]>
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 | |
f8ba1aff | 43 | DEFINE_SPINLOCK(nf_conntrack_lock); |
13b18339 | 44 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 YK |
45 | |
46 | /* nf_conntrack_standalone needs this */ | |
47 | atomic_t nf_conntrack_count = ATOMIC_INIT(0); | |
a999e683 | 48 | EXPORT_SYMBOL_GPL(nf_conntrack_count); |
9fb9cbb1 | 49 | |
e2b7606c | 50 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
51 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
52 | ||
94aec08e | 53 | int nf_conntrack_max __read_mostly; |
a999e683 | 54 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 55 | |
f205c5e0 | 56 | struct hlist_head *nf_conntrack_hash __read_mostly; |
13b18339 PM |
57 | EXPORT_SYMBOL_GPL(nf_conntrack_hash); |
58 | ||
e2b7606c | 59 | struct nf_conn nf_conntrack_untracked __read_mostly; |
13b18339 PM |
60 | EXPORT_SYMBOL_GPL(nf_conntrack_untracked); |
61 | ||
94aec08e | 62 | unsigned int nf_ct_log_invalid __read_mostly; |
f205c5e0 | 63 | HLIST_HEAD(unconfirmed); |
1192e403 | 64 | static int nf_conntrack_vmalloc __read_mostly; |
dacd2a1a | 65 | static struct kmem_cache *nf_conntrack_cachep __read_mostly; |
77ab9cff | 66 | |
9fb9cbb1 YK |
67 | DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); |
68 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); | |
69 | ||
9fb9cbb1 YK |
70 | static int nf_conntrack_hash_rnd_initted; |
71 | static unsigned int nf_conntrack_hash_rnd; | |
72 | ||
73 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
74 | unsigned int size, unsigned int rnd) | |
75 | { | |
0794935e PM |
76 | unsigned int n; |
77 | u_int32_t h; | |
78 | ||
79 | /* The direction must be ignored, so we hash everything up to the | |
80 | * destination ports (which is a multiple of 4) and treat the last | |
81 | * three bytes manually. | |
82 | */ | |
83 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
84 | h = jhash2((u32 *)tuple, n, | |
85 | rnd ^ (((__force __u16)tuple->dst.u.all << 16) | | |
86 | tuple->dst.protonum)); | |
87 | ||
88 | return ((u64)h * size) >> 32; | |
9fb9cbb1 YK |
89 | } |
90 | ||
91 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
92 | { | |
93 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
94 | nf_conntrack_hash_rnd); | |
95 | } | |
96 | ||
5f2b4c90 | 97 | bool |
9fb9cbb1 YK |
98 | nf_ct_get_tuple(const struct sk_buff *skb, |
99 | unsigned int nhoff, | |
100 | unsigned int dataoff, | |
101 | u_int16_t l3num, | |
102 | u_int8_t protonum, | |
103 | struct nf_conntrack_tuple *tuple, | |
104 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 105 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 106 | { |
443a70d5 | 107 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
108 | |
109 | tuple->src.l3num = l3num; | |
110 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 111 | return false; |
9fb9cbb1 YK |
112 | |
113 | tuple->dst.protonum = protonum; | |
114 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
115 | ||
605dcad6 | 116 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 117 | } |
13b18339 | 118 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 119 | |
5f2b4c90 JE |
120 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
121 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
122 | { |
123 | struct nf_conntrack_l3proto *l3proto; | |
124 | struct nf_conntrack_l4proto *l4proto; | |
125 | unsigned int protoff; | |
126 | u_int8_t protonum; | |
127 | int ret; | |
128 | ||
129 | rcu_read_lock(); | |
130 | ||
131 | l3proto = __nf_ct_l3proto_find(l3num); | |
132 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
133 | if (ret != NF_ACCEPT) { | |
134 | rcu_read_unlock(); | |
5f2b4c90 | 135 | return false; |
e2a3123f YK |
136 | } |
137 | ||
138 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
139 | ||
140 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
141 | l3proto, l4proto); | |
142 | ||
143 | rcu_read_unlock(); | |
144 | return ret; | |
145 | } | |
146 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
147 | ||
5f2b4c90 | 148 | bool |
9fb9cbb1 YK |
149 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
150 | const struct nf_conntrack_tuple *orig, | |
151 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 152 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 153 | { |
443a70d5 | 154 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
155 | |
156 | inverse->src.l3num = orig->src.l3num; | |
157 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 158 | return false; |
9fb9cbb1 YK |
159 | |
160 | inverse->dst.dir = !orig->dst.dir; | |
161 | ||
162 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 163 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 164 | } |
13b18339 | 165 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 166 | |
9fb9cbb1 YK |
167 | static void |
168 | clean_from_lists(struct nf_conn *ct) | |
169 | { | |
0d53778e | 170 | pr_debug("clean_from_lists(%p)\n", ct); |
76507f69 PM |
171 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
172 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); | |
9fb9cbb1 YK |
173 | |
174 | /* Destroy all pending expectations */ | |
