Commit | Line | Data |
---|---|---|
9fb9cbb1 YK |
1 | /* Connection state tracking for netfilter. This is separated from, |
2 | but required by, the NAT layer; it can also be used by an iptables | |
3 | extension. */ | |
4 | ||
5 | /* (C) 1999-2001 Paul `Rusty' Russell | |
dc808fe2 | 6 | * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org> |
9fb9cbb1 YK |
7 | * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org> |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
9fb9cbb1 YK |
12 | */ |
13 | ||
9fb9cbb1 YK |
14 | #include <linux/types.h> |
15 | #include <linux/netfilter.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/skbuff.h> | |
18 | #include <linux/proc_fs.h> | |
19 | #include <linux/vmalloc.h> | |
20 | #include <linux/stddef.h> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/random.h> | |
23 | #include <linux/jhash.h> | |
24 | #include <linux/err.h> | |
25 | #include <linux/percpu.h> | |
26 | #include <linux/moduleparam.h> | |
27 | #include <linux/notifier.h> | |
28 | #include <linux/kernel.h> | |
29 | #include <linux/netdevice.h> | |
30 | #include <linux/socket.h> | |
d7fe0f24 | 31 | #include <linux/mm.h> |
9fb9cbb1 | 32 | |
9fb9cbb1 YK |
33 | #include <net/netfilter/nf_conntrack.h> |
34 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
605dcad6 | 35 | #include <net/netfilter/nf_conntrack_l4proto.h> |
77ab9cff | 36 | #include <net/netfilter/nf_conntrack_expect.h> |
9fb9cbb1 YK |
37 | #include <net/netfilter/nf_conntrack_helper.h> |
38 | #include <net/netfilter/nf_conntrack_core.h> | |
ecfab2c9 | 39 | #include <net/netfilter/nf_conntrack_extend.h> |
9fb9cbb1 | 40 | |
dc808fe2 | 41 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 42 | |
9fb9cbb1 | 43 | DEFINE_RWLOCK(nf_conntrack_lock); |
13b18339 | 44 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 YK |
45 | |
46 | /* nf_conntrack_standalone needs this */ | |
47 | atomic_t nf_conntrack_count = ATOMIC_INIT(0); | |
a999e683 | 48 | EXPORT_SYMBOL_GPL(nf_conntrack_count); |
9fb9cbb1 | 49 | |
e2b7606c | 50 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
51 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
52 | ||
94aec08e | 53 | int nf_conntrack_max __read_mostly; |
a999e683 | 54 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 55 | |
f205c5e0 | 56 | struct hlist_head *nf_conntrack_hash __read_mostly; |
13b18339 PM |
57 | EXPORT_SYMBOL_GPL(nf_conntrack_hash); |
58 | ||
e2b7606c | 59 | struct nf_conn nf_conntrack_untracked __read_mostly; |
13b18339 PM |
60 | EXPORT_SYMBOL_GPL(nf_conntrack_untracked); |
61 | ||
94aec08e | 62 | unsigned int nf_ct_log_invalid __read_mostly; |
f205c5e0 | 63 | HLIST_HEAD(unconfirmed); |
1192e403 | 64 | static int nf_conntrack_vmalloc __read_mostly; |
dacd2a1a | 65 | static struct kmem_cache *nf_conntrack_cachep __read_mostly; |
77ab9cff | 66 | |
9fb9cbb1 YK |
67 | DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); |
68 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); | |
69 | ||
9fb9cbb1 YK |
70 | static int nf_conntrack_hash_rnd_initted; |
71 | static unsigned int nf_conntrack_hash_rnd; | |
72 | ||
73 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
74 | unsigned int size, unsigned int rnd) | |
75 | { | |
76 | unsigned int a, b; | |
9b887909 SF |
77 | |
78 | a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all), | |
79 | (tuple->src.l3num << 16) | tuple->dst.protonum); | |
80 | b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all), | |
a34c4589 AV |
81 | ((__force __u16)tuple->src.u.all << 16) | |
82 | (__force __u16)tuple->dst.u.all); | |
9fb9cbb1 YK |
83 | |
84 | return jhash_2words(a, b, rnd) % size; | |
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 | ||
9fb9cbb1 YK |
93 | int |
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 YK |
102 | { |
103 | NF_CT_TUPLE_U_BLANK(tuple); | |
104 | ||
105 | tuple->src.l3num = l3num; | |
106 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
107 | return 0; | |
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 | |
e2a3123f YK |
116 | int nf_ct_get_tuplepr(const struct sk_buff *skb, |
117 | unsigned int nhoff, | |
118 | u_int16_t l3num, | |
119 | struct nf_conntrack_tuple *tuple) | |
120 | { | |
121 | struct nf_conntrack_l3proto *l3proto; | |
122 | struct nf_conntrack_l4proto *l4proto; | |
123 | unsigned int protoff; | |
124 | u_int8_t protonum; | |
125 | int ret; | |
126 | ||
127 | rcu_read_lock(); | |
128 | ||
129 | l3proto = __nf_ct_l3proto_find(l3num); | |
130 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
131 | if (ret != NF_ACCEPT) { | |
132 | rcu_read_unlock(); | |
133 | return 0; | |
134 | } | |
135 | ||
136 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
137 | ||
138 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
139 | l3proto, l4proto); | |
140 | ||
141 | rcu_read_unlock(); | |
142 | return ret; | |
143 | } | |
144 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
145 | ||
9fb9cbb1 YK |
146 | int |
147 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, | |
148 | const struct nf_conntrack_tuple *orig, | |
149 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 150 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 YK |
151 | { |
152 | NF_CT_TUPLE_U_BLANK(inverse); | |
153 | ||
154 | inverse->src.l3num = orig->src.l3num; | |
155 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
156 | return 0; | |
157 | ||
158 | inverse->dst.dir = !orig->dst.dir; | |
159 | ||
160 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 161 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 162 | } |
