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; |
4e3882f7 | 66 | static unsigned int nf_conntrack_next_id; |
77ab9cff | 67 | |
9fb9cbb1 YK |
68 | DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); |
69 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); | |
70 | ||
9fb9cbb1 YK |
71 | static int nf_conntrack_hash_rnd_initted; |
72 | static unsigned int nf_conntrack_hash_rnd; | |
73 | ||
74 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
75 | unsigned int size, unsigned int rnd) | |
76 | { | |
77 | unsigned int a, b; | |
9b887909 SF |
78 | |
79 | a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all), | |
80 | (tuple->src.l3num << 16) | tuple->dst.protonum); | |
81 | b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all), | |
a34c4589 AV |
82 | ((__force __u16)tuple->src.u.all << 16) | |
83 | (__force __u16)tuple->dst.u.all); | |
9fb9cbb1 YK |
84 | |
85 | return jhash_2words(a, b, rnd) % size; | |
86 | } | |
87 | ||
88 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
89 | { | |
90 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
91 | nf_conntrack_hash_rnd); | |
92 | } | |
93 | ||
9fb9cbb1 YK |
94 | int |
95 | nf_ct_get_tuple(const struct sk_buff *skb, | |
96 | unsigned int nhoff, | |
97 | unsigned int dataoff, | |
98 | u_int16_t l3num, | |
99 | u_int8_t protonum, | |
100 | struct nf_conntrack_tuple *tuple, | |
101 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 102 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 YK |
103 | { |
104 | NF_CT_TUPLE_U_BLANK(tuple); | |
105 | ||
106 | tuple->src.l3num = l3num; | |
107 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
108 | return 0; | |
109 | ||
110 | tuple->dst.protonum = protonum; | |
111 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
112 | ||
605dcad6 | 113 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 114 | } |
13b18339 | 115 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 116 | |
e2a3123f YK |
117 | int nf_ct_get_tuplepr(const struct sk_buff *skb, |
118 | unsigned int nhoff, | |
119 | u_int16_t l3num, | |
120 | struct nf_conntrack_tuple *tuple) | |
121 | { | |
122 | struct nf_conntrack_l3proto *l3proto; | |
123 | struct nf_conntrack_l4proto *l4proto; | |
124 | unsigned int protoff; | |
125 | u_int8_t protonum; | |
126 | int ret; | |
127 | ||
128 | rcu_read_lock(); | |
129 | ||
130 | l3proto = __nf_ct_l3proto_find(l3num); | |
131 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
132 | if (ret != NF_ACCEPT) { | |
133 | rcu_read_unlock(); | |
134 | return 0; | |
135 | } | |
136 | ||
137 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
138 | ||
139 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
140 | l3proto, l4proto); | |
141 | ||
142 | rcu_read_unlock(); | |
143 | return ret; | |
144 | } | |
145 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
146 | ||
9fb9cbb1 YK |
147 | int |
148 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, | |
149 | const struct nf_conntrack_tuple *orig, | |
150 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 151 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 YK |
152 | { |
153 | NF_CT_TUPLE_U_BLANK(inverse); | |
154 | ||
155 | inverse->src.l3num = orig->src.l3num; | |
156 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
157 | return 0; | |
158 | ||
159 | inverse->dst.dir = !orig->dst.dir; | |
160 | ||
161 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 162 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 163 | } |
13b18339 | 164 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 165 | |
9fb9cbb1 YK |
166 | static void |
167 | clean_from_lists(struct nf_conn *ct) | |
168 | { | |
0d53778e | 169 | pr_debug("clean_from_lists(%p)\n", ct); |
f205c5e0 PM |
170 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
171 | hlist_del(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); | |
9fb9cbb1 YK |
172 | |
173 | /* Destroy all pending expectations */ | |
c1d10adb | 174 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
175 | } |
176 | ||
177 | static void | |
178 | destroy_conntrack(struct nf_conntrack *nfct) | |
179 | { | |
180 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 181 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 182 | |
0d53778e | 183 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
184 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
185 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
186 | ||
187 | nf_conntrack_event(IPCT_DESTROY, ct); | |
188 | set_bit(IPS_DYING_BIT, &ct->status); | |
189 | ||
190 | /* To make sure we don't get any weird locking issues here: | |
191 | * destroy_conntrack() MUST NOT be called with a write lock | |
192 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 193 | rcu_read_lock(); |
923f4902 PM |
194 | l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, |
195 | ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); | |
605dcad6 MJ |
196 | if (l4proto && l4proto->destroy) |
197 | l4proto->destroy(ct); | |
9fb9cbb1 | 198 | |
ecfab2c9 YK |
199 | nf_ct_ext_destroy(ct); |
200 | ||
982d9a9c | 201 | rcu_read_unlock(); |
9fb9cbb1 YK |
202 | |
203 | write_lock_bh(&nf_conntrack_lock); | |
204 | /* Expectations will have been removed in clean_from_lists, | |
205 | * except TFTP can create an expectation on the first packet, | |