c1d10adb | 175 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
176 | } |
177 | ||
178 | static void | |
179 | destroy_conntrack(struct nf_conntrack *nfct) | |
180 | { | |
181 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 182 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 183 | |
0d53778e | 184 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
185 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
186 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
187 | ||
188 | nf_conntrack_event(IPCT_DESTROY, ct); | |
189 | set_bit(IPS_DYING_BIT, &ct->status); | |
190 | ||
191 | /* To make sure we don't get any weird locking issues here: | |
192 | * destroy_conntrack() MUST NOT be called with a write lock | |
193 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 194 | rcu_read_lock(); |
5e8fbe2a | 195 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
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 | 202 | |
f8ba1aff | 203 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
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); | |
f8ba1aff | 217 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
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 | 239 | |
f8ba1aff | 240 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
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); | |
f8ba1aff | 245 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
246 | nf_ct_put(ct); |
247 | } | |
248 | ||
c1d10adb | 249 | struct nf_conntrack_tuple_hash * |
ba419aff | 250 | __nf_conntrack_find(const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
251 | { |
252 | struct nf_conntrack_tuple_hash *h; | |
f205c5e0 | 253 | struct hlist_node *n; |
9fb9cbb1 YK |
254 | unsigned int hash = hash_conntrack(tuple); |
255 | ||
4e29e9ec PM |
256 | /* Disable BHs the entire time since we normally need to disable them |
257 | * at least once for the stats anyway. | |
258 | */ | |
259 | local_bh_disable(); | |
76507f69 | 260 | hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) { |
ba419aff | 261 | if (nf_ct_tuple_equal(tuple, &h->tuple)) { |
9fb9cbb1 | 262 | NF_CT_STAT_INC(found); |
4e29e9ec | 263 | local_bh_enable(); |
9fb9cbb1 YK |
264 | return h; |
265 | } | |
266 | NF_CT_STAT_INC(searched); | |
267 | } | |
4e29e9ec | 268 | local_bh_enable(); |
9fb9cbb1 YK |
269 | |
270 | return NULL; | |
271 | } | |
13b18339 | 272 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
273 | |
274 | /* Find a connection corresponding to a tuple. */ | |
275 | struct nf_conntrack_tuple_hash * | |
330f7db5 | 276 | nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
277 | { |
278 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 279 | struct nf_conn *ct; |
9fb9cbb1 | 280 | |
76507f69 | 281 | rcu_read_lock(); |
ba419aff | 282 | h = __nf_conntrack_find(tuple); |
76507f69 PM |
283 | if (h) { |
284 | ct = nf_ct_tuplehash_to_ctrack(h); | |
285 | if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
286 | h = NULL; | |
287 | } | |
288 | rcu_read_unlock(); | |
9fb9cbb1 YK |
289 | |
290 | return h; | |
291 | } | |
13b18339 | 292 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 293 | |
c1d10adb PNA |
294 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
295 | unsigned int hash, | |
601e68e1 | 296 | unsigned int repl_hash) |
c1d10adb | 297 | { |
76507f69 PM |
298 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
299 | &nf_conntrack_hash[hash]); | |
300 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, | |
301 | &nf_conntrack_hash[repl_hash]); | |
c1d10adb PNA |
302 | } |
303 | ||
304 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
305 | { | |
306 | unsigned int hash, repl_hash; | |
307 | ||
308 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
309 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
310 | ||
f8ba1aff | 311 | spin_lock_bh(&nf_conntrack_lock); |
c1d10adb | 312 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
f8ba1aff | 313 | spin_unlock_bh(&nf_conntrack_lock); |
c1d10adb | 314 | } |
13b18339 | 315 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 316 | |
9fb9cbb1 YK |
317 | /* Confirm a connection given skb; places it in hash table */ |
318 | int | |
3db05fea | 319 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
320 | { |
321 | unsigned int hash, repl_hash; | |
df0933dc | 322 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 323 | struct nf_conn *ct; |
df0933dc | 324 | struct nf_conn_help *help; |
f205c5e0 | 325 | struct hlist_node *n; |
9fb9cbb1 YK |
326 | enum ip_conntrack_info ctinfo; |
327 | ||
3db05fea | 328 | ct = nf_ct_get(skb, &ctinfo); |
9fb9cbb1 YK |
329 | |
330 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
331 | ICMP/TCP RST packets in other direction. Actual packet | |
332 | which created connection will be IP_CT_NEW or for an | |
333 | expected connection, IP_CT_RELATED. */ | |
334 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
335 | return NF_ACCEPT; | |
336 | ||
337 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
338 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
339 | ||
340 | /* We're not in hash table, and we refuse to set up related | |
341 | connections for unconfirmed conns. But packet copies and | |
