf_conntrack_ecache_init(net); |
1508 | if (ret < 0) | |
1509 | goto err_ecache; | |
7d3cdc6b | 1510 | |
08f6547d | 1511 | return 0; |
9fb9cbb1 | 1512 | |
a0891aa6 PNA |
1513 | err_ecache: |
1514 | nf_conntrack_acct_fini(net); | |
08f6547d | 1515 | err_acct: |
9b03f38d | 1516 | nf_conntrack_expect_fini(net); |
08f6547d | 1517 | err_expect: |
400dad39 | 1518 | nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, |
d696c7bd | 1519 | net->ct.htable_size); |
6058fa6b | 1520 | err_hash: |
5b3501fa ED |
1521 | kmem_cache_destroy(net->ct.nf_conntrack_cachep); |
1522 | err_cache: | |
1523 | kfree(net->ct.slabname); | |
1524 | err_slabname: | |
0d55af87 AD |
1525 | free_percpu(net->ct.stat); |
1526 | err_stat: | |
08f6547d AD |
1527 | return ret; |
1528 | } | |
1529 | ||
f9dd09c7 JK |
1530 | s16 (*nf_ct_nat_offset)(const struct nf_conn *ct, |
1531 | enum ip_conntrack_dir dir, | |
1532 | u32 seq); | |
1533 | EXPORT_SYMBOL_GPL(nf_ct_nat_offset); | |
1534 | ||
08f6547d AD |
1535 | int nf_conntrack_init(struct net *net) |
1536 | { | |
1537 | int ret; | |
1538 | ||
1539 | if (net_eq(net, &init_net)) { | |
1540 | ret = nf_conntrack_init_init_net(); | |
1541 | if (ret < 0) | |
1542 | goto out_init_net; | |
1543 | } | |
1544 | ret = nf_conntrack_init_net(net); | |
1545 | if (ret < 0) | |
1546 | goto out_net; | |
1547 | ||
1548 | if (net_eq(net, &init_net)) { | |
1549 | /* For use by REJECT target */ | |
1550 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); | |
1551 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); | |
f9dd09c7 JK |
1552 | |
1553 | /* Howto get NAT offsets */ | |
1554 | rcu_assign_pointer(nf_ct_nat_offset, NULL); | |
08f6547d AD |
1555 | } |
1556 | return 0; | |
1557 | ||
1558 | out_net: | |
1559 | if (net_eq(net, &init_net)) | |
1560 | nf_conntrack_cleanup_init_net(); | |
1561 | out_init_net: | |
1562 | return ret; | |
9fb9cbb1 | 1563 | } |