]>
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> |
58401572 | 40 | #include <net/netfilter/nf_conntrack_acct.h> |
9fb9cbb1 | 41 | |
dc808fe2 | 42 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 43 | |
f8ba1aff | 44 | DEFINE_SPINLOCK(nf_conntrack_lock); |
13b18339 | 45 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 YK |
46 | |
47 | /* nf_conntrack_standalone needs this */ | |
48 | atomic_t nf_conntrack_count = ATOMIC_INIT(0); | |
a999e683 | 49 | EXPORT_SYMBOL_GPL(nf_conntrack_count); |
9fb9cbb1 | 50 | |
e2b7606c | 51 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
52 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
53 | ||
94aec08e | 54 | int nf_conntrack_max __read_mostly; |
a999e683 | 55 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 56 | |
f205c5e0 | 57 | struct hlist_head *nf_conntrack_hash __read_mostly; |
13b18339 PM |
58 | EXPORT_SYMBOL_GPL(nf_conntrack_hash); |
59 | ||
e2b7606c | 60 | struct nf_conn nf_conntrack_untracked __read_mostly; |
13b18339 PM |
61 | EXPORT_SYMBOL_GPL(nf_conntrack_untracked); |
62 | ||
94aec08e | 63 | unsigned int nf_ct_log_invalid __read_mostly; |
f205c5e0 | 64 | HLIST_HEAD(unconfirmed); |
1192e403 | 65 | static int nf_conntrack_vmalloc __read_mostly; |
dacd2a1a | 66 | static struct kmem_cache *nf_conntrack_cachep __read_mostly; |
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 | { | |
0794935e PM |
77 | unsigned int n; |
78 | u_int32_t h; | |
79 | ||
80 | /* The direction must be ignored, so we hash everything up to the | |
81 | * destination ports (which is a multiple of 4) and treat the last | |
82 | * three bytes manually. | |
83 | */ | |
84 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
85 | h = jhash2((u32 *)tuple, n, | |
86 | rnd ^ (((__force __u16)tuple->dst.u.all << 16) | | |
87 | tuple->dst.protonum)); | |
88 | ||
89 | return ((u64)h * size) >> 32; | |
9fb9cbb1 YK |
90 | } |
91 | ||
92 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
93 | { | |
94 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
95 | nf_conntrack_hash_rnd); | |
96 | } | |
97 | ||
5f2b4c90 | 98 | bool |
9fb9cbb1 YK |
99 | nf_ct_get_tuple(const struct sk_buff *skb, |
100 | unsigned int nhoff, | |
101 | unsigned int dataoff, | |
102 | u_int16_t l3num, | |
103 | u_int8_t protonum, | |
104 | struct nf_conntrack_tuple *tuple, | |
105 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 106 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 107 | { |
443a70d5 | 108 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
109 | |
110 | tuple->src.l3num = l3num; | |
111 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 112 | return false; |
9fb9cbb1 YK |
113 | |
114 | tuple->dst.protonum = protonum; | |
115 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
116 | ||
605dcad6 | 117 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 118 | } |
13b18339 | 119 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 120 | |
5f2b4c90 JE |
121 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
122 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
123 | { |
124 | struct nf_conntrack_l3proto *l3proto; | |
125 | struct nf_conntrack_l4proto *l4proto; | |
126 | unsigned int protoff; | |
127 | u_int8_t protonum; | |
128 | int ret; | |
129 | ||
130 | rcu_read_lock(); | |
131 | ||
132 | l3proto = __nf_ct_l3proto_find(l3num); | |
133 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
134 | if (ret != NF_ACCEPT) { | |
135 | rcu_read_unlock(); | |
5f2b4c90 | 136 | return false; |
e2a3123f YK |
137 | } |
138 | ||
139 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
140 | ||
141 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
142 | l3proto, l4proto); | |
143 | ||
144 | rcu_read_unlock(); | |
145 | return ret; | |
146 | } | |
147 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
148 | ||
5f2b4c90 | 149 | bool |
9fb9cbb1 YK |
150 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
151 | const struct nf_conntrack_tuple *orig, | |
152 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 153 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 154 | { |
443a70d5 | 155 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
156 | |
157 | inverse->src.l3num = orig->src.l3num; | |
158 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 159 | return false; |
9fb9cbb1 YK |
160 | |
161 | inverse->dst.dir = !orig->dst.dir; | |
162 | ||
163 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 164 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 165 | } |
13b18339 | 166 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 167 | |
9fb9cbb1 YK |
168 | static void |
169 | clean_from_lists(struct nf_conn *ct) | |
170 | { | |
0d53778e | 171 | pr_debug("clean_from_lists(%p)\n", ct); |
76507f69 PM |
172 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
173 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); | |
9fb9cbb1 YK |
174 | |
175 | /* Destroy all pending expectations */ | |
c1d10adb | 176 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
177 | } |
178 | ||
179 | static void | |
180 | destroy_conntrack(struct nf_conntrack *nfct) | |
181 | { | |
182 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 183 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 184 | |
0d53778e | 185 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
186 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
187 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
188 | ||
189 | nf_conntrack_event(IPCT_DESTROY, ct); | |
190 | set_bit(IPS_DYING_BIT, &ct->status); | |
191 | ||
