]>
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 | ||
9fb9cbb1 YK |
97 | int |
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 YK |
106 | { |
107 | NF_CT_TUPLE_U_BLANK(tuple); | |
108 | ||
109 | tuple->src.l3num = l3num; | |
110 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
111 | return 0; | |
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 | |
e2a3123f YK |
120 | int nf_ct_get_tuplepr(const struct sk_buff *skb, |
121 | unsigned int nhoff, | |
122 | u_int16_t l3num, | |
123 | struct nf_conntrack_tuple *tuple) | |
124 | { | |
125 | struct nf_conntrack_l3proto *l3proto; | |
126 | struct nf_conntrack_l4proto *l4proto; | |
127 | unsigned int protoff; | |
128 | u_int8_t protonum; | |
129 | int ret; | |
130 | ||
131 | rcu_read_lock(); | |
132 | ||
133 | l3proto = __nf_ct_l3proto_find(l3num); | |
134 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
135 | if (ret != NF_ACCEPT) { | |
136 | rcu_read_unlock(); | |
137 | return 0; | |
138 | } | |
139 | ||
140 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
141 | ||
142 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
143 | l3proto, l4proto); | |
144 | ||
145 | rcu_read_unlock(); | |
146 | return ret; | |
147 | } | |
148 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
149 | ||
9fb9cbb1 YK |
150 | int |
151 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, | |
152 | const struct nf_conntrack_tuple *orig, | |
153 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 154 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 YK |
155 | { |
156 | NF_CT_TUPLE_U_BLANK(inverse); | |
157 | ||
158 | inverse->src.l3num = orig->src.l3num; | |
159 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
160 | return 0; | |
161 | ||
162 | inverse->dst.dir = !orig->dst.dir; | |
163 | ||
164 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 165 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 166 | } |
13b18339 | 167 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 168 | |
9fb9cbb1 YK |
169 | static void |
170 | clean_from_lists(struct nf_conn *ct) | |
171 | { | |
0d53778e | 172 | pr_debug("clean_from_lists(%p)\n", ct); |
76507f69 PM |
173 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
174 | hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); | |
9fb9cbb1 YK |
175 | |
176 | /* Destroy all pending expectations */ | |
c1d10adb | 177 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
178 | } |
179 | ||
180 | static void | |
181 | destroy_conntrack(struct nf_conntrack *nfct) | |
182 | { | |
183 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 184 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 185 | |
0d53778e | 186 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
187 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
188 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
189 | ||
190 | nf_conntrack_event(IPCT_DESTROY, ct); | |
191 | set_bit(IPS_DYING_BIT, &ct->status); | |
192 | ||
193 | /* To make sure we don't get any weird locking issues here: | |
194 | * destroy_conntrack() MUST NOT be called with a write lock | |
195 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 196 | rcu_read_lock(); |
923f4902 PM |
197 | l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, |
198 | ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); | |
605dcad6 MJ |
199 | if (l4proto && l4proto->destroy) |
200 | l4proto->destroy(ct); | |
9fb9cbb1 | 201 | |
ecfab2c9 YK |
202 | nf_ct_ext_destroy(ct); |
203 | ||
982d9a9c | 204 | rcu_read_unlock(); |
9fb9cbb1 | 205 | |
f8ba1aff | 206 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
207 | /* Expectations will have been removed in clean_from_lists, |
208 | * except TFTP can create an expectation on the first packet, | |
209 | * before connection is in the list, so we need to clean here, | |
210 | * too. */ | |
c1d10adb | 211 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
212 | |
213 | /* We overload first tuple to link into unconfirmed list. */ | |
214 | if (!nf_ct_is_confirmed(ct)) { | |
f205c5e0 PM |
215 | BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode)); |
216 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); | |
9fb9cbb1 YK |
217 | } |
218 | ||
219 | NF_CT_STAT_INC(delete); | |
f8ba1aff | 220 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
221 | |
222 | if (ct->master) | |
223 | nf_ct_put(ct->master); | |
224 | ||
0d53778e | 225 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