206 | * before connection is in the list, so we need to clean here, | |
207 | * too. */ | |
c1d10adb | 208 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
209 | |
210 | /* We overload first tuple to link into unconfirmed list. */ | |
211 | if (!nf_ct_is_confirmed(ct)) { | |
f205c5e0 PM |
212 | BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode)); |
213 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); | |
9fb9cbb1 YK |
214 | } |
215 | ||
216 | NF_CT_STAT_INC(delete); | |
217 | write_unlock_bh(&nf_conntrack_lock); | |
218 | ||
219 | if (ct->master) | |
220 | nf_ct_put(ct->master); | |
221 | ||
0d53778e | 222 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
223 | nf_conntrack_free(ct); |
224 | } | |
225 | ||
226 | static void death_by_timeout(unsigned long ul_conntrack) | |
227 | { | |
228 | struct nf_conn *ct = (void *)ul_conntrack; | |
5397e97d | 229 | struct nf_conn_help *help = nfct_help(ct); |
3c158f7f | 230 | struct nf_conntrack_helper *helper; |
5397e97d | 231 | |
3c158f7f PM |
232 | if (help) { |
233 | rcu_read_lock(); | |
234 | helper = rcu_dereference(help->helper); | |
235 | if (helper && helper->destroy) | |
236 | helper->destroy(ct); | |
237 | rcu_read_unlock(); | |
238 | } | |
9fb9cbb1 YK |
239 | |
240 | write_lock_bh(&nf_conntrack_lock); | |
241 | /* Inside lock so preempt is disabled on module removal path. | |
242 | * Otherwise we can get spurious warnings. */ | |
243 | NF_CT_STAT_INC(delete_list); | |
244 | clean_from_lists(ct); | |
245 | write_unlock_bh(&nf_conntrack_lock); | |
246 | nf_ct_put(ct); | |
247 | } | |
248 | ||
c1d10adb | 249 | struct nf_conntrack_tuple_hash * |
9fb9cbb1 YK |
250 | __nf_conntrack_find(const struct nf_conntrack_tuple *tuple, |
251 | const struct nf_conn *ignored_conntrack) | |
252 | { | |
253 | struct nf_conntrack_tuple_hash *h; | |
f205c5e0 | 254 | struct hlist_node *n; |
9fb9cbb1 YK |
255 | unsigned int hash = hash_conntrack(tuple); |
256 | ||
f205c5e0 | 257 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) { |
df0933dc PM |
258 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
259 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
9fb9cbb1 YK |
260 | NF_CT_STAT_INC(found); |
261 | return h; | |
262 | } | |
263 | NF_CT_STAT_INC(searched); | |
264 | } | |
265 | ||
266 | return NULL; | |
267 | } | |
13b18339 | 268 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
269 | |
270 | /* Find a connection corresponding to a tuple. */ | |
271 | struct nf_conntrack_tuple_hash * | |
330f7db5 | 272 | nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
273 | { |
274 | struct nf_conntrack_tuple_hash *h; | |
275 | ||
276 | read_lock_bh(&nf_conntrack_lock); | |
330f7db5 | 277 | h = __nf_conntrack_find(tuple, NULL); |
9fb9cbb1 YK |
278 | if (h) |
279 | atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use); | |
280 | read_unlock_bh(&nf_conntrack_lock); | |
281 | ||
282 | return h; | |
283 | } | |
13b18339 | 284 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 285 | |
c1d10adb PNA |
286 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
287 | unsigned int hash, | |
601e68e1 | 288 | unsigned int repl_hash) |
c1d10adb PNA |
289 | { |
290 | ct->id = ++nf_conntrack_next_id; | |
f205c5e0 PM |
291 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
292 | &nf_conntrack_hash[hash]); | |
293 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, | |
294 | &nf_conntrack_hash[repl_hash]); | |
c1d10adb PNA |
295 | } |
296 | ||
297 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
298 | { | |
299 | unsigned int hash, repl_hash; | |
300 | ||
301 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
302 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
303 | ||
304 | write_lock_bh(&nf_conntrack_lock); | |
305 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
306 | write_unlock_bh(&nf_conntrack_lock); | |
307 | } | |
13b18339 | 308 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 309 | |
9fb9cbb1 YK |
310 | /* Confirm a connection given skb; places it in hash table */ |
311 | int | |
312 | __nf_conntrack_confirm(struct sk_buff **pskb) | |
313 | { | |
314 | unsigned int hash, repl_hash; | |
df0933dc | 315 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 316 | struct nf_conn *ct; |
df0933dc | 317 | struct nf_conn_help *help; |
f205c5e0 | 318 | struct hlist_node *n; |
9fb9cbb1 YK |
319 | enum ip_conntrack_info ctinfo; |
320 | ||
321 | ct = nf_ct_get(*pskb, &ctinfo); | |
322 | ||
323 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
324 | ICMP/TCP RST packets in other direction. Actual packet | |
325 | which created connection will be IP_CT_NEW or for an | |
326 | expected connection, IP_CT_RELATED. */ | |
327 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
328 | return NF_ACCEPT; | |
329 | ||
330 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
331 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
332 | ||
333 | /* We're not in hash table, and we refuse to set up related | |
334 | connections for unconfirmed conns. But packet copies and | |
335 | REJECT will give spurious warnings here. */ | |
336 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
337 | ||
338 | /* No external references means noone else could have | |
339 | confirmed us. */ | |