13b18339 | 163 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 164 | |
9fb9cbb1 YK |
165 | static void |
166 | clean_from_lists(struct nf_conn *ct) | |
167 | { | |
0d53778e | 168 | pr_debug("clean_from_lists(%p)\n", ct); |
f205c5e0 PM |
169 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
170 | hlist_del(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); | |
9fb9cbb1 YK |
171 | |
172 | /* Destroy all pending expectations */ | |
c1d10adb | 173 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
174 | } |
175 | ||
176 | static void | |
177 | destroy_conntrack(struct nf_conntrack *nfct) | |
178 | { | |
179 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 180 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 181 | |
0d53778e | 182 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
183 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
184 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
185 | ||
186 | nf_conntrack_event(IPCT_DESTROY, ct); | |
187 | set_bit(IPS_DYING_BIT, &ct->status); | |
188 | ||
189 | /* To make sure we don't get any weird locking issues here: | |
190 | * destroy_conntrack() MUST NOT be called with a write lock | |
191 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 192 | rcu_read_lock(); |
923f4902 PM |
193 | l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, |
194 | ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); | |
605dcad6 MJ |
195 | if (l4proto && l4proto->destroy) |
196 | l4proto->destroy(ct); | |
9fb9cbb1 | 197 | |
ecfab2c9 YK |
198 | nf_ct_ext_destroy(ct); |
199 | ||
982d9a9c | 200 | rcu_read_unlock(); |
9fb9cbb1 YK |
201 | |
202 | write_lock_bh(&nf_conntrack_lock); | |
203 | /* Expectations will have been removed in clean_from_lists, | |
204 | * except TFTP can create an expectation on the first packet, | |
205 | * before connection is in the list, so we need to clean here, | |
206 | * too. */ | |
c1d10adb | 207 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
208 | |
209 | /* We overload first tuple to link into unconfirmed list. */ | |
210 | if (!nf_ct_is_confirmed(ct)) { | |
f205c5e0 PM |
211 | BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode)); |
212 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); | |
9fb9cbb1 YK |
213 | } |
214 | ||
215 | NF_CT_STAT_INC(delete); | |
216 | write_unlock_bh(&nf_conntrack_lock); | |
217 | ||
218 | if (ct->master) | |
219 | nf_ct_put(ct->master); | |
220 | ||
0d53778e | 221 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
222 | nf_conntrack_free(ct); |
223 | } | |
224 | ||
225 | static void death_by_timeout(unsigned long ul_conntrack) | |
226 | { | |
227 | struct nf_conn *ct = (void *)ul_conntrack; | |
5397e97d | 228 | struct nf_conn_help *help = nfct_help(ct); |
3c158f7f | 229 | struct nf_conntrack_helper *helper; |
5397e97d | 230 | |
3c158f7f PM |
231 | if (help) { |
232 | rcu_read_lock(); | |
233 | helper = rcu_dereference(help->helper); | |
234 | if (helper && helper->destroy) | |
235 | helper->destroy(ct); | |
236 | rcu_read_unlock(); | |
237 | } | |
9fb9cbb1 YK |
238 | |
239 | write_lock_bh(&nf_conntrack_lock); | |
240 | /* Inside lock so preempt is disabled on module removal path. | |
241 | * Otherwise we can get spurious warnings. */ | |
242 | NF_CT_STAT_INC(delete_list); | |
243 | clean_from_lists(ct); | |
244 | write_unlock_bh(&nf_conntrack_lock); | |
245 | nf_ct_put(ct); | |
246 | } | |
247 | ||
c1d10adb | 248 | struct nf_conntrack_tuple_hash * |
9fb9cbb1 YK |
249 | __nf_conntrack_find(const struct nf_conntrack_tuple *tuple, |
250 | const struct nf_conn *ignored_conntrack) | |
251 | { | |
252 | struct nf_conntrack_tuple_hash *h; | |
f205c5e0 | 253 | struct hlist_node *n; |
9fb9cbb1 YK |
254 | unsigned int hash = hash_conntrack(tuple); |
255 | ||
f205c5e0 | 256 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) { |
df0933dc PM |
257 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
258 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
9fb9cbb1 YK |
259 | NF_CT_STAT_INC(found); |
260 | return h; | |
261 | } | |
262 | NF_CT_STAT_INC(searched); | |
263 | } | |
264 | ||
265 | return NULL; | |
266 | } | |
13b18339 | 267 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
268 | |
269 | /* Find a connection corresponding to a tuple. */ | |
270 | struct nf_conntrack_tuple_hash * | |
330f7db5 | 271 | nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
272 | { |
273 | struct nf_conntrack_tuple_hash *h; | |
274 | ||
275 | read_lock_bh(&nf_conntrack_lock); | |
330f7db5 | 276 | h = __nf_conntrack_find(tuple, NULL); |
9fb9cbb1 YK |
277 | if (h) |
278 | atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use); | |
279 | read_unlock_bh(&nf_conntrack_lock); | |
280 | ||
281 | return h; | |
282 | } | |
13b18339 | 283 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 284 | |
c1d10adb PNA |
285 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
286 | unsigned int hash, | |
601e68e1 | 287 | unsigned int repl_hash) |
c1d10adb | 288 | { |
f205c5e0 PM |
289 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
290 | &nf_conntrack_hash[hash]); | |
291 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, | |
292 | &nf_conntrack_hash[repl_hash]); | |
c1d10adb PNA |
293 | } |
294 | ||
295 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
296 | { | |
297 | unsigned int hash, repl_hash; | |
298 | ||
299 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
300 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
301 | ||
302 | write_lock_bh(&nf_conntrack_lock); | |
303 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