226 | nf_conntrack_free(ct); |
227 | } | |
228 | ||
229 | static void death_by_timeout(unsigned long ul_conntrack) | |
230 | { | |
231 | struct nf_conn *ct = (void *)ul_conntrack; | |
5397e97d | 232 | struct nf_conn_help *help = nfct_help(ct); |
3c158f7f | 233 | struct nf_conntrack_helper *helper; |
5397e97d | 234 | |
3c158f7f PM |
235 | if (help) { |
236 | rcu_read_lock(); | |
237 | helper = rcu_dereference(help->helper); | |
238 | if (helper && helper->destroy) | |
239 | helper->destroy(ct); | |
240 | rcu_read_unlock(); | |
241 | } | |
9fb9cbb1 | 242 | |
f8ba1aff | 243 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
244 | /* Inside lock so preempt is disabled on module removal path. |
245 | * Otherwise we can get spurious warnings. */ | |
246 | NF_CT_STAT_INC(delete_list); | |
247 | clean_from_lists(ct); | |
f8ba1aff | 248 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
249 | nf_ct_put(ct); |
250 | } | |
251 | ||
c1d10adb | 252 | struct nf_conntrack_tuple_hash * |
ba419aff | 253 | __nf_conntrack_find(const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
254 | { |
255 | struct nf_conntrack_tuple_hash *h; | |
f205c5e0 | 256 | struct hlist_node *n; |
9fb9cbb1 YK |
257 | unsigned int hash = hash_conntrack(tuple); |
258 | ||
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 YK |
261 | NF_CT_STAT_INC(found); |
262 | return h; | |
263 | } | |
264 | NF_CT_STAT_INC(searched); | |
265 | } | |
266 | ||
267 | return NULL; | |
268 | } | |
13b18339 | 269 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
270 | |
271 | /* Find a connection corresponding to a tuple. */ | |
272 | struct nf_conntrack_tuple_hash * | |
330f7db5 | 273 | nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 YK |
274 | { |
275 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 276 | struct nf_conn *ct; |
9fb9cbb1 | 277 | |
76507f69 | 278 | rcu_read_lock(); |
ba419aff | 279 | h = __nf_conntrack_find(tuple); |
76507f69 PM |
280 | if (h) { |
281 | ct = nf_ct_tuplehash_to_ctrack(h); | |
282 | if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
283 | h = NULL; | |
284 | } | |
285 | rcu_read_unlock(); | |
9fb9cbb1 YK |
286 | |
287 | return h; | |
288 | } | |
13b18339 | 289 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 290 | |
c1d10adb PNA |
291 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
292 | unsigned int hash, | |
601e68e1 | 293 | unsigned int repl_hash) |
c1d10adb | 294 | { |
76507f69 PM |
295 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
296 | &nf_conntrack_hash[hash]); | |
297 | hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, | |
298 | &nf_conntrack_hash[repl_hash]); | |
c1d10adb PNA |
299 | } |
300 | ||
301 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
302 | { | |
303 | unsigned int hash, repl_hash; | |
304 | ||
305 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
306 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
307 | ||
f8ba1aff | 308 | spin_lock_bh(&nf_conntrack_lock); |
c1d10adb | 309 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
f8ba1aff | 310 | spin_unlock_bh(&nf_conntrack_lock); |
c1d10adb | 311 | } |
13b18339 | 312 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 313 | |
9fb9cbb1 YK |
314 | /* Confirm a connection given skb; places it in hash table */ |
315 | int | |
3db05fea | 316 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
317 | { |
318 | unsigned int hash, repl_hash; | |
df0933dc | 319 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 320 | struct nf_conn *ct; |
df0933dc | 321 | struct nf_conn_help *help; |
f205c5e0 | 322 | struct hlist_node *n; |
9fb9cbb1 YK |
323 | enum ip_conntrack_info ctinfo; |
324 | ||
3db05fea | 325 | ct = nf_ct_get(skb, &ctinfo); |
9fb9cbb1 YK |
326 | |
327 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
328 | ICMP/TCP RST packets in other direction. Actual packet | |
329 | which created connection will be IP_CT_NEW or for an | |
330 | expected connection, IP_CT_RELATED. */ | |
331 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
332 | return NF_ACCEPT; | |
333 | ||
334 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
335 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
336 | ||
337 | /* We're not in hash table, and we refuse to set up related | |
338 | connections for unconfirmed conns. But packet copies and | |
339 | REJECT will give spurious warnings here. */ | |
340 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
341 | ||
342 | /* No external references means noone else could have | |