340 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 341 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 YK |
342 | |
343 | write_lock_bh(&nf_conntrack_lock); | |
344 | ||
345 | /* See if there's one in the list already, including reverse: | |
346 | NAT could have grabbed it without realizing, since we're | |
347 | not in the hash. If there is, we lost race. */ | |
f205c5e0 | 348 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) |
df0933dc PM |
349 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
350 | &h->tuple)) | |
351 | goto out; | |
f205c5e0 | 352 | hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode) |
df0933dc PM |
353 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
354 | &h->tuple)) | |
355 | goto out; | |
9fb9cbb1 | 356 | |
df0933dc | 357 | /* Remove from unconfirmed list */ |
f205c5e0 | 358 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
df0933dc PM |
359 | |
360 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
361 | /* Timer relative to confirmation time, not original | |
362 | setting time, otherwise we'd get timer wrap in | |
363 | weird delay cases. */ | |
364 | ct->timeout.expires += jiffies; | |
365 | add_timer(&ct->timeout); | |
366 | atomic_inc(&ct->ct_general.use); | |
367 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
368 | NF_CT_STAT_INC(insert); | |
369 | write_unlock_bh(&nf_conntrack_lock); | |
370 | help = nfct_help(ct); | |
371 | if (help && help->helper) | |
372 | nf_conntrack_event_cache(IPCT_HELPER, *pskb); | |
9fb9cbb1 | 373 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
374 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
375 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
376 | nf_conntrack_event_cache(IPCT_NATINFO, *pskb); | |
9fb9cbb1 | 377 | #endif |
df0933dc PM |
378 | nf_conntrack_event_cache(master_ct(ct) ? |
379 | IPCT_RELATED : IPCT_NEW, *pskb); | |
380 | return NF_ACCEPT; | |
9fb9cbb1 | 381 | |
df0933dc | 382 | out: |
9fb9cbb1 YK |
383 | NF_CT_STAT_INC(insert_failed); |
384 | write_unlock_bh(&nf_conntrack_lock); | |
385 | return NF_DROP; | |
386 | } | |
13b18339 | 387 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
388 | |
389 | /* Returns true if a connection correspondings to the tuple (required | |
390 | for NAT). */ | |
391 | int | |
392 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
393 | const struct nf_conn *ignored_conntrack) | |
394 | { | |
395 | struct nf_conntrack_tuple_hash *h; | |
396 | ||
397 | read_lock_bh(&nf_conntrack_lock); | |
398 | h = __nf_conntrack_find(tuple, ignored_conntrack); | |
399 | read_unlock_bh(&nf_conntrack_lock); | |
400 | ||
401 | return h != NULL; | |
402 | } | |
13b18339 | 403 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 404 | |
7ae7730f PM |
405 | #define NF_CT_EVICTION_RANGE 8 |
406 | ||
9fb9cbb1 YK |
407 | /* There's a small race here where we may free a just-assured |
408 | connection. Too bad: we're in trouble anyway. */ | |
7ae7730f | 409 | static int early_drop(unsigned int hash) |
9fb9cbb1 | 410 | { |
f205c5e0 | 411 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 412 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 413 | struct nf_conn *ct = NULL, *tmp; |
f205c5e0 | 414 | struct hlist_node *n; |
7ae7730f | 415 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
416 | int dropped = 0; |
417 | ||
418 | read_lock_bh(&nf_conntrack_lock); | |
7ae7730f PM |
419 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
420 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) { | |
421 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
422 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
423 | ct = tmp; | |
424 | cnt++; | |
425 | } | |
426 | if (ct || cnt >= NF_CT_EVICTION_RANGE) | |
427 | break; | |
428 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 429 | } |
f205c5e0 PM |
430 | if (ct) |
431 | atomic_inc(&ct->ct_general.use); | |
9fb9cbb1 YK |
432 | read_unlock_bh(&nf_conntrack_lock); |
433 | ||
434 | if (!ct) | |
435 | return dropped; | |
436 | ||
437 | if (del_timer(&ct->timeout)) { | |
438 | death_by_timeout((unsigned long)ct); | |
439 | dropped = 1; | |
c0e912d7 | 440 | NF_CT_STAT_INC_ATOMIC(early_drop); |
9fb9cbb1 YK |
441 | } |
442 | nf_ct_put(ct); | |
443 | return dropped; | |
444 | } | |
445 | ||
dacd2a1a YK |
446 | struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, |
447 | const struct nf_conntrack_tuple *repl) | |
9fb9cbb1 YK |
448 | { |
449 | struct nf_conn *conntrack = NULL; | |
9fb9cbb1 | 450 | |
dc808fe2 | 451 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
452 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
453 | nf_conntrack_hash_rnd_initted = 1; | |
454 | } | |
455 | ||
5251e2d2 PNA |
456 | /* We don't want any race condition at early drop stage */ |
457 | atomic_inc(&nf_conntrack_count); | |
458 | ||
9fb9cbb1 | 459 | if (nf_conntrack_max |
5251e2d2 | 460 | && atomic_read(&nf_conntrack_count) > nf_conntrack_max) { |
9fb9cbb1 | 461 | unsigned int hash = hash_conntrack(orig); |
7ae7730f | 462 | if (!early_drop(hash)) { |
5251e2d2 | 463 | atomic_dec(&nf_conntrack_count); |
9fb9cbb1 YK |
464 | if (net_ratelimit()) |
465 | printk(KERN_WARNING | |
466 | "nf_conntrack: table full, dropping" | |
467 | " packet.\n"); | |
468 | return ERR_PTR(-ENOMEM); | |
469 | } | |
470 | } | |
471 | ||
dacd2a1a | 472 | conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC); |