304 | write_unlock_bh(&nf_conntrack_lock); | |
305 | } | |
13b18339 | 306 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 307 | |
9fb9cbb1 YK |
308 | /* Confirm a connection given skb; places it in hash table */ |
309 | int | |
3db05fea | 310 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
311 | { |
312 | unsigned int hash, repl_hash; | |
df0933dc | 313 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 314 | struct nf_conn *ct; |
df0933dc | 315 | struct nf_conn_help *help; |
f205c5e0 | 316 | struct hlist_node *n; |
9fb9cbb1 YK |
317 | enum ip_conntrack_info ctinfo; |
318 | ||
3db05fea | 319 | ct = nf_ct_get(skb, &ctinfo); |
9fb9cbb1 YK |
320 | |
321 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
322 | ICMP/TCP RST packets in other direction. Actual packet | |
323 | which created connection will be IP_CT_NEW or for an | |
324 | expected connection, IP_CT_RELATED. */ | |
325 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
326 | return NF_ACCEPT; | |
327 | ||
328 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
329 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
330 | ||
331 | /* We're not in hash table, and we refuse to set up related | |
332 | connections for unconfirmed conns. But packet copies and | |
333 | REJECT will give spurious warnings here. */ | |
334 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
335 | ||
336 | /* No external references means noone else could have | |
337 | confirmed us. */ | |
338 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 339 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 YK |
340 | |
341 | write_lock_bh(&nf_conntrack_lock); | |
342 | ||
343 | /* See if there's one in the list already, including reverse: | |
344 | NAT could have grabbed it without realizing, since we're | |
345 | not in the hash. If there is, we lost race. */ | |
f205c5e0 | 346 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) |
df0933dc PM |
347 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
348 | &h->tuple)) | |
349 | goto out; | |
f205c5e0 | 350 | hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode) |
df0933dc PM |
351 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
352 | &h->tuple)) | |
353 | goto out; | |
9fb9cbb1 | 354 | |
df0933dc | 355 | /* Remove from unconfirmed list */ |
f205c5e0 | 356 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
df0933dc PM |
357 | |
358 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
359 | /* Timer relative to confirmation time, not original | |
360 | setting time, otherwise we'd get timer wrap in | |
361 | weird delay cases. */ | |
362 | ct->timeout.expires += jiffies; | |
363 | add_timer(&ct->timeout); | |
364 | atomic_inc(&ct->ct_general.use); | |
365 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
366 | NF_CT_STAT_INC(insert); | |
367 | write_unlock_bh(&nf_conntrack_lock); | |
368 | help = nfct_help(ct); | |
369 | if (help && help->helper) | |
3db05fea | 370 | nf_conntrack_event_cache(IPCT_HELPER, skb); |
9fb9cbb1 | 371 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
372 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
373 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
3db05fea | 374 | nf_conntrack_event_cache(IPCT_NATINFO, skb); |
9fb9cbb1 | 375 | #endif |
df0933dc | 376 | nf_conntrack_event_cache(master_ct(ct) ? |
3db05fea | 377 | IPCT_RELATED : IPCT_NEW, skb); |
df0933dc | 378 | return NF_ACCEPT; |
9fb9cbb1 | 379 | |
df0933dc | 380 | out: |
9fb9cbb1 YK |
381 | NF_CT_STAT_INC(insert_failed); |
382 | write_unlock_bh(&nf_conntrack_lock); | |
383 | return NF_DROP; | |
384 | } | |
13b18339 | 385 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
386 | |
387 | /* Returns true if a connection correspondings to the tuple (required | |
388 | for NAT). */ | |
389 | int | |
390 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
391 | const struct nf_conn *ignored_conntrack) | |
392 | { | |
393 | struct nf_conntrack_tuple_hash *h; | |
394 | ||
395 | read_lock_bh(&nf_conntrack_lock); | |
396 | h = __nf_conntrack_find(tuple, ignored_conntrack); | |
397 | read_unlock_bh(&nf_conntrack_lock); | |
398 | ||
399 | return h != NULL; | |
400 | } | |
13b18339 | 401 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 402 | |
7ae7730f PM |
403 | #define NF_CT_EVICTION_RANGE 8 |
404 | ||
9fb9cbb1 YK |
405 | /* There's a small race here where we may free a just-assured |
406 | connection. Too bad: we're in trouble anyway. */ | |
7ae7730f | 407 | static int early_drop(unsigned int hash) |
9fb9cbb1 | 408 | { |
f205c5e0 | 409 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 410 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 411 | struct nf_conn *ct = NULL, *tmp; |
f205c5e0 | 412 | struct hlist_node *n; |
7ae7730f | 413 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
414 | int dropped = 0; |
415 | ||
416 | read_lock_bh(&nf_conntrack_lock); | |
7ae7730f PM |
417 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
418 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) { | |
419 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
420 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
421 | ct = tmp; | |
422 | cnt++; | |
423 | } | |
424 | if (ct || cnt >= NF_CT_EVICTION_RANGE) | |
425 | break; | |
426 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 427 | } |
f205c5e0 PM |
428 | if (ct) |
429 | atomic_inc(&ct->ct_general.use); | |
9fb9cbb1 YK |
430 | read_unlock_bh(&nf_conntrack_lock); |
431 | ||
432 | if (!ct) | |
433 | return dropped; | |
434 | ||
435 | if (del_timer(&ct->timeout)) { | |