343 | confirmed us. */ | |
344 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 345 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 346 | |
f8ba1aff | 347 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
348 | |
349 | /* See if there's one in the list already, including reverse: | |
350 | NAT could have grabbed it without realizing, since we're | |
351 | not in the hash. If there is, we lost race. */ | |
f205c5e0 | 352 | hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) |
df0933dc PM |
353 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
354 | &h->tuple)) | |
355 | goto out; | |
f205c5e0 | 356 | hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode) |
df0933dc PM |
357 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
358 | &h->tuple)) | |
359 | goto out; | |
9fb9cbb1 | 360 | |
df0933dc | 361 | /* Remove from unconfirmed list */ |
f205c5e0 | 362 | hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); |
df0933dc PM |
363 | |
364 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
365 | /* Timer relative to confirmation time, not original | |
366 | setting time, otherwise we'd get timer wrap in | |
367 | weird delay cases. */ | |
368 | ct->timeout.expires += jiffies; | |
369 | add_timer(&ct->timeout); | |
370 | atomic_inc(&ct->ct_general.use); | |
371 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
372 | NF_CT_STAT_INC(insert); | |
f8ba1aff | 373 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
374 | help = nfct_help(ct); |
375 | if (help && help->helper) | |
3db05fea | 376 | nf_conntrack_event_cache(IPCT_HELPER, skb); |
9fb9cbb1 | 377 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
378 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
379 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
3db05fea | 380 | nf_conntrack_event_cache(IPCT_NATINFO, skb); |
9fb9cbb1 | 381 | #endif |
df0933dc | 382 | nf_conntrack_event_cache(master_ct(ct) ? |
3db05fea | 383 | IPCT_RELATED : IPCT_NEW, skb); |
df0933dc | 384 | return NF_ACCEPT; |
9fb9cbb1 | 385 | |
df0933dc | 386 | out: |
9fb9cbb1 | 387 | NF_CT_STAT_INC(insert_failed); |
f8ba1aff | 388 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
389 | return NF_DROP; |
390 | } | |
13b18339 | 391 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
392 | |
393 | /* Returns true if a connection correspondings to the tuple (required | |
394 | for NAT). */ | |
395 | int | |
396 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
397 | const struct nf_conn *ignored_conntrack) | |
398 | { | |
399 | struct nf_conntrack_tuple_hash *h; | |
ba419aff PM |
400 | struct hlist_node *n; |
401 | unsigned int hash = hash_conntrack(tuple); | |
9fb9cbb1 | 402 | |
76507f69 | 403 | rcu_read_lock(); |
ba419aff PM |
404 | hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) { |
405 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && | |
406 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
407 | NF_CT_STAT_INC(found); | |
408 | rcu_read_unlock(); | |
409 | return 1; | |
410 | } | |
411 | NF_CT_STAT_INC(searched); | |
412 | } | |
76507f69 | 413 | rcu_read_unlock(); |
9fb9cbb1 | 414 | |
ba419aff | 415 | return 0; |
9fb9cbb1 | 416 | } |
13b18339 | 417 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 418 | |
7ae7730f PM |
419 | #define NF_CT_EVICTION_RANGE 8 |
420 | ||
9fb9cbb1 YK |
421 | /* There's a small race here where we may free a just-assured |
422 | connection. Too bad: we're in trouble anyway. */ | |
76eb9460 | 423 | static noinline int early_drop(unsigned int hash) |
9fb9cbb1 | 424 | { |
f205c5e0 | 425 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 426 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 427 | struct nf_conn *ct = NULL, *tmp; |
f205c5e0 | 428 | struct hlist_node *n; |
7ae7730f | 429 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
430 | int dropped = 0; |
431 | ||
76507f69 | 432 | rcu_read_lock(); |
7ae7730f | 433 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
76507f69 PM |
434 | hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], |
435 | hnode) { | |
7ae7730f PM |
436 | tmp = nf_ct_tuplehash_to_ctrack(h); |
437 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
438 | ct = tmp; | |
439 | cnt++; | |
440 | } | |
76507f69 PM |
441 | |
442 | if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) | |
443 | ct = NULL; | |
7ae7730f PM |
444 | if (ct || cnt >= NF_CT_EVICTION_RANGE) |
445 | break; | |