342 | REJECT will give spurious warnings here. */ | |
343 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
344 | ||
345 | /* No external references means noone else could have | |
346 | confirmed us. */ | |
347 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 348 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 349 | |
f8ba1aff | 350 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
351 | |
352 | /* See if there's one in the list already, including reverse: | |
353 | NAT could have grabbed it without realizing, since we're | |
354 | not in the hash. If there is, we lost race. */ | |
f205c5e0 | 355 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) |
df0933dc PM |
356 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
357 | &h->tuple)) | |
358 | goto out; | |
f205c5e0 | 359 | hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode) |
df0933dc PM |
360 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
361 | &h->tuple)) | |
362 | goto out; | |
9fb9cbb1 | 363 | |
df0933dc | 364 | /* Remove from unconfirmed list */ |
f205c5e0 | 365 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
df0933dc PM |
366 | |
367 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
368 | /* Timer relative to confirmation time, not original | |
369 | setting time, otherwise we'd get timer wrap in | |
370 | weird delay cases. */ | |
371 | ct->timeout.expires += jiffies; | |
372 | add_timer(&ct->timeout); | |
373 | atomic_inc(&ct->ct_general.use); | |
374 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
375 | NF_CT_STAT_INC(insert); | |
f8ba1aff | 376 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
377 | help = nfct_help(ct); |
378 | if (help && help->helper) | |
3db05fea | 379 | nf_conntrack_event_cache(IPCT_HELPER, skb); |
9fb9cbb1 | 380 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
381 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
382 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
3db05fea | 383 | nf_conntrack_event_cache(IPCT_NATINFO, skb); |
9fb9cbb1 | 384 | #endif |
df0933dc | 385 | nf_conntrack_event_cache(master_ct(ct) ? |
3db05fea | 386 | IPCT_RELATED : IPCT_NEW, skb); |
df0933dc | 387 | return NF_ACCEPT; |
9fb9cbb1 | 388 | |
df0933dc | 389 | out: |
9fb9cbb1 | 390 | NF_CT_STAT_INC(insert_failed); |
f8ba1aff | 391 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
392 | return NF_DROP; |
393 | } | |
13b18339 | 394 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
395 | |
396 | /* Returns true if a connection correspondings to the tuple (required | |
397 | for NAT). */ | |
398 | int | |
399 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
400 | const struct nf_conn *ignored_conntrack) | |
401 | { | |
402 | struct nf_conntrack_tuple_hash *h; | |
ba419aff PM |
403 | struct hlist_node *n; |
404 | unsigned int hash = hash_conntrack(tuple); | |
9fb9cbb1 | 405 | |
4e29e9ec PM |
406 | /* Disable BHs the entire time since we need to disable them at |
407 | * least once for the stats anyway. | |
408 | */ | |
409 | rcu_read_lock_bh(); | |
ba419aff PM |
410 | hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) { |
411 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && | |
412 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
413 | NF_CT_STAT_INC(found); | |
4e29e9ec | 414 | rcu_read_unlock_bh(); |
ba419aff PM |
415 | return 1; |
416 | } | |
417 | NF_CT_STAT_INC(searched); | |
418 | } | |
4e29e9ec | 419 | rcu_read_unlock_bh(); |
9fb9cbb1 | 420 | |
ba419aff | 421 | return 0; |
9fb9cbb1 | 422 | } |
13b18339 | 423 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 424 | |
7ae7730f PM |
425 | #define NF_CT_EVICTION_RANGE 8 |
426 | ||
9fb9cbb1 YK |
427 | /* There's a small race here where we may free a just-assured |
428 | connection. Too bad: we're in trouble anyway. */ | |
76eb9460 | 429 | static noinline int early_drop(unsigned int hash) |
9fb9cbb1 | 430 | { |
f205c5e0 | 431 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 432 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 433 | struct nf_conn *ct = NULL, *tmp; |
f205c5e0 | 434 | struct hlist_node *n; |
7ae7730f | 435 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
436 | int dropped = 0; |
437 | ||
76507f69 | 438 | rcu_read_lock(); |
7ae7730f | 439 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
76507f69 PM |
440 | hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], |
441 | hnode) { | |
7ae7730f PM |
442 | tmp = nf_ct_tuplehash_to_ctrack(h); |
443 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
444 | ct = tmp; | |
445 | cnt++; | |
446 | } | |
76507f69 PM |
447 | |
448 | if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
449 | ct = NULL; | |
7ae7730f PM |
450 | if (ct || cnt >= NF_CT_EVICTION_RANGE) |
451 | break; | |
452 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 453 | } |
76507f69 | 454 | rcu_read_unlock(); |
9fb9cbb1 YK |
455 | |
456 | if (!ct) | |
457 | return dropped; | |
458 | ||
459 | if (del_timer(&ct->timeout)) { | |
460 | death_by_timeout((unsigned long)ct); | |
461 | dropped = 1; | |
c0e912d7 | 462 | NF_CT_STAT_INC_ATOMIC(early_drop); |
9fb9cbb1 YK |
463 | } |
464 | nf_ct_put(ct); | |