436 | death_by_timeout((unsigned long)ct); | |
437 | dropped = 1; | |
c0e912d7 | 438 | NF_CT_STAT_INC_ATOMIC(early_drop); |
9fb9cbb1 YK |
439 | } |
440 | nf_ct_put(ct); | |
441 | return dropped; | |
442 | } | |
443 | ||
dacd2a1a YK |
444 | struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, |
445 | const struct nf_conntrack_tuple *repl) | |
9fb9cbb1 YK |
446 | { |
447 | struct nf_conn *conntrack = NULL; | |
9fb9cbb1 | 448 | |
dc808fe2 | 449 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
450 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
451 | nf_conntrack_hash_rnd_initted = 1; | |
452 | } | |
453 | ||
5251e2d2 PNA |
454 | /* We don't want any race condition at early drop stage */ |
455 | atomic_inc(&nf_conntrack_count); | |
456 | ||
9fb9cbb1 | 457 | if (nf_conntrack_max |
5251e2d2 | 458 | && atomic_read(&nf_conntrack_count) > nf_conntrack_max) { |
9fb9cbb1 | 459 | unsigned int hash = hash_conntrack(orig); |
7ae7730f | 460 | if (!early_drop(hash)) { |
5251e2d2 | 461 | atomic_dec(&nf_conntrack_count); |
9fb9cbb1 YK |
462 | if (net_ratelimit()) |
463 | printk(KERN_WARNING | |
464 | "nf_conntrack: table full, dropping" | |
465 | " packet.\n"); | |
466 | return ERR_PTR(-ENOMEM); | |
467 | } | |
468 | } | |
469 | ||
dacd2a1a | 470 | conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC); |
9fb9cbb1 | 471 | if (conntrack == NULL) { |
0d53778e | 472 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
dacd2a1a YK |
473 | atomic_dec(&nf_conntrack_count); |
474 | return ERR_PTR(-ENOMEM); | |
9fb9cbb1 YK |
475 | } |
476 | ||
9fb9cbb1 | 477 | atomic_set(&conntrack->ct_general.use, 1); |
9fb9cbb1 YK |
478 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
479 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
480 | /* Don't set timer yet: wait for confirmation */ | |
e6f689db PM |
481 | setup_timer(&conntrack->timeout, death_by_timeout, |
482 | (unsigned long)conntrack); | |
9fb9cbb1 | 483 | |
9fb9cbb1 YK |
484 | return conntrack; |
485 | } | |
13b18339 | 486 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 YK |
487 | |
488 | void nf_conntrack_free(struct nf_conn *conntrack) | |
489 | { | |
ecfab2c9 | 490 | nf_ct_ext_free(conntrack); |
dacd2a1a | 491 | kmem_cache_free(nf_conntrack_cachep, conntrack); |
9fb9cbb1 YK |
492 | atomic_dec(&nf_conntrack_count); |
493 | } | |
13b18339 | 494 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
495 | |
496 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
497 | failed due to stress. Otherwise it really is unclassifiable. */ | |
498 | static struct nf_conntrack_tuple_hash * | |
499 | init_conntrack(const struct nf_conntrack_tuple *tuple, | |
500 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 501 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
502 | struct sk_buff *skb, |
503 | unsigned int dataoff) | |
504 | { | |
505 | struct nf_conn *conntrack; | |
3c158f7f | 506 | struct nf_conn_help *help; |
9fb9cbb1 YK |
507 | struct nf_conntrack_tuple repl_tuple; |
508 | struct nf_conntrack_expect *exp; | |
509 | ||
605dcad6 | 510 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 511 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
512 | return NULL; |
513 | } | |
514 | ||
dacd2a1a | 515 | conntrack = nf_conntrack_alloc(tuple, &repl_tuple); |
9fb9cbb1 | 516 | if (conntrack == NULL || IS_ERR(conntrack)) { |
0d53778e | 517 | pr_debug("Can't allocate conntrack.\n"); |
9fb9cbb1 YK |
518 | return (struct nf_conntrack_tuple_hash *)conntrack; |
519 | } | |
520 | ||
605dcad6 | 521 | if (!l4proto->new(conntrack, skb, dataoff)) { |
9fb9cbb1 | 522 | nf_conntrack_free(conntrack); |
0d53778e | 523 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
524 | return NULL; |
525 | } | |
526 | ||
527 | write_lock_bh(&nf_conntrack_lock); | |
6823645d | 528 | exp = nf_ct_find_expectation(tuple); |
9fb9cbb1 | 529 | if (exp) { |
0d53778e PM |
530 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
531 | conntrack, exp); | |
9fb9cbb1 YK |
532 | /* Welcome, Mr. Bond. We've been expecting you... */ |
533 | __set_bit(IPS_EXPECTED_BIT, &conntrack->status); | |
534 | conntrack->master = exp->master; | |
ceceae1b | 535 | if (exp->helper) { |
b560580a | 536 | help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC); |
ceceae1b YK |
537 | if (help) |
538 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
539 | } |
540 | ||
9fb9cbb1 YK |
541 | #ifdef CONFIG_NF_CONNTRACK_MARK |
542 | conntrack->mark = exp->master->mark; | |
7c9728c3 JM |
543 | #endif |
544 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
545 | conntrack->secmark = exp->master->secmark; | |
9fb9cbb1 YK |
546 | #endif |
547 | nf_conntrack_get(&conntrack->master->ct_general); | |
548 | NF_CT_STAT_INC(expect_new); | |
22e7410b | 549 | } else { |
ceceae1b YK |
550 | struct nf_conntrack_helper *helper; |
551 | ||
552 | helper = __nf_ct_helper_find(&repl_tuple); | |
553 | if (helper) { | |
b560580a | 554 | help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC); |
ceceae1b | 555 | if (help) |
ceceae1b | 556 | rcu_assign_pointer(help->helper, helper); |
3c158f7f | 557 | } |
9fb9cbb1 | 558 | NF_CT_STAT_INC(new); |
22e7410b | 559 | } |
9fb9cbb1 YK |
560 | |
561 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
f205c5e0 PM |
562 | hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
563 | &unconfirmed); | |
9fb9cbb1 YK |
564 | |
565 | write_unlock_bh(&nf_conntrack_lock); | |
566 | ||
567 | if (exp) { | |
568 | if (exp->expectfn) | |
569 | exp->expectfn(conntrack, exp); | |
6823645d | 570 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