446 | hash = (hash + 1) % nf_conntrack_htable_size; | |
9fb9cbb1 | 447 | } |
76507f69 | 448 | rcu_read_unlock(); |
9fb9cbb1 YK |
449 | |
450 | if (!ct) | |
451 | return dropped; | |
452 | ||
453 | if (del_timer(&ct->timeout)) { | |
454 | death_by_timeout((unsigned long)ct); | |
455 | dropped = 1; | |
c0e912d7 | 456 | NF_CT_STAT_INC_ATOMIC(early_drop); |
9fb9cbb1 YK |
457 | } |
458 | nf_ct_put(ct); | |
459 | return dropped; | |
460 | } | |
461 | ||
dacd2a1a YK |
462 | struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, |
463 | const struct nf_conntrack_tuple *repl) | |
9fb9cbb1 YK |
464 | { |
465 | struct nf_conn *conntrack = NULL; | |
9fb9cbb1 | 466 | |
dc808fe2 | 467 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
468 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
469 | nf_conntrack_hash_rnd_initted = 1; | |
470 | } | |
471 | ||
5251e2d2 PNA |
472 | /* We don't want any race condition at early drop stage */ |
473 | atomic_inc(&nf_conntrack_count); | |
474 | ||
76eb9460 PM |
475 | if (nf_conntrack_max && |
476 | unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) { | |
9fb9cbb1 | 477 | unsigned int hash = hash_conntrack(orig); |
7ae7730f | 478 | if (!early_drop(hash)) { |
5251e2d2 | 479 | atomic_dec(&nf_conntrack_count); |
9fb9cbb1 YK |
480 | if (net_ratelimit()) |
481 | printk(KERN_WARNING | |
482 | "nf_conntrack: table full, dropping" | |
483 | " packet.\n"); | |
484 | return ERR_PTR(-ENOMEM); | |
485 | } | |
486 | } | |
487 | ||
dacd2a1a | 488 | conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC); |
9fb9cbb1 | 489 | if (conntrack == NULL) { |
0d53778e | 490 | pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); |
dacd2a1a YK |
491 | atomic_dec(&nf_conntrack_count); |
492 | return ERR_PTR(-ENOMEM); | |
9fb9cbb1 YK |
493 | } |
494 | ||
9fb9cbb1 | 495 | atomic_set(&conntrack->ct_general.use, 1); |
9fb9cbb1 YK |
496 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
497 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
498 | /* Don't set timer yet: wait for confirmation */ | |
e6f689db PM |
499 | setup_timer(&conntrack->timeout, death_by_timeout, |
500 | (unsigned long)conntrack); | |
76507f69 | 501 | INIT_RCU_HEAD(&conntrack->rcu); |
9fb9cbb1 | 502 | |
9fb9cbb1 YK |
503 | return conntrack; |
504 | } | |
13b18339 | 505 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 506 | |
76507f69 | 507 | static void nf_conntrack_free_rcu(struct rcu_head *head) |
9fb9cbb1 | 508 | { |
76507f69 PM |
509 | struct nf_conn *ct = container_of(head, struct nf_conn, rcu); |
510 | ||
511 | nf_ct_ext_free(ct); | |
512 | kmem_cache_free(nf_conntrack_cachep, ct); | |
9fb9cbb1 YK |
513 | atomic_dec(&nf_conntrack_count); |
514 | } | |
76507f69 PM |
515 | |
516 | void nf_conntrack_free(struct nf_conn *conntrack) | |
517 | { | |
518 | call_rcu(&conntrack->rcu, nf_conntrack_free_rcu); | |
519 | } | |
13b18339 | 520 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
521 | |
522 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
523 | failed due to stress. Otherwise it really is unclassifiable. */ | |
524 | static struct nf_conntrack_tuple_hash * | |
525 | init_conntrack(const struct nf_conntrack_tuple *tuple, | |
526 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 527 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
528 | struct sk_buff *skb, |
529 | unsigned int dataoff) | |
530 | { | |
531 | struct nf_conn *conntrack; | |
3c158f7f | 532 | struct nf_conn_help *help; |
9fb9cbb1 YK |
533 | struct nf_conntrack_tuple repl_tuple; |
534 | struct nf_conntrack_expect *exp; | |
535 | ||
605dcad6 | 536 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 537 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
538 | return NULL; |
539 | } | |
540 | ||
dacd2a1a | 541 | conntrack = nf_conntrack_alloc(tuple, &repl_tuple); |
9fb9cbb1 | 542 | if (conntrack == NULL || IS_ERR(conntrack)) { |
0d53778e | 543 | pr_debug("Can't allocate conntrack.\n"); |
9fb9cbb1 YK |
544 | return (struct nf_conntrack_tuple_hash *)conntrack; |
545 | } | |
546 | ||
605dcad6 | 547 | if (!l4proto->new(conntrack, skb, dataoff)) { |
9fb9cbb1 | 548 | nf_conntrack_free(conntrack); |
0d53778e | 549 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
550 | return NULL; |
551 | } | |
552 | ||
f8ba1aff | 553 | spin_lock_bh(&nf_conntrack_lock); |
6823645d | 554 | exp = nf_ct_find_expectation(tuple); |
9fb9cbb1 | 555 | if (exp) { |
0d53778e PM |
556 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
557 | conntrack, exp); | |