192 | /* To make sure we don't get any weird locking issues here: | |
193 | * destroy_conntrack() MUST NOT be called with a write lock | |
194 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 195 | rcu_read_lock(); |
5e8fbe2a | 196 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
197 | if (l4proto && l4proto->destroy) |
198 | l4proto->destroy(ct); | |
9fb9cbb1 | 199 | |
982d9a9c | 200 | rcu_read_unlock(); |
9fb9cbb1 | 201 | |
f8ba1aff | 202 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
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); | |
f8ba1aff | 216 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
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 | 238 | |
f8ba1aff | 239 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
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); | |
f8ba1aff | 244 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
245 | nf_ct_put(ct); |
246 | } | |
247 | ||
c1d10adb | 248 | struct nf_conntrack_tuple_hash * |
ba419aff | 249 | __nf_conntrack_find(const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
250 | { |
251 | struct nf_conntrack_tuple_hash *h; | |
f205c5e0 | 252 | struct hlist_node *n; |
9fb9cbb1 YK |
253 | unsigned int hash = hash_conntrack(tuple); |
254 | ||
4e29e9ec PM |
255 | /* Disable BHs the entire time since we normally need to disable them |
256 | * at least once for the stats anyway. | |
257 | */ | |
258 | local_bh_disable(); | |
76507f69 | 259 | hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) { |
ba419aff | 260 | if (nf_ct_tuple_equal(tuple, &h->tuple)) { |
9fb9cbb1 | 261 | NF_CT_STAT_INC(found); |
4e29e9ec | 262 | local_bh_enable(); |
9fb9cbb1 YK |
263 | return h; |
264 | } | |
265 | NF_CT_STAT_INC(searched); | |
266 | } | |
4e29e9ec | 267 | local_bh_enable(); |
9fb9cbb1 YK |
268 | |
269 | return NULL; | |
270 | } | |
13b18339 | 271 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
272 | |
273 | /* Find a connection corresponding to a tuple. */ | |
274 | struct nf_conntrack_tuple_hash * | |
330f7db5 | 275 | nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
276 | { |
277 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 278 | struct nf_conn *ct; |
9fb9cbb1 | 279 | |
76507f69 | 280 | rcu_read_lock(); |
ba419aff | 281 | h = __nf_conntrack_find(tuple); |
76507f69 PM |
282 | if (h) { |
283 | ct = nf_ct_tuplehash_to_ctrack(h); | |
284 | if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
285 | h = NULL; | |
286 | } | |
287 | rcu_read_unlock(); | |
9fb9cbb1 YK |
288 | |
289 | return h; | |
290 | } | |
13b18339 | 291 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 292 | |
c1d10adb PNA |
293 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
294 | unsigned int hash, | |
601e68e1 | 295 | unsigned int repl_hash) |
c1d10adb | 296 | { |
76507f69 PM |
297 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
298 | &nf_conntrack_hash[hash]); | |
299 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, | |
300 | &nf_conntrack_hash[repl_hash]); | |
c1d10adb PNA |
301 | } |
302 | ||
303 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
304 | { | |
305 | unsigned int hash, repl_hash; | |
306 | ||
307 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
308 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
309 | ||
f8ba1aff | 310 | spin_lock_bh(&nf_conntrack_lock); |
c1d10adb | 311 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
f8ba1aff | 312 | spin_unlock_bh(&nf_conntrack_lock); |
c1d10adb | 313 | } |
13b18339 | 314 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 315 | |
9fb9cbb1 YK |
316 | /* Confirm a connection given skb; places it in hash table */ |
317 | int | |
3db05fea | 318 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
319 | { |
320 | unsigned int hash, repl_hash; | |
df0933dc | 321 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 322 | struct nf_conn *ct; |
df0933dc | 323 | struct nf_conn_help *help; |
f205c5e0 | 324 | struct hlist_node *n; |
9fb9cbb1 YK |
325 | enum ip_conntrack_info ctinfo; |
326 | ||
3db05fea | 327 | ct = nf_ct_get(skb, &ctinfo); |
9fb9cbb1 YK |
328 | |
329 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
330 | ICMP/TCP RST packets in other direction. Actual packet | |
331 | which created connection will be IP_CT_NEW or for an | |
332 | expected connection, IP_CT_RELATED. */ | |
333 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
334 | return NF_ACCEPT; | |
335 | ||
336 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
337 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
338 | ||
339 | /* We're not in hash table, and we refuse to set up related | |
340 | connections for unconfirmed conns. But packet copies and | |
341 | REJECT will give spurious warnings here. */ | |
342 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
343 | ||
344 | /* No external references means noone else could have | |
345 | confirmed us. */ | |
346 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 347 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 348 | |
f8ba1aff | 349 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
350 | |
351 | /* See if there's one in the list already, including reverse: | |
352 | NAT could have grabbed it without realizing, since we're | |
353 | not in the hash. If there is, we lost race. */ | |
f205c5e0 | 354 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) |
df0933dc PM |
355 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
356 | &h->tuple)) | |
357 | goto out; | |
f205c5e0 | 358 | hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode) |