571 | } |
572 | ||
573 | return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; | |
574 | } | |
575 | ||
576 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
577 | static inline struct nf_conn * | |
578 | resolve_normal_ct(struct sk_buff *skb, | |
579 | unsigned int dataoff, | |
580 | u_int16_t l3num, | |
581 | u_int8_t protonum, | |
582 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 583 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
584 | int *set_reply, |
585 | enum ip_conntrack_info *ctinfo) | |
586 | { | |
587 | struct nf_conntrack_tuple tuple; | |
588 | struct nf_conntrack_tuple_hash *h; | |
589 | struct nf_conn *ct; | |
590 | ||
bbe735e4 | 591 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 592 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 593 | l4proto)) { |
0d53778e | 594 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
595 | return NULL; |
596 | } | |
597 | ||
598 | /* look for tuple match */ | |
330f7db5 | 599 | h = nf_conntrack_find_get(&tuple); |
9fb9cbb1 | 600 | if (!h) { |
605dcad6 | 601 | h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
602 | if (!h) |
603 | return NULL; | |
604 | if (IS_ERR(h)) | |
605 | return (void *)h; | |
606 | } | |
607 | ct = nf_ct_tuplehash_to_ctrack(h); | |
608 | ||
609 | /* It exists; we have (non-exclusive) reference. */ | |
610 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
611 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
612 | /* Please set reply bit if this packet OK */ | |
613 | *set_reply = 1; | |
614 | } else { | |
615 | /* Once we've had two way comms, always ESTABLISHED. */ | |
616 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 617 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
618 | *ctinfo = IP_CT_ESTABLISHED; |
619 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
620 | pr_debug("nf_conntrack_in: related packet for %p\n", |
621 | ct); | |
9fb9cbb1 YK |
622 | *ctinfo = IP_CT_RELATED; |
623 | } else { | |
0d53778e | 624 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
625 | *ctinfo = IP_CT_NEW; |
626 | } | |
627 | *set_reply = 0; | |
628 | } | |
629 | skb->nfct = &ct->ct_general; | |
630 | skb->nfctinfo = *ctinfo; | |
631 | return ct; | |
632 | } | |
633 | ||
634 | unsigned int | |
3db05fea | 635 | nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb) |
9fb9cbb1 YK |
636 | { |
637 | struct nf_conn *ct; | |
638 | enum ip_conntrack_info ctinfo; | |
639 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 640 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
641 | unsigned int dataoff; |
642 | u_int8_t protonum; | |
643 | int set_reply = 0; | |
644 | int ret; | |
645 | ||
646 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 647 | if (skb->nfct) { |
c0e912d7 | 648 | NF_CT_STAT_INC_ATOMIC(ignore); |
9fb9cbb1 YK |
649 | return NF_ACCEPT; |
650 | } | |
651 | ||
923f4902 | 652 | /* rcu_read_lock()ed by nf_hook_slow */ |
c1d10adb | 653 | l3proto = __nf_ct_l3proto_find((u_int16_t)pf); |
3db05fea | 654 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
655 | &dataoff, &protonum); |
656 | if (ret <= 0) { | |
0d53778e | 657 | pr_debug("not prepared to track yet or error occured\n"); |
d87d8469 YK |
658 | NF_CT_STAT_INC_ATOMIC(error); |
659 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
660 | return -ret; |
661 | } | |
662 | ||
605dcad6 | 663 | l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); |
9fb9cbb1 YK |
664 | |
665 | /* It may be an special packet, error, unclean... | |
666 | * inverse of the return code tells to the netfilter | |
667 | * core what to do with the packet. */ | |
605dcad6 | 668 | if (l4proto->error != NULL && |
3db05fea | 669 | (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) { |
c0e912d7 PM |
670 | NF_CT_STAT_INC_ATOMIC(error); |
671 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
672 | return -ret; |
673 | } | |
674 | ||
3db05fea | 675 | ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto, |
9fb9cbb1 YK |
676 | &set_reply, &ctinfo); |
677 | if (!ct) { | |
678 | /* Not valid part of a connection */ | |
c0e912d7 | 679 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
680 | return NF_ACCEPT; |
681 | } | |
682 | ||
683 | if (IS_ERR(ct)) { | |
684 | /* Too stressed to deal. */ | |
c0e912d7 | 685 | NF_CT_STAT_INC_ATOMIC(drop); |
9fb9cbb1 YK |
686 | return NF_DROP; |
687 | } | |
688 | ||
3db05fea | 689 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 690 | |
3db05fea | 691 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
692 | if (ret < 0) { |
693 | /* Invalid: inverse of the return code tells | |
694 | * the netfilter core what to do */ | |
0d53778e | 695 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
696 | nf_conntrack_put(skb->nfct); |
697 | skb->nfct = NULL; | |
c0e912d7 | 698 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
699 | return -ret; |
700 | } | |
701 | ||
702 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
3db05fea | 703 | nf_conntrack_event_cache(IPCT_STATUS, skb); |
9fb9cbb1 YK |
704 | |
705 | return ret; | |
706 | } | |
13b18339 | 707 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 YK |
708 | |
709 | int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, | |
710 | const struct nf_conntrack_tuple *orig) | |
711 | { | |
923f4902 PM |
712 | int ret; |
713 | ||
714 | rcu_read_lock(); | |
715 | ret = nf_ct_invert_tuple(inverse, orig, | |
716 | __nf_ct_l3proto_find(orig->src.l3num), | |
717 | __nf_ct_l4proto_find(orig->src.l3num, | |
718 | orig->dst.protonum)); | |
719 | rcu_read_unlock(); | |