9fb9cbb1 YK |
558 | /* Welcome, Mr. Bond. We've been expecting you... */ |
559 | __set_bit(IPS_EXPECTED_BIT, &conntrack->status); | |
560 | conntrack->master = exp->master; | |
ceceae1b | 561 | if (exp->helper) { |
b560580a | 562 | help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC); |
ceceae1b YK |
563 | if (help) |
564 | rcu_assign_pointer(help->helper, exp->helper); | |
ceceae1b YK |
565 | } |
566 | ||
9fb9cbb1 YK |
567 | #ifdef CONFIG_NF_CONNTRACK_MARK |
568 | conntrack->mark = exp->master->mark; | |
7c9728c3 JM |
569 | #endif |
570 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
571 | conntrack->secmark = exp->master->secmark; | |
9fb9cbb1 YK |
572 | #endif |
573 | nf_conntrack_get(&conntrack->master->ct_general); | |
574 | NF_CT_STAT_INC(expect_new); | |
22e7410b | 575 | } else { |
ceceae1b YK |
576 | struct nf_conntrack_helper *helper; |
577 | ||
578 | helper = __nf_ct_helper_find(&repl_tuple); | |
579 | if (helper) { | |
b560580a | 580 | help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC); |
ceceae1b | 581 | if (help) |
ceceae1b | 582 | rcu_assign_pointer(help->helper, helper); |
3c158f7f | 583 | } |
9fb9cbb1 | 584 | NF_CT_STAT_INC(new); |
22e7410b | 585 | } |
9fb9cbb1 YK |
586 | |
587 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
f205c5e0 PM |
588 | hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode, |
589 | &unconfirmed); | |
9fb9cbb1 | 590 | |
f8ba1aff | 591 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
592 | |
593 | if (exp) { | |
594 | if (exp->expectfn) | |
595 | exp->expectfn(conntrack, exp); | |
6823645d | 596 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
597 | } |
598 | ||
599 | return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; | |
600 | } | |
601 | ||
602 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
603 | static inline struct nf_conn * | |
604 | resolve_normal_ct(struct sk_buff *skb, | |
605 | unsigned int dataoff, | |
606 | u_int16_t l3num, | |
607 | u_int8_t protonum, | |
608 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 609 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
610 | int *set_reply, |
611 | enum ip_conntrack_info *ctinfo) | |
612 | { | |
613 | struct nf_conntrack_tuple tuple; | |
614 | struct nf_conntrack_tuple_hash *h; | |
615 | struct nf_conn *ct; | |
616 | ||
bbe735e4 | 617 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 618 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 619 | l4proto)) { |
0d53778e | 620 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
621 | return NULL; |
622 | } | |
623 | ||
624 | /* look for tuple match */ | |
330f7db5 | 625 | h = nf_conntrack_find_get(&tuple); |
9fb9cbb1 | 626 | if (!h) { |
605dcad6 | 627 | h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
628 | if (!h) |
629 | return NULL; | |
630 | if (IS_ERR(h)) | |
631 | return (void *)h; | |
632 | } | |
633 | ct = nf_ct_tuplehash_to_ctrack(h); | |
634 | ||
635 | /* It exists; we have (non-exclusive) reference. */ | |
636 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
637 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
638 | /* Please set reply bit if this packet OK */ | |
639 | *set_reply = 1; | |
640 | } else { | |
641 | /* Once we've had two way comms, always ESTABLISHED. */ | |
642 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 643 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
644 | *ctinfo = IP_CT_ESTABLISHED; |
645 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
646 | pr_debug("nf_conntrack_in: related packet for %p\n", |
647 | ct); | |
9fb9cbb1 YK |
648 | *ctinfo = IP_CT_RELATED; |
649 | } else { | |
0d53778e | 650 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
651 | *ctinfo = IP_CT_NEW; |
652 | } | |
653 | *set_reply = 0; | |
654 | } | |
655 | skb->nfct = &ct->ct_general; | |
656 | skb->nfctinfo = *ctinfo; | |
657 | return ct; | |
658 | } | |
659 | ||
660 | unsigned int | |
3db05fea | 661 | nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb) |
9fb9cbb1 YK |
662 | { |
663 | struct nf_conn *ct; | |
664 | enum ip_conntrack_info ctinfo; | |
665 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 666 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
667 | unsigned int dataoff; |
668 | u_int8_t protonum; | |
669 | int set_reply = 0; | |
670 | int ret; | |
671 | ||
672 | /* Previously seen (loopback or untracked)? Ignore. */ | |
3db05fea | 673 | if (skb->nfct) { |
c0e912d7 | 674 | NF_CT_STAT_INC_ATOMIC(ignore); |