df0933dc PM |
359 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
360 | &h->tuple)) | |
361 | goto out; | |
9fb9cbb1 | 362 | |
df0933dc | 363 | /* Remove from unconfirmed list */ |
f205c5e0 | 364 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
df0933dc PM |
365 | |
366 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
367 | /* Timer relative to confirmation time, not original | |
368 | setting time, otherwise we'd get timer wrap in | |
369 | weird delay cases. */ | |
370 | ct->timeout.expires += jiffies; | |
371 | add_timer(&ct->timeout); | |
372 | atomic_inc(&ct->ct_general.use); | |
373 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
374 | NF_CT_STAT_INC(insert); | |
f8ba1aff | 375 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
376 | help = nfct_help(ct); |
377 | if (help && help->helper) | |
3db05fea | 378 | nf_conntrack_event_cache(IPCT_HELPER, skb); |
9fb9cbb1 | 379 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
380 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
381 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
3db05fea | 382 | nf_conntrack_event_cache(IPCT_NATINFO, skb); |
9fb9cbb1 | 383 | #endif |
df0933dc | 384 | nf_conntrack_event_cache(master_ct(ct) ? |
3db05fea | 385 | IPCT_RELATED : IPCT_NEW, skb); |
df0933dc | 386 | return NF_ACCEPT; |
9fb9cbb1 | 387 | |
df0933dc | 388 | out: |
9fb9cbb1 | 389 | NF_CT_STAT_INC(insert_failed); |
f8ba1aff | 390 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
391 | return NF_DROP; |
392 | } | |
13b18339 | 393 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
394 | |
395 | /* Returns true if a connection correspondings to the tuple (required | |
396 | for NAT). */ | |
397 | int | |
398 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
399 | const struct nf_conn *ignored_conntrack) | |
400 | { | |
401 | struct nf_conntrack_tuple_hash *h; | |
ba419aff PM |
402 | struct hlist_node *n; |
403 | unsigned int hash = hash_conntrack(tuple); | |
9fb9cbb1 | 404 | |
4e29e9ec PM |
405 | /* Disable BHs the entire time since we need to disable them at |
406 | * least once for the stats anyway. | |
407 | */ | |
408 | rcu_read_lock_bh(); | |
ba419aff PM |
409 | hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) { |
410 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && | |
411 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
412 | NF_CT_STAT_INC(found); | |
4e29e9ec | 413 | rcu_read_unlock_bh(); |
ba419aff PM |
414 | return 1; |
415 | } | |
416 | NF_CT_STAT_INC(searched); | |
417 | } | |
4e29e9ec | 418 | rcu_read_unlock_bh(); |
9fb9cbb1 | 419 | |
ba419aff | 420 | return 0; |
9fb9cbb1 | 421 | } |
13b18339 | 422 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 423 | |
7ae7730f PM |
424 | #define NF_CT_EVICTION_RANGE 8 |
425 | ||
9fb9cbb1 YK |
426 | /* There's a small race here where we may free a just-assured |
427 | connection. Too bad: we're in trouble anyway. */ | |
76eb9460 | 428 | static noinline int early_drop(unsigned int hash) |
9fb9cbb1 | 429 | { |
f205c5e0 | 430 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 431 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 432 | struct nf_conn *ct = NULL, *tmp; |
f205c5e0 | 433 | struct hlist_node *n; |
7ae7730f | 434 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
435 | int dropped = 0; |
436 | ||
76507f69 | 437 | rcu_read_lock(); |
7ae7730f | 438 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
76507f69 PM |
439 | hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], |
440 | hnode) { | |
7ae7730f PM |
441 | tmp = nf_ct_tuplehash_to_ctrack(h); |
442 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
443 | ct = tmp; | |
444 | cnt++; | |
445 | } | |
76507f69 PM |
446 | |
447 | if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
448 | ct = NULL; | |
7ae7730f PM |
449 | if (ct || cnt >= NF_CT_EVICTION_RANGE) |
450 | break; | |
451 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 452 | } |
76507f69 | 453 | rcu_read_unlock(); |
9fb9cbb1 YK |
454 | |
455 | if (!ct) | |
456 | return dropped; | |
457 | ||
458 | if (del_timer(&ct->timeout)) { | |
459 | death_by_timeout((unsigned long)ct); | |
460 | dropped = 1; | |
c0e912d7 | 461 | NF_CT_STAT_INC_ATOMIC(early_drop); |
9fb9cbb1 YK |
462 | } |
463 | nf_ct_put(ct); | |
464 | return dropped; | |
465 | } | |
466 | ||
dacd2a1a | 467 | struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, |
b891c5a8 PNA |
468 | const struct nf_conntrack_tuple *repl, |
469 | gfp_t gfp) | |
9fb9cbb1 | 470 | { |
c88130bc | 471 | struct nf_conn *ct = NULL; |
9fb9cbb1 | 472 | |
dc808fe2 | 473 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
474 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
475 | nf_conntrack_hash_rnd_initted = 1; | |
476 | } | |
477 | ||
5251e2d2 PNA |
478 | /* We don't want any race condition at early drop stage */ |
479 | atomic_inc(&nf_conntrack_count); | |
480 | ||
76eb9460 PM |
481 | if (nf_conntrack_max && |
482 | unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) { | |
9fb9cbb1 | 483 | unsigned int hash = hash_conntrack(orig); |
7ae7730f | 484 | if (!early_drop(hash)) { |
5251e2d2 | 485 | atomic_dec(&nf_conntrack_count); |
9fb9cbb1 YK |
486 | if (net_ratelimit()) |
487 | printk(KERN_WARNING | |
488 | "nf_conntrack: table full, dropping" | |
489 | " packet.\n"); | |
490 | return ERR_PTR(-ENOMEM); | |
491 | } | |
492 | } | |
493 | ||
b891c5a8 | 494 | ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp); |
c88130bc | 495 | if (ct == NULL) { |
0d53778e | 496 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