9fb9cbb1 | 473 | if (conntrack == NULL) { |
0d53778e | 474 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
dacd2a1a YK |
475 | atomic_dec(&nf_conntrack_count); |
476 | return ERR_PTR(-ENOMEM); | |
9fb9cbb1 YK |
477 | } |
478 | ||
9fb9cbb1 | 479 | atomic_set(&conntrack->ct_general.use, 1); |
9fb9cbb1 YK |
480 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
481 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
482 | /* Don't set timer yet: wait for confirmation */ | |
e6f689db PM |
483 | setup_timer(&conntrack->timeout, death_by_timeout, |
484 | (unsigned long)conntrack); | |
9fb9cbb1 | 485 | |
9fb9cbb1 YK |
486 | return conntrack; |
487 | } | |
13b18339 | 488 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 YK |
489 | |
490 | void nf_conntrack_free(struct nf_conn *conntrack) | |
491 | { | |
ecfab2c9 | 492 | nf_ct_ext_free(conntrack); |
dacd2a1a | 493 | kmem_cache_free(nf_conntrack_cachep, conntrack); |
9fb9cbb1 YK |
494 | atomic_dec(&nf_conntrack_count); |
495 | } | |
13b18339 | 496 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
497 | |
498 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
499 | failed due to stress. Otherwise it really is unclassifiable. */ | |
500 | static struct nf_conntrack_tuple_hash * | |
501 | init_conntrack(const struct nf_conntrack_tuple *tuple, | |
502 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 503 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
504 | struct sk_buff *skb, |
505 | unsigned int dataoff) | |
506 | { | |
507 | struct nf_conn *conntrack; | |
3c158f7f | 508 | struct nf_conn_help *help; |
9fb9cbb1 YK |
509 | struct nf_conntrack_tuple repl_tuple; |
510 | struct nf_conntrack_expect *exp; | |
511 | ||
605dcad6 | 512 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 513 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
514 | return NULL; |
515 | } | |
516 | ||
dacd2a1a | 517 | conntrack = nf_conntrack_alloc(tuple, &repl_tuple); |
9fb9cbb1 | 518 | if (conntrack == NULL || IS_ERR(conntrack)) { |
0d53778e | 519 | pr_debug("Can't allocate conntrack.\n"); |
9fb9cbb1 YK |
520 | return (struct nf_conntrack_tuple_hash *)conntrack; |
521 | } | |
522 | ||
605dcad6 | 523 | if (!l4proto->new(conntrack, skb, dataoff)) { |
9fb9cbb1 | 524 | nf_conntrack_free(conntrack); |
0d53778e | 525 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
526 | return NULL; |
527 | } | |
528 | ||
529 | write_lock_bh(&nf_conntrack_lock); | |
6823645d | 530 | exp = nf_ct_find_expectation(tuple); |
9fb9cbb1 | 531 | if (exp) { |
0d53778e PM |
532 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
533 | conntrack, exp); | |
9fb9cbb1 YK |
534 | /* Welcome, Mr. Bond. We've been expecting you... */ |
535 | __set_bit(IPS_EXPECTED_BIT, &conntrack->status); | |
536 | conntrack->master = exp->master; | |
ceceae1b | 537 | if (exp->helper) { |
b560580a | 538 | help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC); |
ceceae1b YK |
539 | if (help) |
540 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
541 | } |
542 | ||
9fb9cbb1 YK |
543 | #ifdef CONFIG_NF_CONNTRACK_MARK |
544 | conntrack->mark = exp->master->mark; | |
7c9728c3 JM |
545 | #endif |
546 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
547 | conntrack->secmark = exp->master->secmark; | |
9fb9cbb1 YK |
548 | #endif |
549 | nf_conntrack_get(&conntrack->master->ct_general); | |
550 | NF_CT_STAT_INC(expect_new); | |
22e7410b | 551 | } else { |
ceceae1b YK |
552 | struct nf_conntrack_helper *helper; |
553 | ||
554 | helper = __nf_ct_helper_find(&repl_tuple); | |
555 | if (helper) { | |
b560580a | 556 | help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC); |
ceceae1b | 557 | if (help) |
ceceae1b | 558 | rcu_assign_pointer(help->helper, helper); |
3c158f7f | 559 | } |
9fb9cbb1 | 560 | NF_CT_STAT_INC(new); |
22e7410b | 561 | } |
9fb9cbb1 YK |
562 | |
563 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
f205c5e0 PM |
564 | hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
565 | &unconfirmed); | |
9fb9cbb1 YK |
566 | |
567 | write_unlock_bh(&nf_conntrack_lock); | |
568 | ||
569 | if (exp) { | |
570 | if (exp->expectfn) | |
571 | exp->expectfn(conntrack, exp); | |
6823645d | 572 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
573 | } |
574 | ||
575 | return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; | |
576 | } | |
577 | ||
578 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
579 | static inline struct nf_conn * | |
580 | resolve_normal_ct(struct sk_buff *skb, | |
581 | unsigned int dataoff, | |
582 | u_int16_t l3num, | |
583 | u_int8_t protonum, | |
584 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 585 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
586 | int *set_reply, |
587 | enum ip_conntrack_info *ctinfo) | |
588 | { | |
589 | struct nf_conntrack_tuple tuple; | |
590 | struct nf_conntrack_tuple_hash *h; | |
591 | struct nf_conn *ct; | |
592 | ||
bbe735e4 | 593 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 594 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 595 | l4proto)) { |
0d53778e | 596 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
597 | return NULL; |
598 | } | |
599 | ||
600 | /* look for tuple match */ | |
330f7db5 | 601 | h = nf_conntrack_find_get(&tuple); |
9fb9cbb1 | 602 | if (!h) { |