9fb9cbb1 YK |
675 | return NF_ACCEPT; |
676 | } | |
677 | ||
923f4902 | 678 | /* rcu_read_lock()ed by nf_hook_slow */ |
c1d10adb | 679 | l3proto = __nf_ct_l3proto_find((u_int16_t)pf); |
3db05fea | 680 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
681 | &dataoff, &protonum); |
682 | if (ret <= 0) { | |
0d53778e | 683 | pr_debug("not prepared to track yet or error occured\n"); |
d87d8469 YK |
684 | NF_CT_STAT_INC_ATOMIC(error); |
685 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
686 | return -ret; |
687 | } | |
688 | ||
605dcad6 | 689 | l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); |
9fb9cbb1 YK |
690 | |
691 | /* It may be an special packet, error, unclean... | |
692 | * inverse of the return code tells to the netfilter | |
693 | * core what to do with the packet. */ | |
605dcad6 | 694 | if (l4proto->error != NULL && |
3db05fea | 695 | (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) { |
c0e912d7 PM |
696 | NF_CT_STAT_INC_ATOMIC(error); |
697 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
698 | return -ret; |
699 | } | |
700 | ||
3db05fea | 701 | ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto, |
9fb9cbb1 YK |
702 | &set_reply, &ctinfo); |
703 | if (!ct) { | |
704 | /* Not valid part of a connection */ | |
c0e912d7 | 705 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
706 | return NF_ACCEPT; |
707 | } | |
708 | ||
709 | if (IS_ERR(ct)) { | |
710 | /* Too stressed to deal. */ | |
c0e912d7 | 711 | NF_CT_STAT_INC_ATOMIC(drop); |
9fb9cbb1 YK |
712 | return NF_DROP; |
713 | } | |
714 | ||
3db05fea | 715 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 716 | |
3db05fea | 717 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
718 | if (ret < 0) { |
719 | /* Invalid: inverse of the return code tells | |
720 | * the netfilter core what to do */ | |
0d53778e | 721 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
722 | nf_conntrack_put(skb->nfct); |
723 | skb->nfct = NULL; | |
c0e912d7 | 724 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
725 | return -ret; |
726 | } | |
727 | ||
728 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
3db05fea | 729 | nf_conntrack_event_cache(IPCT_STATUS, skb); |
9fb9cbb1 YK |
730 | |
731 | return ret; | |
732 | } | |
13b18339 | 733 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 YK |
734 | |
735 | int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, | |
736 | const struct nf_conntrack_tuple *orig) | |
737 | { | |
923f4902 PM |
738 | int ret; |
739 | ||
740 | rcu_read_lock(); | |
741 | ret = nf_ct_invert_tuple(inverse, orig, | |
742 | __nf_ct_l3proto_find(orig->src.l3num), | |
743 | __nf_ct_l4proto_find(orig->src.l3num, | |
744 | orig->dst.protonum)); | |
745 | rcu_read_unlock(); | |
746 | return ret; | |
9fb9cbb1 | 747 | } |
13b18339 | 748 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 749 | |
5b1158e9 JK |
750 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
751 | implicitly racy: see __nf_conntrack_confirm */ | |
752 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
753 | const struct nf_conntrack_tuple *newreply) | |
754 | { | |
755 | struct nf_conn_help *help = nfct_help(ct); | |
ceceae1b | 756 | struct nf_conntrack_helper *helper; |
5b1158e9 | 757 | |
5b1158e9 JK |
758 | /* Should be unconfirmed, so not in hash table yet */ |
759 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
760 | ||
0d53778e | 761 | pr_debug("Altering reply tuple of %p to ", ct); |
5b1158e9 JK |
762 | NF_CT_DUMP_TUPLE(newreply); |
763 | ||
764 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ceceae1b | 765 | if (ct->master || (help && help->expecting != 0)) |
c52fbb41 | 766 | return; |
ceceae1b | 767 | |
c52fbb41 | 768 | rcu_read_lock(); |
ceceae1b YK |
769 | helper = __nf_ct_helper_find(newreply); |
770 | if (helper == NULL) { | |
771 | if (help) | |
772 | rcu_assign_pointer(help->helper, NULL); | |
773 | goto out; | |
5d78a849 | 774 | } |
ceceae1b YK |
775 | |
776 | if (help == NULL) { | |
b560580a PM |
777 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
778 | if (help == NULL) | |
ceceae1b | 779 | goto out; |
ceceae1b YK |
780 | } else { |
781 | memset(&help->help, 0, sizeof(help->help)); | |
782 | } | |
783 | ||
784 | rcu_assign_pointer(help->helper, helper); | |
785 | out: | |
c52fbb41 | 786 | rcu_read_unlock(); |
5b1158e9 | 787 | } |
13b18339 | 788 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 789 | |
9fb9cbb1 YK |
790 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