dacd2a1a YK |
497 | atomic_dec(&nf_conntrack_count); |
498 | return ERR_PTR(-ENOMEM); | |
9fb9cbb1 YK |
499 | } |
500 | ||
c88130bc PM |
501 | atomic_set(&ct->ct_general.use, 1); |
502 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; | |
503 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
9fb9cbb1 | 504 | /* Don't set timer yet: wait for confirmation */ |
c88130bc PM |
505 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
506 | INIT_RCU_HEAD(&ct->rcu); | |
9fb9cbb1 | 507 | |
c88130bc | 508 | return ct; |
9fb9cbb1 | 509 | } |
13b18339 | 510 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 511 | |
76507f69 | 512 | static void nf_conntrack_free_rcu(struct rcu_head *head) |
9fb9cbb1 | 513 | { |
76507f69 PM |
514 | struct nf_conn *ct = container_of(head, struct nf_conn, rcu); |
515 | ||
516 | nf_ct_ext_free(ct); | |
517 | kmem_cache_free(nf_conntrack_cachep, ct); | |
9fb9cbb1 YK |
518 | atomic_dec(&nf_conntrack_count); |
519 | } | |
76507f69 | 520 | |
c88130bc | 521 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 522 | { |
ceeff754 | 523 | nf_ct_ext_destroy(ct); |
c88130bc | 524 | call_rcu(&ct->rcu, nf_conntrack_free_rcu); |
76507f69 | 525 | } |
13b18339 | 526 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
527 | |
528 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
529 | failed due to stress. Otherwise it really is unclassifiable. */ | |
530 | static struct nf_conntrack_tuple_hash * | |
531 | init_conntrack(const struct nf_conntrack_tuple *tuple, | |
532 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 533 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
534 | struct sk_buff *skb, |
535 | unsigned int dataoff) | |
536 | { | |
c88130bc | 537 | struct nf_conn *ct; |
3c158f7f | 538 | struct nf_conn_help *help; |
9fb9cbb1 YK |
539 | struct nf_conntrack_tuple repl_tuple; |
540 | struct nf_conntrack_expect *exp; | |
541 | ||
605dcad6 | 542 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 543 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
544 | return NULL; |
545 | } | |
546 | ||
b891c5a8 | 547 | ct = nf_conntrack_alloc(tuple, &repl_tuple, GFP_ATOMIC); |
c88130bc | 548 | if (ct == NULL || IS_ERR(ct)) { |
0d53778e | 549 | pr_debug("Can't allocate conntrack.\n"); |
c88130bc | 550 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 YK |
551 | } |
552 | ||
c88130bc PM |
553 | if (!l4proto->new(ct, skb, dataoff)) { |
554 | nf_conntrack_free(ct); | |
0d53778e | 555 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
556 | return NULL; |
557 | } | |
558 | ||
58401572 KPO |
559 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
560 | ||
f8ba1aff | 561 | spin_lock_bh(&nf_conntrack_lock); |
6823645d | 562 | exp = nf_ct_find_expectation(tuple); |
9fb9cbb1 | 563 | if (exp) { |
0d53778e | 564 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 565 | ct, exp); |
9fb9cbb1 | 566 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
567 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
568 | ct->master = exp->master; | |
ceceae1b | 569 | if (exp->helper) { |
c88130bc | 570 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b YK |
571 | if (help) |
572 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
573 | } |
574 | ||
9fb9cbb1 | 575 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 576 | ct->mark = exp->master->mark; |
7c9728c3 JM |
577 | #endif |
578 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 579 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 580 | #endif |
c88130bc | 581 | nf_conntrack_get(&ct->master->ct_general); |
9fb9cbb1 | 582 | NF_CT_STAT_INC(expect_new); |
22e7410b | 583 | } else { |
ceceae1b YK |
584 | struct nf_conntrack_helper *helper; |
585 | ||
586 | helper = __nf_ct_helper_find(&repl_tuple); | |
587 | if (helper) { | |
c88130bc | 588 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b | 589 | if (help) |
ceceae1b | 590 | rcu_assign_pointer(help->helper, helper); |
3c158f7f | 591 | } |
9fb9cbb1 | 592 | NF_CT_STAT_INC(new); |
22e7410b | 593 | } |
9fb9cbb1 YK |
594 | |
595 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
c88130bc | 596 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed); |
9fb9cbb1 | 597 | |
f8ba1aff | 598 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
599 | |
600 | if (exp) { | |
601 | if (exp->expectfn) | |
c88130bc | 602 | exp->expectfn(ct, exp); |
6823645d | 603 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
604 | } |
605 | ||
c88130bc | 606 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
607 | } |
608 | ||
609 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
610 | static inline struct nf_conn * | |
611 | resolve_normal_ct(struct sk_buff *skb, | |
612 | unsigned int dataoff, | |
613 | u_int16_t l3num, | |
614 | u_int8_t protonum, | |
615 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 616 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
617 | int *set_reply, |
618 | enum ip_conntrack_info *ctinfo) | |
619 | { | |
620 | struct nf_conntrack_tuple tuple; | |
621 | struct nf_conntrack_tuple_hash *h; | |
622 | struct nf_conn *ct; | |
623 | ||
bbe735e4 | 624 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 625 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 626 | l4proto)) { |
0d53778e | 627 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
628 | return NULL; |
629 | } | |
630 | ||
631 | /* look for tuple match */ | |
330f7db5 | 632 | h = nf_conntrack_find_get(&tuple); |
9fb9cbb1 | 633 | if (!h) { |
605dcad6 | 634 | h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