465 | return dropped; | |
466 | } | |
467 | ||
dacd2a1a YK |
468 | struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, |
469 | const struct nf_conntrack_tuple *repl) | |
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 | ||
c88130bc PM |
494 | ct = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC); |
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 | { |
c88130bc | 523 | call_rcu(&ct->rcu, nf_conntrack_free_rcu); |
76507f69 | 524 | } |
13b18339 | 525 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
526 | |
527 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
528 | failed due to stress. Otherwise it really is unclassifiable. */ | |
529 | static struct nf_conntrack_tuple_hash * | |
530 | init_conntrack(const struct nf_conntrack_tuple *tuple, | |
531 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 532 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
533 | struct sk_buff *skb, |
534 | unsigned int dataoff) | |
535 | { | |
c88130bc | 536 | struct nf_conn *ct; |
3c158f7f | 537 | struct nf_conn_help *help; |
9fb9cbb1 YK |
538 | struct nf_conntrack_tuple repl_tuple; |
539 | struct nf_conntrack_expect *exp; | |
540 | ||
605dcad6 | 541 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 542 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
543 | return NULL; |
544 | } | |
545 | ||
c88130bc PM |
546 | ct = nf_conntrack_alloc(tuple, &repl_tuple); |
547 | if (ct == NULL || IS_ERR(ct)) { | |
0d53778e | 548 | pr_debug("Can't allocate conntrack.\n"); |
c88130bc | 549 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 YK |
550 | } |
551 | ||
c88130bc PM |
552 | if (!l4proto->new(ct, skb, dataoff)) { |
553 | nf_conntrack_free(ct); | |
0d53778e | 554 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
555 | return NULL; |
556 | } | |
557 | ||
f8ba1aff | 558 | spin_lock_bh(&nf_conntrack_lock); |
6823645d | 559 | exp = nf_ct_find_expectation(tuple); |
9fb9cbb1 | 560 | if (exp) { |
0d53778e | 561 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 562 | ct, exp); |
9fb9cbb1 | 563 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
564 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
565 | ct->master = exp->master; | |
ceceae1b | 566 | if (exp->helper) { |
c88130bc | 567 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b YK |
568 | if (help) |
569 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
570 | } |
571 | ||
9fb9cbb1 | 572 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 573 | ct->mark = exp->master->mark; |
7c9728c3 JM |
574 | #endif |
575 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 576 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 577 | #endif |
c88130bc | 578 | nf_conntrack_get(&ct->master->ct_general); |
9fb9cbb1 | 579 | NF_CT_STAT_INC(expect_new); |
22e7410b | 580 | } else { |
ceceae1b YK |
581 | struct nf_conntrack_helper *helper; |
582 | ||
583 | helper = __nf_ct_helper_find(&repl_tuple); | |
584 | if (helper) { | |
c88130bc | 585 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b | 586 | if (help) |
ceceae1b | 587 | rcu_assign_pointer(help->helper, helper); |
3c158f7f | 588 | } |
9fb9cbb1 | 589 | NF_CT_STAT_INC(new); |
22e7410b | 590 | } |
9fb9cbb1 YK |
591 | |
592 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
c88130bc | 593 | hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed); |
9fb9cbb1 | 594 | |
f8ba1aff | 595 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
596 | |
597 | if (exp) { | |
598 | if (exp->expectfn) | |
c88130bc | 599 | exp->expectfn(ct, exp); |
6823645d | 600 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
601 | } |
602 | ||
c88130bc | 603 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
604 | } |
605 | ||
606 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
607 | static inline struct nf_conn * | |
608 | resolve_normal_ct(struct sk_buff *skb, | |
609 | unsigned int dataoff, | |
610 | u_int16_t l3num, | |
611 | u_int8_t protonum, | |
612 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 613 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
614 | int *set_reply, |
615 | enum ip_conntrack_info *ctinfo) | |
616 | { | |
617 | struct nf_conntrack_tuple tuple; | |
618 | struct nf_conntrack_tuple_hash *h; | |
619 | struct nf_conn *ct; | |
620 | ||
bbe735e4 | 621 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 622 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 623 | l4proto)) { |
0d53778e | 624 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
625 | return NULL; |
626 | } | |
627 | ||
628 | /* look for tuple match */ | |
330f7db5 | 629 | h = nf_conntrack_find_get(&tuple); |
9fb9cbb1 | 630 | if (!h) { |
605dcad6 | 631 | h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
632 | if (!h) |
633 | return NULL; | |
634 | if (IS_ERR(h)) | |
635 | return (void *)h; | |
636 | } | |
637 | ct = nf_ct_tuplehash_to_ctrack(h); | |
638 | ||
639 | /* It exists; we have (non-exclusive) reference. */ | |
640 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
641 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
642 | /* Please set reply bit if this packet OK */ | |
643 | *set_reply = 1; | |
644 | } else { | |