605dcad6 | 603 | h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
604 | if (!h) |
605 | return NULL; | |
606 | if (IS_ERR(h)) | |
607 | return (void *)h; | |
608 | } | |
609 | ct = nf_ct_tuplehash_to_ctrack(h); | |
610 | ||
611 | /* It exists; we have (non-exclusive) reference. */ | |
612 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
613 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
614 | /* Please set reply bit if this packet OK */ | |
615 | *set_reply = 1; | |
616 | } else { | |
617 | /* Once we've had two way comms, always ESTABLISHED. */ | |
618 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 619 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
620 | *ctinfo = IP_CT_ESTABLISHED; |
621 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
622 | pr_debug("nf_conntrack_in: related packet for %p\n", |
623 | ct); | |
9fb9cbb1 YK |
624 | *ctinfo = IP_CT_RELATED; |
625 | } else { | |
0d53778e | 626 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
627 | *ctinfo = IP_CT_NEW; |
628 | } | |
629 | *set_reply = 0; | |
630 | } | |
631 | skb->nfct = &ct->ct_general; | |
632 | skb->nfctinfo = *ctinfo; | |
633 | return ct; | |
634 | } | |
635 | ||
636 | unsigned int | |
637 | nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb) | |
638 | { | |
639 | struct nf_conn *ct; | |
640 | enum ip_conntrack_info ctinfo; | |
641 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 642 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
643 | unsigned int dataoff; |
644 | u_int8_t protonum; | |
645 | int set_reply = 0; | |
646 | int ret; | |
647 | ||
648 | /* Previously seen (loopback or untracked)? Ignore. */ | |
649 | if ((*pskb)->nfct) { | |
c0e912d7 | 650 | NF_CT_STAT_INC_ATOMIC(ignore); |
9fb9cbb1 YK |
651 | return NF_ACCEPT; |
652 | } | |
653 | ||
923f4902 | 654 | /* rcu_read_lock()ed by nf_hook_slow */ |
c1d10adb | 655 | l3proto = __nf_ct_l3proto_find((u_int16_t)pf); |
ffc30690 YK |
656 | ret = l3proto->get_l4proto(*pskb, skb_network_offset(*pskb), |
657 | &dataoff, &protonum); | |
658 | if (ret <= 0) { | |
0d53778e | 659 | pr_debug("not prepared to track yet or error occured\n"); |
d87d8469 YK |
660 | NF_CT_STAT_INC_ATOMIC(error); |
661 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
662 | return -ret; |
663 | } | |
664 | ||
605dcad6 | 665 | l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); |
9fb9cbb1 YK |
666 | |
667 | /* It may be an special packet, error, unclean... | |
668 | * inverse of the return code tells to the netfilter | |
669 | * core what to do with the packet. */ | |
605dcad6 MJ |
670 | if (l4proto->error != NULL && |
671 | (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) { | |
c0e912d7 PM |
672 | NF_CT_STAT_INC_ATOMIC(error); |
673 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
674 | return -ret; |
675 | } | |
676 | ||
605dcad6 | 677 | ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto, |
9fb9cbb1 YK |
678 | &set_reply, &ctinfo); |
679 | if (!ct) { | |
680 | /* Not valid part of a connection */ | |
c0e912d7 | 681 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
682 | return NF_ACCEPT; |
683 | } | |
684 | ||
685 | if (IS_ERR(ct)) { | |
686 | /* Too stressed to deal. */ | |
c0e912d7 | 687 | NF_CT_STAT_INC_ATOMIC(drop); |
9fb9cbb1 YK |
688 | return NF_DROP; |
689 | } | |
690 | ||
691 | NF_CT_ASSERT((*pskb)->nfct); | |
692 | ||
605dcad6 | 693 | ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
694 | if (ret < 0) { |
695 | /* Invalid: inverse of the return code tells | |
696 | * the netfilter core what to do */ | |
0d53778e | 697 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
9fb9cbb1 YK |
698 | nf_conntrack_put((*pskb)->nfct); |
699 | (*pskb)->nfct = NULL; | |
c0e912d7 | 700 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
701 | return -ret; |
702 | } | |
703 | ||
704 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
705 | nf_conntrack_event_cache(IPCT_STATUS, *pskb); | |
706 | ||
707 | return ret; | |
708 | } | |
13b18339 | 709 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 YK |
710 | |
711 | int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, | |
712 | const struct nf_conntrack_tuple *orig) | |
713 | { | |
923f4902 PM |
714 | int ret; |
715 | ||
716 | rcu_read_lock(); | |
717 | ret = nf_ct_invert_tuple(inverse, orig, | |
718 | __nf_ct_l3proto_find(orig->src.l3num), | |
719 | __nf_ct_l4proto_find(orig->src.l3num, | |
720 | orig->dst.protonum)); | |
721 | rcu_read_unlock(); | |
722 | return ret; | |
9fb9cbb1 | 723 | } |
13b18339 | 724 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 725 | |
5b1158e9 JK |
726 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
727 | implicitly racy: see __nf_conntrack_confirm */ | |
728 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
729 | const struct nf_conntrack_tuple *newreply) | |
730 | { | |
731 | struct nf_conn_help *help = nfct_help(ct); | |
ceceae1b | 732 | struct nf_conntrack_helper *helper; |
5b1158e9 JK |
733 | |
734 | write_lock_bh(&nf_conntrack_lock); | |
735 | /* Should be unconfirmed, so not in hash table yet */ | |
736 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
737 | ||
0d53778e | 738 | pr_debug("Altering reply tuple of %p to ", ct); |
5b1158e9 JK |
739 | NF_CT_DUMP_TUPLE(newreply); |
740 | ||