635 | if (!h) |
636 | return NULL; | |
637 | if (IS_ERR(h)) | |
638 | return (void *)h; | |
639 | } | |
640 | ct = nf_ct_tuplehash_to_ctrack(h); | |
641 | ||
642 | /* It exists; we have (non-exclusive) reference. */ | |
643 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
644 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
645 | /* Please set reply bit if this packet OK */ | |
646 | *set_reply = 1; | |
647 | } else { | |
648 | /* Once we've had two way comms, always ESTABLISHED. */ | |
649 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 650 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
651 | *ctinfo = IP_CT_ESTABLISHED; |
652 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
653 | pr_debug("nf_conntrack_in: related packet for %p\n", |
654 | ct); | |
9fb9cbb1 YK |
655 | *ctinfo = IP_CT_RELATED; |
656 | } else { | |
0d53778e | 657 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
658 | *ctinfo = IP_CT_NEW; |
659 | } | |
660 | *set_reply = 0; | |
661 | } | |
662 | skb->nfct = &ct->ct_general; | |
663 | skb->nfctinfo = *ctinfo; | |
664 | return ct; | |
665 | } | |
666 | ||
667 | unsigned int | |
76108cea | 668 | nf_conntrack_in(u_int8_t pf, unsigned int hooknum, struct sk_buff *skb) |
9fb9cbb1 YK |
669 | { |
670 | struct nf_conn *ct; | |
671 | enum ip_conntrack_info ctinfo; | |
672 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 673 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
674 | unsigned int dataoff; |
675 | u_int8_t protonum; | |
676 | int set_reply = 0; | |
677 | int ret; | |
678 | ||
679 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 680 | if (skb->nfct) { |
c0e912d7 | 681 | NF_CT_STAT_INC_ATOMIC(ignore); |
9fb9cbb1 YK |
682 | return NF_ACCEPT; |
683 | } | |
684 | ||
923f4902 | 685 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 686 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 687 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
688 | &dataoff, &protonum); |
689 | if (ret <= 0) { | |
0d53778e | 690 | pr_debug("not prepared to track yet or error occured\n"); |
d87d8469 YK |
691 | NF_CT_STAT_INC_ATOMIC(error); |
692 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
693 | return -ret; |
694 | } | |
695 | ||
76108cea | 696 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
697 | |
698 | /* It may be an special packet, error, unclean... | |
699 | * inverse of the return code tells to the netfilter | |
700 | * core what to do with the packet. */ | |
605dcad6 | 701 | if (l4proto->error != NULL && |
3db05fea | 702 | (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) { |
c0e912d7 PM |
703 | NF_CT_STAT_INC_ATOMIC(error); |
704 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
705 | return -ret; |
706 | } | |
707 | ||
3db05fea | 708 | ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto, |
9fb9cbb1 YK |
709 | &set_reply, &ctinfo); |
710 | if (!ct) { | |
711 | /* Not valid part of a connection */ | |
c0e912d7 | 712 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
713 | return NF_ACCEPT; |
714 | } | |
715 | ||
716 | if (IS_ERR(ct)) { | |
717 | /* Too stressed to deal. */ | |
c0e912d7 | 718 | NF_CT_STAT_INC_ATOMIC(drop); |
9fb9cbb1 YK |
719 | return NF_DROP; |
720 | } | |
721 | ||
3db05fea | 722 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 723 | |
3db05fea | 724 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
725 | if (ret < 0) { |
726 | /* Invalid: inverse of the return code tells | |
727 | * the netfilter core what to do */ | |
0d53778e | 728 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
729 | nf_conntrack_put(skb->nfct); |
730 | skb->nfct = NULL; | |
c0e912d7 | 731 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
732 | return -ret; |
733 | } | |
734 | ||
735 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
3db05fea | 736 | nf_conntrack_event_cache(IPCT_STATUS, skb); |
9fb9cbb1 YK |
737 | |
738 | return ret; | |
739 | } | |
13b18339 | 740 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 741 | |
5f2b4c90 JE |
742 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
743 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 744 | { |
5f2b4c90 | 745 | bool ret; |
923f4902 PM |
746 | |
747 | rcu_read_lock(); | |
748 | ret = nf_ct_invert_tuple(inverse, orig, | |
749 | __nf_ct_l3proto_find(orig->src.l3num), | |
750 | __nf_ct_l4proto_find(orig->src.l3num, | |
751 | orig->dst.protonum)); | |
752 | rcu_read_unlock(); | |
753 | return ret; | |
9fb9cbb1 | 754 | } |
13b18339 | 755 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 756 | |
5b1158e9 JK |
757 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
758 | implicitly racy: see __nf_conntrack_confirm */ | |
759 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
760 | const struct nf_conntrack_tuple *newreply) | |
761 | { | |
762 | struct nf_conn_help *help = nfct_help(ct); | |
ceceae1b | 763 | struct nf_conntrack_helper *helper; |
5b1158e9 | 764 | |
5b1158e9 JK |
765 | /* Should be unconfirmed, so not in hash table yet */ |
766 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
767 | ||
0d53778e | 768 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 769 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
770 | |
771 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 772 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 773 | return; |
ceceae1b | 774 | |
c52fbb41 | 775 | rcu_read_lock(); |
ceceae1b YK |
776 | helper = __nf_ct_helper_find(newreply); |
777 | if (helper == NULL) { | |
778 | if (help) | |
779 | rcu_assign_pointer(help->helper, NULL); | |