791 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
792 | enum ip_conntrack_info ctinfo, | |
793 | const struct sk_buff *skb, | |
794 | unsigned long extra_jiffies, | |
795 | int do_acct) | |
796 | { | |
797 | int event = 0; | |
798 | ||
799 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
800 | NF_CT_ASSERT(skb); | |
801 | ||
f8ba1aff | 802 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 803 | |
997ae831 | 804 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
805 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
806 | goto acct; | |
997ae831 | 807 | |
9fb9cbb1 YK |
808 | /* If not in hash table, timer will not be active yet */ |
809 | if (!nf_ct_is_confirmed(ct)) { | |
810 | ct->timeout.expires = extra_jiffies; | |
811 | event = IPCT_REFRESH; | |
812 | } else { | |
be00c8e4 MJ |
813 | unsigned long newtime = jiffies + extra_jiffies; |
814 | ||
815 | /* Only update the timeout if the new timeout is at least | |
816 | HZ jiffies from the old timeout. Need del_timer for race | |
817 | avoidance (may already be dying). */ | |
818 | if (newtime - ct->timeout.expires >= HZ | |
819 | && del_timer(&ct->timeout)) { | |
820 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
821 | add_timer(&ct->timeout); |
822 | event = IPCT_REFRESH; | |
823 | } | |
824 | } | |
825 | ||
47d95045 | 826 | acct: |
9fb9cbb1 YK |
827 | #ifdef CONFIG_NF_CT_ACCT |
828 | if (do_acct) { | |
829 | ct->counters[CTINFO2DIR(ctinfo)].packets++; | |
830 | ct->counters[CTINFO2DIR(ctinfo)].bytes += | |
bbe735e4 | 831 | skb->len - skb_network_offset(skb); |
3ffd5eeb MJ |
832 | |
833 | if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) | |
834 | || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) | |
835 | event |= IPCT_COUNTER_FILLING; | |
9fb9cbb1 YK |
836 | } |
837 | #endif | |
838 | ||
f8ba1aff | 839 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
840 | |
841 | /* must be unlocked when calling event cache */ | |
842 | if (event) | |
843 | nf_conntrack_event_cache(event, skb); | |
844 | } | |
13b18339 | 845 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 846 | |
e281db5c | 847 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
848 | |
849 | #include <linux/netfilter/nfnetlink.h> | |
850 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
851 | #include <linux/mutex.h> |
852 | ||
c1d10adb PNA |
853 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
854 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
855 | * or depend on ctnetlink */ | |
fdf70832 | 856 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
857 | const struct nf_conntrack_tuple *tuple) |
858 | { | |
77236b6e PM |
859 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
860 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
861 | return 0; |
862 | ||
df6fb868 | 863 | nla_put_failure: |
c1d10adb PNA |
864 | return -1; |
865 | } | |
fdf70832 | 866 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 867 | |
f73e924c PM |
868 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
869 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
870 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 871 | }; |
f73e924c | 872 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 873 | |
fdf70832 | 874 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
875 | struct nf_conntrack_tuple *t) |
876 | { | |
df6fb868 | 877 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
878 | return -EINVAL; |
879 | ||
77236b6e PM |
880 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
881 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
882 | |
883 | return 0; | |
884 | } | |
fdf70832 | 885 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
c1d10adb PNA |
886 | #endif |
887 | ||
9fb9cbb1 | 888 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 889 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
890 | { |
891 | struct nf_conn *ct; | |
892 | enum ip_conntrack_info ctinfo; | |
893 | ||
894 | /* This ICMP is in reverse direction to the packet which caused it */ | |
895 | ct = nf_ct_get(skb, &ctinfo); | |
896 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
897 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
898 | else | |
899 | ctinfo = IP_CT_RELATED; | |
900 | ||
901 | /* Attach to new skbuff, and increment count */ | |
902 | nskb->nfct = &ct->ct_general; | |
903 | nskb->nfctinfo = ctinfo; | |
904 | nf_conntrack_get(nskb->nfct); | |
905 | } | |
906 | ||
907 | static inline int | |