741 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ceceae1b YK |
742 | if (ct->master || (help && help->expecting != 0)) |
743 | goto out; | |
744 | ||
745 | helper = __nf_ct_helper_find(newreply); | |
746 | if (helper == NULL) { | |
747 | if (help) | |
748 | rcu_assign_pointer(help->helper, NULL); | |
749 | goto out; | |
5d78a849 | 750 | } |
ceceae1b YK |
751 | |
752 | if (help == NULL) { | |
b560580a PM |
753 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
754 | if (help == NULL) | |
ceceae1b | 755 | goto out; |
ceceae1b YK |
756 | } else { |
757 | memset(&help->help, 0, sizeof(help->help)); | |
758 | } | |
759 | ||
760 | rcu_assign_pointer(help->helper, helper); | |
761 | out: | |
5b1158e9 JK |
762 | write_unlock_bh(&nf_conntrack_lock); |
763 | } | |
13b18339 | 764 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 765 | |
9fb9cbb1 YK |
766 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
767 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
768 | enum ip_conntrack_info ctinfo, | |
769 | const struct sk_buff *skb, | |
770 | unsigned long extra_jiffies, | |
771 | int do_acct) | |
772 | { | |
773 | int event = 0; | |
774 | ||
775 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
776 | NF_CT_ASSERT(skb); | |
777 | ||
778 | write_lock_bh(&nf_conntrack_lock); | |
779 | ||
997ae831 EL |
780 | /* Only update if this is not a fixed timeout */ |
781 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) { | |
782 | write_unlock_bh(&nf_conntrack_lock); | |
783 | return; | |
784 | } | |
785 | ||
9fb9cbb1 YK |
786 | /* If not in hash table, timer will not be active yet */ |
787 | if (!nf_ct_is_confirmed(ct)) { | |
788 | ct->timeout.expires = extra_jiffies; | |
789 | event = IPCT_REFRESH; | |
790 | } else { | |
be00c8e4 MJ |
791 | unsigned long newtime = jiffies + extra_jiffies; |
792 | ||
793 | /* Only update the timeout if the new timeout is at least | |
794 | HZ jiffies from the old timeout. Need del_timer for race | |
795 | avoidance (may already be dying). */ | |
796 | if (newtime - ct->timeout.expires >= HZ | |
797 | && del_timer(&ct->timeout)) { | |
798 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
799 | add_timer(&ct->timeout); |
800 | event = IPCT_REFRESH; | |
801 | } | |
802 | } | |
803 | ||
804 | #ifdef CONFIG_NF_CT_ACCT | |
805 | if (do_acct) { | |
806 | ct->counters[CTINFO2DIR(ctinfo)].packets++; | |
807 | ct->counters[CTINFO2DIR(ctinfo)].bytes += | |
bbe735e4 | 808 | skb->len - skb_network_offset(skb); |
3ffd5eeb MJ |
809 | |
810 | if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) | |
811 | || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) | |
812 | event |= IPCT_COUNTER_FILLING; | |
9fb9cbb1 YK |
813 | } |
814 | #endif | |
815 | ||
816 | write_unlock_bh(&nf_conntrack_lock); | |
817 | ||
818 | /* must be unlocked when calling event cache */ | |
819 | if (event) | |
820 | nf_conntrack_event_cache(event, skb); | |
821 | } | |
13b18339 | 822 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 823 | |
e281db5c | 824 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
825 | |
826 | #include <linux/netfilter/nfnetlink.h> | |
827 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
828 | #include <linux/mutex.h> |
829 | ||
c1d10adb PNA |
830 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
831 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
832 | * or depend on ctnetlink */ | |
fdf70832 | 833 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
834 | const struct nf_conntrack_tuple *tuple) |
835 | { | |
df6fb868 | 836 | NLA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t), |
c1d10adb | 837 | &tuple->src.u.tcp.port); |
df6fb868 | 838 | NLA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t), |
c1d10adb PNA |
839 | &tuple->dst.u.tcp.port); |
840 | return 0; | |
841 | ||
df6fb868 | 842 | nla_put_failure: |
c1d10adb PNA |
843 | return -1; |
844 | } | |
fdf70832 | 845 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 846 | |
df6fb868 PM |
847 | static const size_t cta_min_proto[CTA_PROTO_MAX+1] = { |
848 | [CTA_PROTO_SRC_PORT] = sizeof(u_int16_t), | |
849 | [CTA_PROTO_DST_PORT] = sizeof(u_int16_t) | |
c1d10adb PNA |
850 | }; |
851 | ||
fdf70832 | 852 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
853 | struct nf_conntrack_tuple *t) |
854 | { | |
df6fb868 | 855 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
856 | return -EINVAL; |
857 | ||
fdf70832 | 858 | if (nlattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto)) |
c1d10adb PNA |
859 | return -EINVAL; |
860 | ||
df6fb868 PM |
861 | t->src.u.tcp.port = *(__be16 *)nla_data(tb[CTA_PROTO_SRC_PORT]); |
862 | t->dst.u.tcp.port = *(__be16 *)nla_data(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
863 | |
864 | return 0; | |
865 | } | |
fdf70832 | 866 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
c1d10adb PNA |
867 | #endif |
868 | ||
9fb9cbb1 YK |
869 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
870 | void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) | |
871 | { | |
872 | struct nf_conn *ct; | |
873 | enum ip_conntrack_info ctinfo; | |
874 | ||
875 | /* This ICMP is in reverse direction to the packet which caused it */ | |
876 | ct = nf_ct_get(skb, &ctinfo); | |
877 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
878 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