645 | /* Once we've had two way comms, always ESTABLISHED. */ | |
646 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 647 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
648 | *ctinfo = IP_CT_ESTABLISHED; |
649 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
650 | pr_debug("nf_conntrack_in: related packet for %p\n", |
651 | ct); | |
9fb9cbb1 YK |
652 | *ctinfo = IP_CT_RELATED; |
653 | } else { | |
0d53778e | 654 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
655 | *ctinfo = IP_CT_NEW; |
656 | } | |
657 | *set_reply = 0; | |
658 | } | |
659 | skb->nfct = &ct->ct_general; | |
660 | skb->nfctinfo = *ctinfo; | |
661 | return ct; | |
662 | } | |
663 | ||
664 | unsigned int | |
3db05fea | 665 | nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb) |
9fb9cbb1 YK |
666 | { |
667 | struct nf_conn *ct; | |
668 | enum ip_conntrack_info ctinfo; | |
669 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 670 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
671 | unsigned int dataoff; |
672 | u_int8_t protonum; | |
673 | int set_reply = 0; | |
674 | int ret; | |
675 | ||
676 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 677 | if (skb->nfct) { |
c0e912d7 | 678 | NF_CT_STAT_INC_ATOMIC(ignore); |
9fb9cbb1 YK |
679 | return NF_ACCEPT; |
680 | } | |
681 | ||
923f4902 | 682 | /* rcu_read_lock()ed by nf_hook_slow */ |
c1d10adb | 683 | l3proto = __nf_ct_l3proto_find((u_int16_t)pf); |
3db05fea | 684 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
685 | &dataoff, &protonum); |
686 | if (ret <= 0) { | |
0d53778e | 687 | pr_debug("not prepared to track yet or error occured\n"); |
d87d8469 YK |
688 | NF_CT_STAT_INC_ATOMIC(error); |
689 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
690 | return -ret; |
691 | } | |
692 | ||
605dcad6 | 693 | l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); |
9fb9cbb1 YK |
694 | |
695 | /* It may be an special packet, error, unclean... | |
696 | * inverse of the return code tells to the netfilter | |
697 | * core what to do with the packet. */ | |
605dcad6 | 698 | if (l4proto->error != NULL && |
3db05fea | 699 | (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) { |
c0e912d7 PM |
700 | NF_CT_STAT_INC_ATOMIC(error); |
701 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
702 | return -ret; |
703 | } | |
704 | ||
3db05fea | 705 | ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto, |
9fb9cbb1 YK |
706 | &set_reply, &ctinfo); |
707 | if (!ct) { | |
708 | /* Not valid part of a connection */ | |
c0e912d7 | 709 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
710 | return NF_ACCEPT; |
711 | } | |
712 | ||
713 | if (IS_ERR(ct)) { | |
714 | /* Too stressed to deal. */ | |
c0e912d7 | 715 | NF_CT_STAT_INC_ATOMIC(drop); |
9fb9cbb1 YK |
716 | return NF_DROP; |
717 | } | |
718 | ||
3db05fea | 719 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 720 | |
3db05fea | 721 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
722 | if (ret < 0) { |
723 | /* Invalid: inverse of the return code tells | |
724 | * the netfilter core what to do */ | |
0d53778e | 725 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
726 | nf_conntrack_put(skb->nfct); |
727 | skb->nfct = NULL; | |
c0e912d7 | 728 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
729 | return -ret; |
730 | } | |
731 | ||
732 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
3db05fea | 733 | nf_conntrack_event_cache(IPCT_STATUS, skb); |
9fb9cbb1 YK |
734 | |
735 | return ret; | |
736 | } | |
13b18339 | 737 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 738 | |
5f2b4c90 JE |
739 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
740 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 741 | { |
5f2b4c90 | 742 | bool ret; |
923f4902 PM |
743 | |
744 | rcu_read_lock(); | |
745 | ret = nf_ct_invert_tuple(inverse, orig, | |
746 | __nf_ct_l3proto_find(orig->src.l3num), | |
747 | __nf_ct_l4proto_find(orig->src.l3num, | |
748 | orig->dst.protonum)); | |
749 | rcu_read_unlock(); | |
750 | return ret; | |
9fb9cbb1 | 751 | } |
13b18339 | 752 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 753 | |
5b1158e9 JK |
754 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
755 | implicitly racy: see __nf_conntrack_confirm */ | |
756 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
757 | const struct nf_conntrack_tuple *newreply) | |
758 | { | |
759 | struct nf_conn_help *help = nfct_help(ct); | |
ceceae1b | 760 | struct nf_conntrack_helper *helper; |
5b1158e9 | 761 | |
5b1158e9 JK |
762 | /* Should be unconfirmed, so not in hash table yet */ |
763 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
764 | ||
0d53778e | 765 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 766 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
767 | |
768 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 769 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 770 | return; |
ceceae1b | 771 | |
c52fbb41 | 772 | rcu_read_lock(); |
ceceae1b YK |
773 | helper = __nf_ct_helper_find(newreply); |
774 | if (helper == NULL) { | |
775 | if (help) | |
776 | rcu_assign_pointer(help->helper, NULL); | |
777 | goto out; | |
5d78a849 | 778 | } |
ceceae1b YK |