720 | return ret; | |
9fb9cbb1 | 721 | } |
13b18339 | 722 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 723 | |
5b1158e9 JK |
724 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
725 | implicitly racy: see __nf_conntrack_confirm */ | |
726 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
727 | const struct nf_conntrack_tuple *newreply) | |
728 | { | |
729 | struct nf_conn_help *help = nfct_help(ct); | |
ceceae1b | 730 | struct nf_conntrack_helper *helper; |
5b1158e9 JK |
731 | |
732 | write_lock_bh(&nf_conntrack_lock); | |
733 | /* Should be unconfirmed, so not in hash table yet */ | |
734 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
735 | ||
0d53778e | 736 | pr_debug("Altering reply tuple of %p to ", ct); |
5b1158e9 JK |
737 | NF_CT_DUMP_TUPLE(newreply); |
738 | ||
739 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ceceae1b YK |
740 | if (ct->master || (help && help->expecting != 0)) |
741 | goto out; | |
742 | ||
743 | helper = __nf_ct_helper_find(newreply); | |
744 | if (helper == NULL) { | |
745 | if (help) | |
746 | rcu_assign_pointer(help->helper, NULL); | |
747 | goto out; | |
5d78a849 | 748 | } |
ceceae1b YK |
749 | |
750 | if (help == NULL) { | |
b560580a PM |
751 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
752 | if (help == NULL) | |
ceceae1b | 753 | goto out; |
ceceae1b YK |
754 | } else { |
755 | memset(&help->help, 0, sizeof(help->help)); | |
756 | } | |
757 | ||
758 | rcu_assign_pointer(help->helper, helper); | |
759 | out: | |
5b1158e9 JK |
760 | write_unlock_bh(&nf_conntrack_lock); |
761 | } | |
13b18339 | 762 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 763 | |
9fb9cbb1 YK |
764 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
765 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
766 | enum ip_conntrack_info ctinfo, | |
767 | const struct sk_buff *skb, | |
768 | unsigned long extra_jiffies, | |
769 | int do_acct) | |
770 | { | |
771 | int event = 0; | |
772 | ||
773 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
774 | NF_CT_ASSERT(skb); | |
775 | ||
776 | write_lock_bh(&nf_conntrack_lock); | |
777 | ||
997ae831 EL |
778 | /* Only update if this is not a fixed timeout */ |
779 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) { | |
780 | write_unlock_bh(&nf_conntrack_lock); | |
781 | return; | |
782 | } | |
783 | ||
9fb9cbb1 YK |
784 | /* If not in hash table, timer will not be active yet */ |
785 | if (!nf_ct_is_confirmed(ct)) { | |
786 | ct->timeout.expires = extra_jiffies; | |
787 | event = IPCT_REFRESH; | |
788 | } else { | |
be00c8e4 MJ |
789 | unsigned long newtime = jiffies + extra_jiffies; |
790 | ||
791 | /* Only update the timeout if the new timeout is at least | |
792 | HZ jiffies from the old timeout. Need del_timer for race | |
793 | avoidance (may already be dying). */ | |
794 | if (newtime - ct->timeout.expires >= HZ | |
795 | && del_timer(&ct->timeout)) { | |
796 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
797 | add_timer(&ct->timeout); |
798 | event = IPCT_REFRESH; | |
799 | } | |
800 | } | |
801 | ||
802 | #ifdef CONFIG_NF_CT_ACCT | |
803 | if (do_acct) { | |
804 | ct->counters[CTINFO2DIR(ctinfo)].packets++; | |
805 | ct->counters[CTINFO2DIR(ctinfo)].bytes += | |
bbe735e4 | 806 | skb->len - skb_network_offset(skb); |
3ffd5eeb MJ |
807 | |
808 | if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) | |
809 | || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) | |
810 | event |= IPCT_COUNTER_FILLING; | |
9fb9cbb1 YK |
811 | } |
812 | #endif | |
813 | ||
814 | write_unlock_bh(&nf_conntrack_lock); | |
815 | ||
816 | /* must be unlocked when calling event cache */ | |
817 | if (event) | |
818 | nf_conntrack_event_cache(event, skb); | |
819 | } | |
13b18339 | 820 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 821 | |
e281db5c | 822 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
823 | |
824 | #include <linux/netfilter/nfnetlink.h> | |
825 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
826 | #include <linux/mutex.h> |
827 | ||
c1d10adb PNA |
828 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
829 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
830 | * or depend on ctnetlink */ | |
fdf70832 | 831 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
832 | const struct nf_conntrack_tuple *tuple) |
833 | { | |
df6fb868 | 834 | NLA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t), |
c1d10adb | 835 | &tuple->src.u.tcp.port); |
df6fb868 | 836 | NLA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t), |
c1d10adb PNA |
837 | &tuple->dst.u.tcp.port); |
838 | return 0; | |
839 | ||
df6fb868 | 840 | nla_put_failure: |
c1d10adb PNA |
841 | return -1; |
842 | } | |
fdf70832 | 843 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 844 | |
f73e924c PM |
845 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
846 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
847 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 848 | }; |
f73e924c | 849 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 850 | |
fdf70832 | 851 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
852 | struct nf_conntrack_tuple *t) |
853 | { | |
df6fb868 | 854 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
855 | return -EINVAL; |
856 | ||
df6fb868 PM |
857 | t->src.u.tcp.port = *(__be16 *)nla_data(tb[CTA_PROTO_SRC_PORT]); |
858 | t->dst.u.tcp.port = *(__be16 *)nla_data(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
859 | |
860 | return 0; | |
861 | } | |
fdf70832 | 862 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