879 | else | |
880 | ctinfo = IP_CT_RELATED; | |
881 | ||
882 | /* Attach to new skbuff, and increment count */ | |
883 | nskb->nfct = &ct->ct_general; | |
884 | nskb->nfctinfo = ctinfo; | |
885 | nf_conntrack_get(nskb->nfct); | |
886 | } | |
13b18339 | 887 | EXPORT_SYMBOL_GPL(__nf_conntrack_attach); |
9fb9cbb1 YK |
888 | |
889 | static inline int | |
890 | do_iter(const struct nf_conntrack_tuple_hash *i, | |
891 | int (*iter)(struct nf_conn *i, void *data), | |
892 | void *data) | |
893 | { | |
894 | return iter(nf_ct_tuplehash_to_ctrack(i), data); | |
895 | } | |
896 | ||
897 | /* Bring out ya dead! */ | |
df0933dc | 898 | static struct nf_conn * |
9fb9cbb1 YK |
899 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
900 | void *data, unsigned int *bucket) | |
901 | { | |
df0933dc PM |
902 | struct nf_conntrack_tuple_hash *h; |
903 | struct nf_conn *ct; | |
f205c5e0 | 904 | struct hlist_node *n; |
9fb9cbb1 YK |
905 | |
906 | write_lock_bh(&nf_conntrack_lock); | |
907 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { | |
f205c5e0 | 908 | hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) { |
df0933dc PM |
909 | ct = nf_ct_tuplehash_to_ctrack(h); |
910 | if (iter(ct, data)) | |
911 | goto found; | |
912 | } | |
601e68e1 | 913 | } |
f205c5e0 | 914 | hlist_for_each_entry(h, n, &unconfirmed, hnode) { |
df0933dc PM |
915 | ct = nf_ct_tuplehash_to_ctrack(h); |
916 | if (iter(ct, data)) | |
ec68e97d | 917 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 918 | } |
c073e3fa | 919 | write_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
920 | return NULL; |
921 | found: | |
c073e3fa | 922 | atomic_inc(&ct->ct_general.use); |
9fb9cbb1 | 923 | write_unlock_bh(&nf_conntrack_lock); |
df0933dc | 924 | return ct; |
9fb9cbb1 YK |
925 | } |
926 | ||
927 | void | |
928 | nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) | |
929 | { | |
df0933dc | 930 | struct nf_conn *ct; |
9fb9cbb1 YK |
931 | unsigned int bucket = 0; |
932 | ||
df0933dc | 933 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
934 | /* Time to push up daises... */ |
935 | if (del_timer(&ct->timeout)) | |
936 | death_by_timeout((unsigned long)ct); | |
937 | /* ... else the timer will get him soon. */ | |
938 | ||
939 | nf_ct_put(ct); | |
940 | } | |
941 | } | |
13b18339 | 942 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 YK |
943 | |
944 | static int kill_all(struct nf_conn *i, void *data) | |
945 | { | |
946 | return 1; | |
947 | } | |
948 | ||
ac565e5f | 949 | void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, int size) |
9fb9cbb1 YK |
950 | { |
951 | if (vmalloced) | |
952 | vfree(hash); | |
953 | else | |
601e68e1 | 954 | free_pages((unsigned long)hash, |
f205c5e0 | 955 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 956 | } |
ac565e5f | 957 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 958 | |
272491ef | 959 | void nf_conntrack_flush(void) |
c1d10adb PNA |
960 | { |
961 | nf_ct_iterate_cleanup(kill_all, NULL); | |
962 | } | |
13b18339 | 963 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 964 | |
9fb9cbb1 YK |
965 | /* Mishearing the voices in his head, our hero wonders how he's |
966 | supposed to kill the mall. */ | |
967 | void nf_conntrack_cleanup(void) | |
968 | { | |
c3a47ab3 | 969 | rcu_assign_pointer(ip_ct_attach, NULL); |
7d3cdc6b | 970 | |
9fb9cbb1 YK |
971 | /* This makes sure all current packets have passed through |
972 | netfilter framework. Roll on, two-stage module | |
973 | delete... */ | |
974 | synchronize_net(); | |
975 | ||
976 | nf_ct_event_cache_flush(); | |
977 | i_see_dead_people: | |
c1d10adb | 978 | nf_conntrack_flush(); |
9fb9cbb1 YK |
979 | if (atomic_read(&nf_conntrack_count) != 0) { |
980 | schedule(); | |
981 | goto i_see_dead_people; | |
982 | } | |
6636568c PM |
983 | /* wait until all references to nf_conntrack_untracked are dropped */ |
984 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
985 | schedule(); | |
9fb9cbb1 | 986 | |
de6e05c4 YK |
987 | rcu_assign_pointer(nf_ct_destroy, NULL); |
988 | ||
dacd2a1a | 989 | kmem_cache_destroy(nf_conntrack_cachep); |
ac565e5f PM |
990 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
991 | nf_conntrack_htable_size); | |
5a6f294e | 992 | |
ac5357eb | 993 | nf_conntrack_proto_fini(); |
ceceae1b | 994 | nf_conntrack_helper_fini(); |
e9c1b084 | 995 | nf_conntrack_expect_fini(); |
9fb9cbb1 YK |
996 | } |
997 | ||
ac565e5f | 998 | struct hlist_head *nf_ct_alloc_hashtable(int *sizep, int *vmalloced) |
9fb9cbb1 | 999 | { |
f205c5e0 | 1000 | struct hlist_head *hash; |
8e5105a0 | 1001 | unsigned int size, i; |
9fb9cbb1 | 1002 | |
601e68e1 | 1003 | *vmalloced = 0; |
8e5105a0 | 1004 | |
f205c5e0 | 1005 | size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); |
601e68e1 | 1006 | hash = (void*)__get_free_pages(GFP_KERNEL, |
f205c5e0 | 1007 | get_order(sizeof(struct hlist_head) |
9fb9cbb1 | 1008 | * size)); |
601e68e1 | 1009 | if (!hash) { |
9fb9cbb1 YK |
1010 | *vmalloced = 1; |
1011 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
f205c5e0 | 1012 | hash = vmalloc(sizeof(struct hlist_head) * size); |
9fb9cbb1 YK |
1013 | } |
1014 | ||
1015 | if (hash) | |
601e68e1 | 1016 | for (i = 0; i < size; i++) |
f205c5e0 | 1017 | INIT_HLIST_HEAD(&hash[i]); |
9fb9cbb1 YK |
1018 | |
1019 | return hash; | |
1020 | } | |