779 | |
780 | if (help == NULL) { | |
b560580a PM |
781 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
782 | if (help == NULL) | |
ceceae1b | 783 | goto out; |
ceceae1b YK |
784 | } else { |
785 | memset(&help->help, 0, sizeof(help->help)); | |
786 | } | |
787 | ||
788 | rcu_assign_pointer(help->helper, helper); | |
789 | out: | |
c52fbb41 | 790 | rcu_read_unlock(); |
5b1158e9 | 791 | } |
13b18339 | 792 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 793 | |
9fb9cbb1 YK |
794 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
795 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
796 | enum ip_conntrack_info ctinfo, | |
797 | const struct sk_buff *skb, | |
798 | unsigned long extra_jiffies, | |
799 | int do_acct) | |
800 | { | |
801 | int event = 0; | |
802 | ||
803 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
804 | NF_CT_ASSERT(skb); | |
805 | ||
f8ba1aff | 806 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 807 | |
997ae831 | 808 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
809 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
810 | goto acct; | |
997ae831 | 811 | |
9fb9cbb1 YK |
812 | /* If not in hash table, timer will not be active yet */ |
813 | if (!nf_ct_is_confirmed(ct)) { | |
814 | ct->timeout.expires = extra_jiffies; | |
815 | event = IPCT_REFRESH; | |
816 | } else { | |
be00c8e4 MJ |
817 | unsigned long newtime = jiffies + extra_jiffies; |
818 | ||
819 | /* Only update the timeout if the new timeout is at least | |
820 | HZ jiffies from the old timeout. Need del_timer for race | |
821 | avoidance (may already be dying). */ | |
822 | if (newtime - ct->timeout.expires >= HZ | |
823 | && del_timer(&ct->timeout)) { | |
824 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
825 | add_timer(&ct->timeout); |
826 | event = IPCT_REFRESH; | |
827 | } | |
828 | } | |
829 | ||
47d95045 | 830 | acct: |
9fb9cbb1 YK |
831 | #ifdef CONFIG_NF_CT_ACCT |
832 | if (do_acct) { | |
833 | ct->counters[CTINFO2DIR(ctinfo)].packets++; | |
834 | ct->counters[CTINFO2DIR(ctinfo)].bytes += | |
bbe735e4 | 835 | skb->len - skb_network_offset(skb); |
3ffd5eeb MJ |
836 | |
837 | if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) | |
838 | || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) | |
839 | event |= IPCT_COUNTER_FILLING; | |
9fb9cbb1 YK |
840 | } |
841 | #endif | |
842 | ||
f8ba1aff | 843 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
844 | |
845 | /* must be unlocked when calling event cache */ | |
846 | if (event) | |
847 | nf_conntrack_event_cache(event, skb); | |
848 | } | |
13b18339 | 849 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 850 | |
e281db5c | 851 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
852 | |
853 | #include <linux/netfilter/nfnetlink.h> | |
854 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
855 | #include <linux/mutex.h> |
856 | ||
c1d10adb PNA |
857 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
858 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
859 | * or depend on ctnetlink */ | |
fdf70832 | 860 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
861 | const struct nf_conntrack_tuple *tuple) |
862 | { | |
77236b6e PM |
863 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
864 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
865 | return 0; |
866 | ||
df6fb868 | 867 | nla_put_failure: |
c1d10adb PNA |
868 | return -1; |
869 | } | |
fdf70832 | 870 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 871 | |
f73e924c PM |
872 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
873 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
874 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 875 | }; |
f73e924c | 876 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 877 | |
fdf70832 | 878 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
879 | struct nf_conntrack_tuple *t) |
880 | { | |
df6fb868 | 881 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
882 | return -EINVAL; |
883 | ||
77236b6e PM |
884 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
885 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
886 | |
887 | return 0; | |
888 | } | |
fdf70832 | 889 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
c1d10adb PNA |
890 | #endif |
891 | ||
9fb9cbb1 | 892 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 893 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
894 | { |
895 | struct nf_conn *ct; | |
896 | enum ip_conntrack_info ctinfo; | |
897 | ||
898 | /* This ICMP is in reverse direction to the packet which caused it */ | |
899 | ct = nf_ct_get(skb, &ctinfo); | |
900 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
901 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
902 | else | |
903 | ctinfo = IP_CT_RELATED; | |
904 | ||
905 | /* Attach to new skbuff, and increment count */ | |
906 | nskb->nfct = &ct->ct_general; | |
907 | nskb->nfctinfo = ctinfo; | |
908 | nf_conntrack_get(nskb->nfct); | |
909 | } | |
910 | ||
9fb9cbb1 | 911 | /* Bring out ya dead! */ |
df0933dc | 912 | static struct nf_conn * |
9fb9cbb1 YK |
913 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
914 | void *data, unsigned int *bucket) | |