780 | goto out; | |
5d78a849 | 781 | } |
ceceae1b YK |
782 | |
783 | if (help == NULL) { | |
b560580a PM |
784 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
785 | if (help == NULL) | |
ceceae1b | 786 | goto out; |
ceceae1b YK |
787 | } else { |
788 | memset(&help->help, 0, sizeof(help->help)); | |
789 | } | |
790 | ||
791 | rcu_assign_pointer(help->helper, helper); | |
792 | out: | |
c52fbb41 | 793 | rcu_read_unlock(); |
5b1158e9 | 794 | } |
13b18339 | 795 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 796 | |
9fb9cbb1 YK |
797 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
798 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
799 | enum ip_conntrack_info ctinfo, | |
800 | const struct sk_buff *skb, | |
801 | unsigned long extra_jiffies, | |
802 | int do_acct) | |
803 | { | |
804 | int event = 0; | |
805 | ||
806 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
807 | NF_CT_ASSERT(skb); | |
808 | ||
f8ba1aff | 809 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 810 | |
997ae831 | 811 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
812 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
813 | goto acct; | |
997ae831 | 814 | |
9fb9cbb1 YK |
815 | /* If not in hash table, timer will not be active yet */ |
816 | if (!nf_ct_is_confirmed(ct)) { | |
817 | ct->timeout.expires = extra_jiffies; | |
818 | event = IPCT_REFRESH; | |
819 | } else { | |
be00c8e4 MJ |
820 | unsigned long newtime = jiffies + extra_jiffies; |
821 | ||
822 | /* Only update the timeout if the new timeout is at least | |
823 | HZ jiffies from the old timeout. Need del_timer for race | |
824 | avoidance (may already be dying). */ | |
825 | if (newtime - ct->timeout.expires >= HZ | |
826 | && del_timer(&ct->timeout)) { | |
827 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
828 | add_timer(&ct->timeout); |
829 | event = IPCT_REFRESH; | |
830 | } | |
831 | } | |
832 | ||
47d95045 | 833 | acct: |
9fb9cbb1 | 834 | if (do_acct) { |
58401572 | 835 | struct nf_conn_counter *acct; |
3ffd5eeb | 836 | |
58401572 KPO |
837 | acct = nf_conn_acct_find(ct); |
838 | if (acct) { | |
839 | acct[CTINFO2DIR(ctinfo)].packets++; | |
840 | acct[CTINFO2DIR(ctinfo)].bytes += | |
841 | skb->len - skb_network_offset(skb); | |
842 | } | |
9fb9cbb1 | 843 | } |
9fb9cbb1 | 844 | |
f8ba1aff | 845 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
846 | |
847 | /* must be unlocked when calling event cache */ | |
848 | if (event) | |
849 | nf_conntrack_event_cache(event, skb); | |
850 | } | |
13b18339 | 851 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 852 | |
4c889498 DM |
853 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
854 | enum ip_conntrack_info ctinfo, | |
855 | const struct sk_buff *skb, | |
856 | int do_acct) | |
51091764 | 857 | { |
718d4ad9 | 858 | if (do_acct) { |
58401572 KPO |
859 | struct nf_conn_counter *acct; |
860 | ||
718d4ad9 | 861 | spin_lock_bh(&nf_conntrack_lock); |
58401572 KPO |
862 | acct = nf_conn_acct_find(ct); |
863 | if (acct) { | |
864 | acct[CTINFO2DIR(ctinfo)].packets++; | |
865 | acct[CTINFO2DIR(ctinfo)].bytes += | |
866 | skb->len - skb_network_offset(skb); | |
867 | } | |
718d4ad9 FH |
868 | spin_unlock_bh(&nf_conntrack_lock); |
869 | } | |
58401572 | 870 | |
4c889498 | 871 | if (del_timer(&ct->timeout)) { |
51091764 | 872 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
873 | return true; |
874 | } | |
875 | return false; | |
51091764 | 876 | } |
718d4ad9 | 877 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 878 | |
e281db5c | 879 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
880 | |
881 | #include <linux/netfilter/nfnetlink.h> | |
882 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
883 | #include <linux/mutex.h> |
884 | ||
c1d10adb PNA |
885 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
886 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
887 | * or depend on ctnetlink */ | |
fdf70832 | 888 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
889 | const struct nf_conntrack_tuple *tuple) |
890 | { | |
77236b6e PM |
891 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
892 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
893 | return 0; |
894 | ||
df6fb868 | 895 | nla_put_failure: |
c1d10adb PNA |
896 | return -1; |
897 | } | |
fdf70832 | 898 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 899 | |
f73e924c PM |
900 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
901 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
902 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 903 | }; |
f73e924c | 904 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 905 | |
fdf70832 | 906 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
907 | struct nf_conntrack_tuple *t) |
908 | { | |
df6fb868 | 909 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
910 | return -EINVAL; |
911 | ||
77236b6e PM |
912 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
913 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
914 | |
915 | return 0; | |
916 | } | |
fdf70832 | 917 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
c1d10adb PNA |
918 | #endif |
919 | ||
9fb9cbb1 | 920 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 921 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
922 | { |
923 | struct nf_conn *ct; | |
924 | enum ip_conntrack_info ctinfo; | |
925 | ||
926 | /* This ICMP is in reverse direction to the packet which caused it */ | |
927 | ct = nf_ct_get(skb, &ctinfo); | |
928 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