c1d10adb PNA |
863 | #endif |
864 | ||
9fb9cbb1 YK |
865 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
866 | void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) | |
867 | { | |
868 | struct nf_conn *ct; | |
869 | enum ip_conntrack_info ctinfo; | |
870 | ||
871 | /* This ICMP is in reverse direction to the packet which caused it */ | |
872 | ct = nf_ct_get(skb, &ctinfo); | |
873 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
874 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
875 | else | |
876 | ctinfo = IP_CT_RELATED; | |
877 | ||
878 | /* Attach to new skbuff, and increment count */ | |
879 | nskb->nfct = &ct->ct_general; | |
880 | nskb->nfctinfo = ctinfo; | |
881 | nf_conntrack_get(nskb->nfct); | |
882 | } | |
13b18339 | 883 | EXPORT_SYMBOL_GPL(__nf_conntrack_attach); |
9fb9cbb1 YK |
884 | |
885 | static inline int | |
886 | do_iter(const struct nf_conntrack_tuple_hash *i, | |
887 | int (*iter)(struct nf_conn *i, void *data), | |
888 | void *data) | |
889 | { | |
890 | return iter(nf_ct_tuplehash_to_ctrack(i), data); | |
891 | } | |
892 | ||
893 | /* Bring out ya dead! */ | |
df0933dc | 894 | static struct nf_conn * |
9fb9cbb1 YK |
895 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
896 | void *data, unsigned int *bucket) | |
897 | { | |
df0933dc PM |
898 | struct nf_conntrack_tuple_hash *h; |
899 | struct nf_conn *ct; | |
f205c5e0 | 900 | struct hlist_node *n; |
9fb9cbb1 YK |
901 | |
902 | write_lock_bh(&nf_conntrack_lock); | |
903 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { | |
f205c5e0 | 904 | hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) { |
df0933dc PM |
905 | ct = nf_ct_tuplehash_to_ctrack(h); |
906 | if (iter(ct, data)) | |
907 | goto found; | |
908 | } | |
601e68e1 | 909 | } |
f205c5e0 | 910 | hlist_for_each_entry(h, n, &unconfirmed, hnode) { |
df0933dc PM |
911 | ct = nf_ct_tuplehash_to_ctrack(h); |
912 | if (iter(ct, data)) | |
ec68e97d | 913 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 914 | } |
c073e3fa | 915 | write_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
916 | return NULL; |
917 | found: | |
c073e3fa | 918 | atomic_inc(&ct->ct_general.use); |
9fb9cbb1 | 919 | write_unlock_bh(&nf_conntrack_lock); |
df0933dc | 920 | return ct; |
9fb9cbb1 YK |
921 | } |
922 | ||
923 | void | |
924 | nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) | |
925 | { | |
df0933dc | 926 | struct nf_conn *ct; |
9fb9cbb1 YK |
927 | unsigned int bucket = 0; |
928 | ||
df0933dc | 929 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
930 | /* Time to push up daises... */ |
931 | if (del_timer(&ct->timeout)) | |
932 | death_by_timeout((unsigned long)ct); | |
933 | /* ... else the timer will get him soon. */ | |
934 | ||
935 | nf_ct_put(ct); | |
936 | } | |
937 | } | |
13b18339 | 938 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 YK |
939 | |
940 | static int kill_all(struct nf_conn *i, void *data) | |
941 | { | |
942 | return 1; | |
943 | } | |
944 | ||
ac565e5f | 945 | void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, int size) |
9fb9cbb1 YK |
946 | { |
947 | if (vmalloced) | |
948 | vfree(hash); | |
949 | else | |
601e68e1 | 950 | free_pages((unsigned long)hash, |
f205c5e0 | 951 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 952 | } |
ac565e5f | 953 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 954 | |
272491ef | 955 | void nf_conntrack_flush(void) |
c1d10adb PNA |
956 | { |
957 | nf_ct_iterate_cleanup(kill_all, NULL); | |
958 | } | |
13b18339 | 959 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 960 | |
9fb9cbb1 YK |
961 | /* Mishearing the voices in his head, our hero wonders how he's |
962 | supposed to kill the mall. */ | |
963 | void nf_conntrack_cleanup(void) | |
964 | { | |
c3a47ab3 | 965 | rcu_assign_pointer(ip_ct_attach, NULL); |
7d3cdc6b | 966 | |
9fb9cbb1 YK |
967 | /* This makes sure all current packets have passed through |
968 | netfilter framework. Roll on, two-stage module | |
969 | delete... */ | |
970 | synchronize_net(); | |
971 | ||
972 | nf_ct_event_cache_flush(); | |
973 | i_see_dead_people: | |
c1d10adb | 974 | nf_conntrack_flush(); |
9fb9cbb1 YK |
975 | if (atomic_read(&nf_conntrack_count) != 0) { |
976 | schedule(); | |
977 | goto i_see_dead_people; | |
978 | } | |
6636568c PM |
979 | /* wait until all references to nf_conntrack_untracked are dropped */ |
980 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
981 | schedule(); | |
9fb9cbb1 | 982 | |
de6e05c4 YK |
983 | rcu_assign_pointer(nf_ct_destroy, NULL); |
984 | ||
dacd2a1a | 985 | kmem_cache_destroy(nf_conntrack_cachep); |
ac565e5f PM |
986 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
987 | nf_conntrack_htable_size); | |
5a6f294e | 988 | |
ac5357eb | 989 | nf_conntrack_proto_fini(); |
ceceae1b | 990 | nf_conntrack_helper_fini(); |
e9c1b084 | 991 | nf_conntrack_expect_fini(); |
9fb9cbb1 YK |
992 | } |
993 | ||
ac565e5f | 994 | struct hlist_head *nf_ct_alloc_hashtable(int *sizep, int *vmalloced) |
9fb9cbb1 | 995 | { |
f205c5e0 | 996 | struct hlist_head *hash; |
8e5105a0 | 997 | unsigned int size, i; |
9fb9cbb1 | 998 | |
601e68e1 | 999 | *vmalloced = 0; |
8e5105a0 | 1000 | |
f205c5e0 | 1001 | size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); |
601e68e1 | 1002 | hash = (void*)__get_free_pages(GFP_KERNEL, |
f205c5e0 | 1003 | get_order(sizeof(struct hlist_head) |
9fb9cbb1 | 1004 | * size)); |
601e68e1 | 1005 | if (!hash) { |
9fb9cbb1 YK |
1006 | *vmalloced = 1; |
1007 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