915 | { | |
df0933dc PM |
916 | struct nf_conntrack_tuple_hash *h; |
917 | struct nf_conn *ct; | |
f205c5e0 | 918 | struct hlist_node *n; |
9fb9cbb1 | 919 | |
f8ba1aff | 920 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 921 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
f205c5e0 | 922 | hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) { |
df0933dc PM |
923 | ct = nf_ct_tuplehash_to_ctrack(h); |
924 | if (iter(ct, data)) | |
925 | goto found; | |
926 | } | |
601e68e1 | 927 | } |
f205c5e0 | 928 | hlist_for_each_entry(h, n, &unconfirmed, hnode) { |
df0933dc PM |
929 | ct = nf_ct_tuplehash_to_ctrack(h); |
930 | if (iter(ct, data)) | |
ec68e97d | 931 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 932 | } |
f8ba1aff | 933 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
934 | return NULL; |
935 | found: | |
c073e3fa | 936 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 937 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 938 | return ct; |
9fb9cbb1 YK |
939 | } |
940 | ||
941 | void | |
942 | nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) | |
943 | { | |
df0933dc | 944 | struct nf_conn *ct; |
9fb9cbb1 YK |
945 | unsigned int bucket = 0; |
946 | ||
df0933dc | 947 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
948 | /* Time to push up daises... */ |
949 | if (del_timer(&ct->timeout)) | |
950 | death_by_timeout((unsigned long)ct); | |
951 | /* ... else the timer will get him soon. */ | |
952 | ||
953 | nf_ct_put(ct); | |
954 | } | |
955 | } | |
13b18339 | 956 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 YK |
957 | |
958 | static int kill_all(struct nf_conn *i, void *data) | |
959 | { | |
960 | return 1; | |
961 | } | |
962 | ||
96eb24d7 | 963 | void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
964 | { |
965 | if (vmalloced) | |
966 | vfree(hash); | |
967 | else | |
601e68e1 | 968 | free_pages((unsigned long)hash, |
f205c5e0 | 969 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 970 | } |
ac565e5f | 971 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 972 | |
272491ef | 973 | void nf_conntrack_flush(void) |
c1d10adb PNA |
974 | { |
975 | nf_ct_iterate_cleanup(kill_all, NULL); | |
976 | } | |
13b18339 | 977 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 978 | |
9fb9cbb1 YK |
979 | /* Mishearing the voices in his head, our hero wonders how he's |
980 | supposed to kill the mall. */ | |
981 | void nf_conntrack_cleanup(void) | |
982 | { | |
c3a47ab3 | 983 | rcu_assign_pointer(ip_ct_attach, NULL); |
7d3cdc6b | 984 | |
9fb9cbb1 YK |
985 | /* This makes sure all current packets have passed through |
986 | netfilter framework. Roll on, two-stage module | |
987 | delete... */ | |
988 | synchronize_net(); | |
989 | ||
990 | nf_ct_event_cache_flush(); | |
991 | i_see_dead_people: | |
c1d10adb | 992 | nf_conntrack_flush(); |
9fb9cbb1 YK |
993 | if (atomic_read(&nf_conntrack_count) != 0) { |
994 | schedule(); | |
995 | goto i_see_dead_people; | |
996 | } | |
6636568c PM |
997 | /* wait until all references to nf_conntrack_untracked are dropped */ |
998 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
999 | schedule(); | |
9fb9cbb1 | 1000 | |
de6e05c4 YK |
1001 | rcu_assign_pointer(nf_ct_destroy, NULL); |
1002 | ||
dacd2a1a | 1003 | kmem_cache_destroy(nf_conntrack_cachep); |
ac565e5f PM |
1004 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
1005 | nf_conntrack_htable_size); | |
5a6f294e | 1006 | |
ac5357eb | 1007 | nf_conntrack_proto_fini(); |
ceceae1b | 1008 | nf_conntrack_helper_fini(); |
e9c1b084 | 1009 | nf_conntrack_expect_fini(); |
9fb9cbb1 YK |
1010 | } |
1011 | ||
96eb24d7 | 1012 | struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced) |
9fb9cbb1 | 1013 | { |
f205c5e0 | 1014 | struct hlist_head *hash; |
8e5105a0 | 1015 | unsigned int size, i; |
9fb9cbb1 | 1016 | |
601e68e1 | 1017 | *vmalloced = 0; |
8e5105a0 | 1018 | |
f205c5e0 | 1019 | size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); |
29b67497 | 1020 | hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN, |
f205c5e0 | 1021 | get_order(sizeof(struct hlist_head) |
9fb9cbb1 | 1022 | * size)); |
601e68e1 | 1023 | if (!hash) { |
9fb9cbb1 YK |
1024 | *vmalloced = 1; |
1025 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
f205c5e0 | 1026 | hash = vmalloc(sizeof(struct hlist_head) * size); |
9fb9cbb1 YK |
1027 | } |
1028 | ||
1029 | if (hash) | |
601e68e1 | 1030 | for (i = 0; i < size; i++) |
f205c5e0 | 1031 | INIT_HLIST_HEAD(&hash[i]); |
9fb9cbb1 YK |
1032 | |
1033 | return hash; | |
1034 | } | |
ac565e5f | 1035 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1036 | |
fae718dd | 1037 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1038 | { |
96eb24d7 SH |
1039 | int i, bucket, vmalloced, old_vmalloced; |
1040 | unsigned int hashsize, old_size; | |
9fb9cbb1 | 1041 | int rnd; |
f205c5e0 | 1042 | struct hlist_head *hash, *old_hash; |
9fb9cbb1 YK |
1043 | struct nf_conntrack_tuple_hash *h; |
1044 | ||
1045 | /* On boot, we can set this without any fancy locking. */ | |
1046 | if (!nf_conntrack_htable_size) | |
1047 | return param_set_uint(val, kp); | |
1048 | ||