ac565e5f | 1021 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 YK |
1022 | |
1023 | int set_hashsize(const char *val, struct kernel_param *kp) | |
1024 | { | |
1025 | int i, bucket, hashsize, vmalloced; | |
1026 | int old_vmalloced, old_size; | |
1027 | int rnd; | |
f205c5e0 | 1028 | struct hlist_head *hash, *old_hash; |
9fb9cbb1 YK |
1029 | struct nf_conntrack_tuple_hash *h; |
1030 | ||
1031 | /* On boot, we can set this without any fancy locking. */ | |
1032 | if (!nf_conntrack_htable_size) | |
1033 | return param_set_uint(val, kp); | |
1034 | ||
1035 | hashsize = simple_strtol(val, NULL, 0); | |
1036 | if (!hashsize) | |
1037 | return -EINVAL; | |
1038 | ||
ac565e5f | 1039 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); |
9fb9cbb1 YK |
1040 | if (!hash) |
1041 | return -ENOMEM; | |
1042 | ||
1043 | /* We have to rehahs for the new table anyway, so we also can | |
1044 | * use a newrandom seed */ | |
1045 | get_random_bytes(&rnd, 4); | |
1046 | ||
1047 | write_lock_bh(&nf_conntrack_lock); | |
1048 | for (i = 0; i < nf_conntrack_htable_size; i++) { | |
f205c5e0 PM |
1049 | while (!hlist_empty(&nf_conntrack_hash[i])) { |
1050 | h = hlist_entry(nf_conntrack_hash[i].first, | |
1051 | struct nf_conntrack_tuple_hash, hnode); | |
1052 | hlist_del(&h->hnode); | |
9fb9cbb1 | 1053 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
f205c5e0 | 1054 | hlist_add_head(&h->hnode, &hash[bucket]); |
9fb9cbb1 YK |
1055 | } |
1056 | } | |
1057 | old_size = nf_conntrack_htable_size; | |
1058 | old_vmalloced = nf_conntrack_vmalloc; | |
1059 | old_hash = nf_conntrack_hash; | |
1060 | ||
1061 | nf_conntrack_htable_size = hashsize; | |
1062 | nf_conntrack_vmalloc = vmalloced; | |
1063 | nf_conntrack_hash = hash; | |
1064 | nf_conntrack_hash_rnd = rnd; | |
1065 | write_unlock_bh(&nf_conntrack_lock); | |
1066 | ||
ac565e5f | 1067 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1068 | return 0; |
1069 | } | |
1070 | ||
1071 | module_param_call(hashsize, set_hashsize, param_get_uint, | |
1072 | &nf_conntrack_htable_size, 0600); | |
1073 | ||
1074 | int __init nf_conntrack_init(void) | |
1075 | { | |
f205c5e0 | 1076 | int max_factor = 8; |
9fb9cbb1 YK |
1077 | int ret; |
1078 | ||
1079 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1080 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1081 | if (!nf_conntrack_htable_size) { |
1082 | nf_conntrack_htable_size | |
1083 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
f205c5e0 | 1084 | / sizeof(struct hlist_head)); |
9fb9cbb1 | 1085 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1086 | nf_conntrack_htable_size = 16384; |
1087 | if (nf_conntrack_htable_size < 32) | |
1088 | nf_conntrack_htable_size = 32; | |
1089 | ||
1090 | /* Use a max. factor of four by default to get the same max as | |
1091 | * with the old struct list_heads. When a table size is given | |
1092 | * we use the old value of 8 to avoid reducing the max. | |
1093 | * entries. */ | |
1094 | max_factor = 4; | |
9fb9cbb1 | 1095 | } |
ac565e5f PM |
1096 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
1097 | &nf_conntrack_vmalloc); | |
9fb9cbb1 YK |
1098 | if (!nf_conntrack_hash) { |
1099 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1100 | goto err_out; | |
1101 | } | |
1102 | ||
f205c5e0 | 1103 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1104 | |
1105 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1106 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1107 | nf_conntrack_max); | |
1108 | ||
dacd2a1a YK |
1109 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1110 | sizeof(struct nf_conn), | |
20c2df83 | 1111 | 0, 0, NULL); |
dacd2a1a | 1112 | if (!nf_conntrack_cachep) { |
9fb9cbb1 YK |
1113 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
1114 | goto err_free_hash; | |
1115 | } | |
1116 | ||
e9c1b084 PM |
1117 | ret = nf_conntrack_proto_init(); |
1118 | if (ret < 0) | |
9fb9cbb1 | 1119 | goto err_free_conntrack_slab; |
9fb9cbb1 | 1120 | |
e9c1b084 | 1121 | ret = nf_conntrack_expect_init(); |
933a41e7 | 1122 | if (ret < 0) |
e9c1b084 | 1123 | goto out_fini_proto; |
933a41e7 | 1124 | |
ceceae1b YK |
1125 | ret = nf_conntrack_helper_init(); |
1126 | if (ret < 0) | |
e9c1b084 | 1127 | goto out_fini_expect; |
ceceae1b | 1128 | |
7d3cdc6b | 1129 | /* For use by REJECT target */ |
c3a47ab3 | 1130 | rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach); |
de6e05c4 | 1131 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); |
7d3cdc6b | 1132 | |
9fb9cbb1 YK |
1133 | /* Set up fake conntrack: |
1134 | - to never be deleted, not in any hashes */ | |
1135 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); | |
1136 | /* - and look it like as a confirmed connection */ | |
1137 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1138 | ||
1139 | return ret; | |
1140 | ||
e9c1b084 PM |
1141 | out_fini_expect: |
1142 | nf_conntrack_expect_fini(); | |
ceceae1b YK |
1143 | out_fini_proto: |
1144 | nf_conntrack_proto_fini(); | |
9fb9cbb1 | 1145 | err_free_conntrack_slab: |
dacd2a1a | 1146 | kmem_cache_destroy(nf_conntrack_cachep); |
9fb9cbb1 | 1147 | err_free_hash: |
ac565e5f PM |
1148 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
1149 | nf_conntrack_htable_size); | |
9fb9cbb1 YK |
1150 | err_out: |
1151 | return -ENOMEM; | |
1152 | } |