929 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
930 | else | |
931 | ctinfo = IP_CT_RELATED; | |
932 | ||
933 | /* Attach to new skbuff, and increment count */ | |
934 | nskb->nfct = &ct->ct_general; | |
935 | nskb->nfctinfo = ctinfo; | |
936 | nf_conntrack_get(nskb->nfct); | |
937 | } | |
938 | ||
9fb9cbb1 | 939 | /* Bring out ya dead! */ |
df0933dc | 940 | static struct nf_conn * |
9fb9cbb1 YK |
941 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
942 | void *data, unsigned int *bucket) | |
943 | { | |
df0933dc PM |
944 | struct nf_conntrack_tuple_hash *h; |
945 | struct nf_conn *ct; | |
f205c5e0 | 946 | struct hlist_node *n; |
9fb9cbb1 | 947 | |
f8ba1aff | 948 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 949 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
f205c5e0 | 950 | hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) { |
df0933dc PM |
951 | ct = nf_ct_tuplehash_to_ctrack(h); |
952 | if (iter(ct, data)) | |
953 | goto found; | |
954 | } | |
601e68e1 | 955 | } |
f205c5e0 | 956 | hlist_for_each_entry(h, n, &unconfirmed, hnode) { |
df0933dc PM |
957 | ct = nf_ct_tuplehash_to_ctrack(h); |
958 | if (iter(ct, data)) | |
ec68e97d | 959 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 960 | } |
f8ba1aff | 961 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
962 | return NULL; |
963 | found: | |
c073e3fa | 964 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 965 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 966 | return ct; |
9fb9cbb1 YK |
967 | } |
968 | ||
969 | void | |
970 | nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) | |
971 | { | |
df0933dc | 972 | struct nf_conn *ct; |
9fb9cbb1 YK |
973 | unsigned int bucket = 0; |
974 | ||
df0933dc | 975 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
976 | /* Time to push up daises... */ |
977 | if (del_timer(&ct->timeout)) | |
978 | death_by_timeout((unsigned long)ct); | |
979 | /* ... else the timer will get him soon. */ | |
980 | ||
981 | nf_ct_put(ct); | |
982 | } | |
983 | } | |
13b18339 | 984 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 YK |
985 | |
986 | static int kill_all(struct nf_conn *i, void *data) | |
987 | { | |
988 | return 1; | |
989 | } | |
990 | ||
96eb24d7 | 991 | void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
992 | { |
993 | if (vmalloced) | |
994 | vfree(hash); | |
995 | else | |
601e68e1 | 996 | free_pages((unsigned long)hash, |
f205c5e0 | 997 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 998 | } |
ac565e5f | 999 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1000 | |
272491ef | 1001 | void nf_conntrack_flush(void) |
c1d10adb PNA |
1002 | { |
1003 | nf_ct_iterate_cleanup(kill_all, NULL); | |
1004 | } | |
13b18339 | 1005 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 1006 | |
9fb9cbb1 YK |
1007 | /* Mishearing the voices in his head, our hero wonders how he's |
1008 | supposed to kill the mall. */ | |
dfdb8d79 | 1009 | void nf_conntrack_cleanup(struct net *net) |
9fb9cbb1 | 1010 | { |
c3a47ab3 | 1011 | rcu_assign_pointer(ip_ct_attach, NULL); |
7d3cdc6b | 1012 | |
9fb9cbb1 YK |
1013 | /* This makes sure all current packets have passed through |
1014 | netfilter framework. Roll on, two-stage module | |
1015 | delete... */ | |
1016 | synchronize_net(); | |
1017 | ||
1018 | nf_ct_event_cache_flush(); | |
1019 | i_see_dead_people: | |
c1d10adb | 1020 | nf_conntrack_flush(); |
9fb9cbb1 YK |
1021 | if (atomic_read(&nf_conntrack_count) != 0) { |
1022 | schedule(); | |
1023 | goto i_see_dead_people; | |
1024 | } | |
6636568c PM |
1025 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1026 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1027 | schedule(); | |
9fb9cbb1 | 1028 | |
de6e05c4 YK |
1029 | rcu_assign_pointer(nf_ct_destroy, NULL); |
1030 | ||
dacd2a1a | 1031 | kmem_cache_destroy(nf_conntrack_cachep); |
ac565e5f PM |
1032 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
1033 | nf_conntrack_htable_size); | |
5a6f294e | 1034 | |
58401572 | 1035 | nf_conntrack_acct_fini(); |
9714be7d KPO |
1036 | nf_conntrack_expect_fini(); |
1037 | nf_conntrack_helper_fini(); | |
1038 | nf_conntrack_proto_fini(); | |
9fb9cbb1 YK |
1039 | } |
1040 | ||
96eb24d7 | 1041 | struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced) |
9fb9cbb1 | 1042 | { |
f205c5e0 | 1043 | struct hlist_head *hash; |
8e5105a0 | 1044 | unsigned int size, i; |
9fb9cbb1 | 1045 | |
601e68e1 | 1046 | *vmalloced = 0; |
8e5105a0 | 1047 | |
f205c5e0 | 1048 | size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); |
29b67497 | 1049 | hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN, |
f205c5e0 | 1050 | get_order(sizeof(struct hlist_head) |
9fb9cbb1 | 1051 | * size)); |
601e68e1 | 1052 | if (!hash) { |
9fb9cbb1 YK |
1053 | *vmalloced = 1; |
1054 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
f205c5e0 | 1055 | hash = vmalloc(sizeof(struct hlist_head) * size); |
9fb9cbb1 YK |
1056 | } |
1057 | ||
1058 | if (hash) | |
601e68e1 | 1059 | for (i = 0; i < size; i++) |
f205c5e0 | 1060 | INIT_HLIST_HEAD(&hash[i]); |
9fb9cbb1 YK |
1061 | |
1062 | return hash; | |
1063 | } | |
ac565e5f | 1064 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1065 | |
fae718dd | 1066 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1067 | { |
96eb24d7 SH |
1068 | int i, bucket, vmalloced, old_vmalloced; |
1069 | unsigned int hashsize, old_size; | |
9fb9cbb1 | 1070 | int rnd; |
f205c5e0 | 1071 | struct hlist_head *hash, *old_hash; |
9fb9cbb1 YK |
1072 | struct nf_conntrack_tuple_hash *h; |
1073 | ||