96eb24d7 | 1049 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1050 | if (!hashsize) |
1051 | return -EINVAL; | |
1052 | ||
ac565e5f | 1053 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); |
9fb9cbb1 YK |
1054 | if (!hash) |
1055 | return -ENOMEM; | |
1056 | ||
1057 | /* We have to rehahs for the new table anyway, so we also can | |
1058 | * use a newrandom seed */ | |
1059 | get_random_bytes(&rnd, 4); | |
1060 | ||
76507f69 PM |
1061 | /* Lookups in the old hash might happen in parallel, which means we |
1062 | * might get false negatives during connection lookup. New connections | |
1063 | * created because of a false negative won't make it into the hash | |
1064 | * though since that required taking the lock. | |
1065 | */ | |
f8ba1aff | 1066 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1067 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
f205c5e0 PM |
1068 | while (!hlist_empty(&nf_conntrack_hash[i])) { |
1069 | h = hlist_entry(nf_conntrack_hash[i].first, | |
1070 | struct nf_conntrack_tuple_hash, hnode); | |
76507f69 | 1071 | hlist_del_rcu(&h->hnode); |
9fb9cbb1 | 1072 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
f205c5e0 | 1073 | hlist_add_head(&h->hnode, &hash[bucket]); |
9fb9cbb1 YK |
1074 | } |
1075 | } | |
1076 | old_size = nf_conntrack_htable_size; | |
1077 | old_vmalloced = nf_conntrack_vmalloc; | |
1078 | old_hash = nf_conntrack_hash; | |
1079 | ||
1080 | nf_conntrack_htable_size = hashsize; | |
1081 | nf_conntrack_vmalloc = vmalloced; | |
1082 | nf_conntrack_hash = hash; | |
1083 | nf_conntrack_hash_rnd = rnd; | |
f8ba1aff | 1084 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1085 | |
ac565e5f | 1086 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1087 | return 0; |
1088 | } | |
fae718dd | 1089 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1090 | |
fae718dd | 1091 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1092 | &nf_conntrack_htable_size, 0600); |
1093 | ||
1094 | int __init nf_conntrack_init(void) | |
1095 | { | |
f205c5e0 | 1096 | int max_factor = 8; |
9fb9cbb1 YK |
1097 | int ret; |
1098 | ||
1099 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1100 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1101 | if (!nf_conntrack_htable_size) { |
1102 | nf_conntrack_htable_size | |
1103 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
f205c5e0 | 1104 | / sizeof(struct hlist_head)); |
9fb9cbb1 | 1105 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1106 | nf_conntrack_htable_size = 16384; |
1107 | if (nf_conntrack_htable_size < 32) | |
1108 | nf_conntrack_htable_size = 32; | |
1109 | ||
1110 | /* Use a max. factor of four by default to get the same max as | |
1111 | * with the old struct list_heads. When a table size is given | |
1112 | * we use the old value of 8 to avoid reducing the max. | |
1113 | * entries. */ | |
1114 | max_factor = 4; | |
9fb9cbb1 | 1115 | } |
ac565e5f PM |
1116 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
1117 | &nf_conntrack_vmalloc); | |
9fb9cbb1 YK |
1118 | if (!nf_conntrack_hash) { |
1119 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1120 | goto err_out; | |
1121 | } | |
1122 | ||
f205c5e0 | 1123 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1124 | |
1125 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1126 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1127 | nf_conntrack_max); | |
1128 | ||
dacd2a1a YK |
1129 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1130 | sizeof(struct nf_conn), | |
20c2df83 | 1131 | 0, 0, NULL); |
dacd2a1a | 1132 | if (!nf_conntrack_cachep) { |
9fb9cbb1 YK |
1133 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
1134 | goto err_free_hash; | |
1135 | } | |
1136 | ||
e9c1b084 PM |
1137 | ret = nf_conntrack_proto_init(); |
1138 | if (ret < 0) | |
9fb9cbb1 | 1139 | goto err_free_conntrack_slab; |
9fb9cbb1 | 1140 | |
e9c1b084 | 1141 | ret = nf_conntrack_expect_init(); |
933a41e7 | 1142 | if (ret < 0) |
e9c1b084 | 1143 | goto out_fini_proto; |
933a41e7 | 1144 | |
ceceae1b YK |
1145 | ret = nf_conntrack_helper_init(); |
1146 | if (ret < 0) | |
e9c1b084 | 1147 | goto out_fini_expect; |
ceceae1b | 1148 | |
7d3cdc6b | 1149 | /* For use by REJECT target */ |
b334aadc | 1150 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); |
de6e05c4 | 1151 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); |
7d3cdc6b | 1152 | |
9fb9cbb1 YK |
1153 | /* Set up fake conntrack: |
1154 | - to never be deleted, not in any hashes */ | |
1155 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); | |
1156 | /* - and look it like as a confirmed connection */ | |
1157 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1158 | ||
1159 | return ret; | |
1160 | ||
e9c1b084 PM |
1161 | out_fini_expect: |
1162 | nf_conntrack_expect_fini(); | |
ceceae1b YK |
1163 | out_fini_proto: |
1164 | nf_conntrack_proto_fini(); | |
9fb9cbb1 | 1165 | err_free_conntrack_slab: |
dacd2a1a | 1166 | kmem_cache_destroy(nf_conntrack_cachep); |
9fb9cbb1 | 1167 | err_free_hash: |
ac565e5f PM |
1168 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
1169 | nf_conntrack_htable_size); | |
9fb9cbb1 YK |
1170 | err_out: |
1171 | return -ENOMEM; | |
1172 | } |