f205c5e0 | 1008 | hash = vmalloc(sizeof(struct hlist_head) * size); |
9fb9cbb1 YK |
1009 | } |
1010 | ||
1011 | if (hash) | |
601e68e1 | 1012 | for (i = 0; i < size; i++) |
f205c5e0 | 1013 | INIT_HLIST_HEAD(&hash[i]); |
9fb9cbb1 YK |
1014 | |
1015 | return hash; | |
1016 | } | |
ac565e5f | 1017 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 YK |
1018 | |
1019 | int set_hashsize(const char *val, struct kernel_param *kp) | |
1020 | { | |
1021 | int i, bucket, hashsize, vmalloced; | |
1022 | int old_vmalloced, old_size; | |
1023 | int rnd; | |
f205c5e0 | 1024 | struct hlist_head *hash, *old_hash; |
9fb9cbb1 YK |
1025 | struct nf_conntrack_tuple_hash *h; |
1026 | ||
1027 | /* On boot, we can set this without any fancy locking. */ | |
1028 | if (!nf_conntrack_htable_size) | |
1029 | return param_set_uint(val, kp); | |
1030 | ||
1031 | hashsize = simple_strtol(val, NULL, 0); | |
1032 | if (!hashsize) | |
1033 | return -EINVAL; | |
1034 | ||
ac565e5f | 1035 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); |
9fb9cbb1 YK |
1036 | if (!hash) |
1037 | return -ENOMEM; | |
1038 | ||
1039 | /* We have to rehahs for the new table anyway, so we also can | |
1040 | * use a newrandom seed */ | |
1041 | get_random_bytes(&rnd, 4); | |
1042 | ||
1043 | write_lock_bh(&nf_conntrack_lock); | |
1044 | for (i = 0; i < nf_conntrack_htable_size; i++) { | |
f205c5e0 PM |
1045 | while (!hlist_empty(&nf_conntrack_hash[i])) { |
1046 | h = hlist_entry(nf_conntrack_hash[i].first, | |
1047 | struct nf_conntrack_tuple_hash, hnode); | |
1048 | hlist_del(&h->hnode); | |
9fb9cbb1 | 1049 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
f205c5e0 | 1050 | hlist_add_head(&h->hnode, &hash[bucket]); |
9fb9cbb1 YK |
1051 | } |
1052 | } | |
1053 | old_size = nf_conntrack_htable_size; | |
1054 | old_vmalloced = nf_conntrack_vmalloc; | |
1055 | old_hash = nf_conntrack_hash; | |
1056 | ||
1057 | nf_conntrack_htable_size = hashsize; | |
1058 | nf_conntrack_vmalloc = vmalloced; | |
1059 | nf_conntrack_hash = hash; | |
1060 | nf_conntrack_hash_rnd = rnd; | |
1061 | write_unlock_bh(&nf_conntrack_lock); | |
1062 | ||
ac565e5f | 1063 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1064 | return 0; |
1065 | } | |
1066 | ||
1067 | module_param_call(hashsize, set_hashsize, param_get_uint, | |
1068 | &nf_conntrack_htable_size, 0600); | |
1069 | ||
1070 | int __init nf_conntrack_init(void) | |
1071 | { | |
f205c5e0 | 1072 | int max_factor = 8; |
9fb9cbb1 YK |
1073 | int ret; |
1074 | ||
1075 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1076 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1077 | if (!nf_conntrack_htable_size) { |
1078 | nf_conntrack_htable_size | |
1079 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
f205c5e0 | 1080 | / sizeof(struct hlist_head)); |
9fb9cbb1 | 1081 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1082 | nf_conntrack_htable_size = 16384; |
1083 | if (nf_conntrack_htable_size < 32) | |
1084 | nf_conntrack_htable_size = 32; | |
1085 | ||
1086 | /* Use a max. factor of four by default to get the same max as | |
1087 | * with the old struct list_heads. When a table size is given | |
1088 | * we use the old value of 8 to avoid reducing the max. | |
1089 | * entries. */ | |
1090 | max_factor = 4; | |
9fb9cbb1 | 1091 | } |
ac565e5f PM |
1092 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
1093 | &nf_conntrack_vmalloc); | |
9fb9cbb1 YK |
1094 | if (!nf_conntrack_hash) { |
1095 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1096 | goto err_out; | |
1097 | } | |
1098 | ||
f205c5e0 | 1099 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1100 | |
1101 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1102 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1103 | nf_conntrack_max); | |
1104 | ||
dacd2a1a YK |
1105 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1106 | sizeof(struct nf_conn), | |
20c2df83 | 1107 | 0, 0, NULL); |
dacd2a1a | 1108 | if (!nf_conntrack_cachep) { |
9fb9cbb1 YK |
1109 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
1110 | goto err_free_hash; | |
1111 | } | |
1112 | ||
e9c1b084 PM |
1113 | ret = nf_conntrack_proto_init(); |
1114 | if (ret < 0) | |
9fb9cbb1 | 1115 | goto err_free_conntrack_slab; |
9fb9cbb1 | 1116 | |
e9c1b084 | 1117 | ret = nf_conntrack_expect_init(); |
933a41e7 | 1118 | if (ret < 0) |
e9c1b084 | 1119 | goto out_fini_proto; |
933a41e7 | 1120 | |
ceceae1b YK |
1121 | ret = nf_conntrack_helper_init(); |
1122 | if (ret < 0) | |
e9c1b084 | 1123 | goto out_fini_expect; |
ceceae1b | 1124 | |
7d3cdc6b | 1125 | /* For use by REJECT target */ |
c3a47ab3 | 1126 | rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach); |
de6e05c4 | 1127 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); |
7d3cdc6b | 1128 | |
9fb9cbb1 YK |
1129 | /* Set up fake conntrack: |
1130 | - to never be deleted, not in any hashes */ | |
1131 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); | |
1132 | /* - and look it like as a confirmed connection */ | |
1133 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1134 | ||
1135 | return ret; | |
1136 | ||
e9c1b084 PM |
1137 | out_fini_expect: |
1138 | nf_conntrack_expect_fini(); | |
ceceae1b YK |
1139 | out_fini_proto: |
1140 | nf_conntrack_proto_fini(); | |
9fb9cbb1 | 1141 | err_free_conntrack_slab: |
dacd2a1a | 1142 | kmem_cache_destroy(nf_conntrack_cachep); |
9fb9cbb1 | 1143 | err_free_hash: |
ac565e5f PM |
1144 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
1145 | nf_conntrack_htable_size); | |
9fb9cbb1 YK |
1146 | err_out: |
1147 | return -ENOMEM; | |
1148 | } |