1074 | /* On boot, we can set this without any fancy locking. */ | |
1075 | if (!nf_conntrack_htable_size) | |
1076 | return param_set_uint(val, kp); | |
1077 | ||
96eb24d7 | 1078 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1079 | if (!hashsize) |
1080 | return -EINVAL; | |
1081 | ||
ac565e5f | 1082 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); |
9fb9cbb1 YK |
1083 | if (!hash) |
1084 | return -ENOMEM; | |
1085 | ||
1086 | /* We have to rehahs for the new table anyway, so we also can | |
1087 | * use a newrandom seed */ | |
1088 | get_random_bytes(&rnd, 4); | |
1089 | ||
76507f69 PM |
1090 | /* Lookups in the old hash might happen in parallel, which means we |
1091 | * might get false negatives during connection lookup. New connections | |
1092 | * created because of a false negative won't make it into the hash | |
1093 | * though since that required taking the lock. | |
1094 | */ | |
f8ba1aff | 1095 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1096 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
f205c5e0 PM |
1097 | while (!hlist_empty(&nf_conntrack_hash[i])) { |
1098 | h = hlist_entry(nf_conntrack_hash[i].first, | |
1099 | struct nf_conntrack_tuple_hash, hnode); | |
76507f69 | 1100 | hlist_del_rcu(&h->hnode); |
9fb9cbb1 | 1101 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
f205c5e0 | 1102 | hlist_add_head(&h->hnode, &hash[bucket]); |
9fb9cbb1 YK |
1103 | } |
1104 | } | |
1105 | old_size = nf_conntrack_htable_size; | |
1106 | old_vmalloced = nf_conntrack_vmalloc; | |
1107 | old_hash = nf_conntrack_hash; | |
1108 | ||
1109 | nf_conntrack_htable_size = hashsize; | |
1110 | nf_conntrack_vmalloc = vmalloced; | |
1111 | nf_conntrack_hash = hash; | |
1112 | nf_conntrack_hash_rnd = rnd; | |
f8ba1aff | 1113 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1114 | |
ac565e5f | 1115 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1116 | return 0; |
1117 | } | |
fae718dd | 1118 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1119 | |
fae718dd | 1120 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1121 | &nf_conntrack_htable_size, 0600); |
1122 | ||
dfdb8d79 | 1123 | int nf_conntrack_init(struct net *net) |
9fb9cbb1 | 1124 | { |
f205c5e0 | 1125 | int max_factor = 8; |
9fb9cbb1 YK |
1126 | int ret; |
1127 | ||
1128 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1129 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1130 | if (!nf_conntrack_htable_size) { |
1131 | nf_conntrack_htable_size | |
1132 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
f205c5e0 | 1133 | / sizeof(struct hlist_head)); |
9fb9cbb1 | 1134 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1135 | nf_conntrack_htable_size = 16384; |
1136 | if (nf_conntrack_htable_size < 32) | |
1137 | nf_conntrack_htable_size = 32; | |
1138 | ||
1139 | /* Use a max. factor of four by default to get the same max as | |
1140 | * with the old struct list_heads. When a table size is given | |
1141 | * we use the old value of 8 to avoid reducing the max. | |
1142 | * entries. */ | |
1143 | max_factor = 4; | |
9fb9cbb1 | 1144 | } |
ac565e5f PM |
1145 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
1146 | &nf_conntrack_vmalloc); | |
9fb9cbb1 YK |
1147 | if (!nf_conntrack_hash) { |
1148 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1149 | goto err_out; | |
1150 | } | |
1151 | ||
f205c5e0 | 1152 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1153 | |
1154 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1155 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1156 | nf_conntrack_max); | |
1157 | ||
dacd2a1a YK |
1158 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1159 | sizeof(struct nf_conn), | |
20c2df83 | 1160 | 0, 0, NULL); |
dacd2a1a | 1161 | if (!nf_conntrack_cachep) { |
9fb9cbb1 YK |
1162 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
1163 | goto err_free_hash; | |
1164 | } | |
1165 | ||
e9c1b084 PM |
1166 | ret = nf_conntrack_proto_init(); |
1167 | if (ret < 0) | |
9fb9cbb1 | 1168 | goto err_free_conntrack_slab; |
9fb9cbb1 | 1169 | |
e9c1b084 | 1170 | ret = nf_conntrack_expect_init(); |
933a41e7 | 1171 | if (ret < 0) |
e9c1b084 | 1172 | goto out_fini_proto; |
933a41e7 | 1173 | |
ceceae1b YK |
1174 | ret = nf_conntrack_helper_init(); |
1175 | if (ret < 0) | |
e9c1b084 | 1176 | goto out_fini_expect; |
ceceae1b | 1177 | |
58401572 KPO |
1178 | ret = nf_conntrack_acct_init(); |
1179 | if (ret < 0) | |
1180 | goto out_fini_helper; | |
1181 | ||
7d3cdc6b | 1182 | /* For use by REJECT target */ |
b334aadc | 1183 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); |
de6e05c4 | 1184 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); |
7d3cdc6b | 1185 | |
9fb9cbb1 YK |
1186 | /* Set up fake conntrack: |
1187 | - to never be deleted, not in any hashes */ | |
1188 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); | |
1189 | /* - and look it like as a confirmed connection */ | |
1190 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1191 | ||
1192 | return ret; | |
1193 | ||
58401572 KPO |
1194 | out_fini_helper: |
1195 | nf_conntrack_helper_fini(); | |
e9c1b084 PM |
1196 | out_fini_expect: |
1197 | nf_conntrack_expect_fini(); | |
ceceae1b YK |
1198 | out_fini_proto: |
1199 | nf_conntrack_proto_fini(); | |
9fb9cbb1 | 1200 | err_free_conntrack_slab: |
dacd2a1a | 1201 | kmem_cache_destroy(nf_conntrack_cachep); |
9fb9cbb1 | 1202 | err_free_hash: |
ac565e5f PM |
1203 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
1204 | nf_conntrack_htable_size); | |
9fb9cbb1 YK |
1205 | err_out: |
1206 | return -ENOMEM; | |
1207 | } |