908 | do_iter(const struct nf_conntrack_tuple_hash *i, | |
909 | int (*iter)(struct nf_conn *i, void *data), | |
910 | void *data) | |
911 | { | |
912 | return iter(nf_ct_tuplehash_to_ctrack(i), data); | |
913 | } | |
914 | ||
915 | /* Bring out ya dead! */ | |
df0933dc | 916 | static struct nf_conn * |
9fb9cbb1 YK |
917 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
918 | void *data, unsigned int *bucket) | |
919 | { | |
df0933dc PM |
920 | struct nf_conntrack_tuple_hash *h; |
921 | struct nf_conn *ct; | |
f205c5e0 | 922 | struct hlist_node *n; |
9fb9cbb1 | 923 | |
f8ba1aff | 924 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 925 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
f205c5e0 | 926 | hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) { |
df0933dc PM |
927 | ct = nf_ct_tuplehash_to_ctrack(h); |
928 | if (iter(ct, data)) | |
929 | goto found; | |
930 | } | |
601e68e1 | 931 | } |
f205c5e0 | 932 | hlist_for_each_entry(h, n, &unconfirmed, hnode) { |
df0933dc PM |
933 | ct = nf_ct_tuplehash_to_ctrack(h); |
934 | if (iter(ct, data)) | |
ec68e97d | 935 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 936 | } |
f8ba1aff | 937 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
938 | return NULL; |
939 | found: | |
c073e3fa | 940 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 941 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 942 | return ct; |
9fb9cbb1 YK |
943 | } |
944 | ||
945 | void | |
946 | nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) | |
947 | { | |
df0933dc | 948 | struct nf_conn *ct; |
9fb9cbb1 YK |
949 | unsigned int bucket = 0; |
950 | ||
df0933dc | 951 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
952 | /* Time to push up daises... */ |
953 | if (del_timer(&ct->timeout)) | |
954 | death_by_timeout((unsigned long)ct); | |
955 | /* ... else the timer will get him soon. */ | |
956 | ||
957 | nf_ct_put(ct); | |
958 | } | |
959 | } | |
13b18339 | 960 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 YK |
961 | |
962 | static int kill_all(struct nf_conn *i, void *data) | |
963 | { | |
964 | return 1; | |
965 | } | |
966 | ||
96eb24d7 | 967 | void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size) |
9fb9cbb1 YK |
968 | { |
969 | if (vmalloced) | |
970 | vfree(hash); | |
971 | else | |
601e68e1 | 972 | free_pages((unsigned long)hash, |
f205c5e0 | 973 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 974 | } |
ac565e5f | 975 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 976 | |
272491ef | 977 | void nf_conntrack_flush(void) |
c1d10adb PNA |
978 | { |
979 | nf_ct_iterate_cleanup(kill_all, NULL); | |
980 | } | |
13b18339 | 981 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 982 | |
9fb9cbb1 YK |
983 | /* Mishearing the voices in his head, our hero wonders how he's |
984 | supposed to kill the mall. */ | |
985 | void nf_conntrack_cleanup(void) | |
986 | { | |
c3a47ab3 | 987 | rcu_assign_pointer(ip_ct_attach, NULL); |
7d3cdc6b | 988 | |
9fb9cbb1 YK |
989 | /* This makes sure all current packets have passed through |
990 | netfilter framework. Roll on, two-stage module | |
991 | delete... */ | |
992 | synchronize_net(); | |
993 | ||
994 | nf_ct_event_cache_flush(); | |
995 | i_see_dead_people: | |
c1d10adb | 996 | nf_conntrack_flush(); |
9fb9cbb1 YK |
997 | if (atomic_read(&nf_conntrack_count) != 0) { |
998 | schedule(); | |
999 | goto i_see_dead_people; | |
1000 | } | |
6636568c PM |
1001 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1002 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1003 | schedule(); | |
9fb9cbb1 | 1004 | |
de6e05c4 YK |
1005 | rcu_assign_pointer(nf_ct_destroy, NULL); |
1006 | ||
dacd2a1a | 1007 | kmem_cache_destroy(nf_conntrack_cachep); |
ac565e5f PM |
1008 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
1009 | nf_conntrack_htable_size); | |
5a6f294e | 1010 | |
ac5357eb | 1011 | nf_conntrack_proto_fini(); |
ceceae1b | 1012 | nf_conntrack_helper_fini(); |
e9c1b084 | 1013 | nf_conntrack_expect_fini(); |
9fb9cbb1 YK |
1014 | } |
1015 | ||
96eb24d7 | 1016 | struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced) |
9fb9cbb1 | 1017 | { |
f205c5e0 | 1018 | struct hlist_head *hash; |
8e5105a0 | 1019 | unsigned int size, i; |
9fb9cbb1 | 1020 | |
601e68e1 | 1021 | *vmalloced = 0; |
8e5105a0 | 1022 | |
f205c5e0 | 1023 | size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); |
29b67497 | 1024 | hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN, |
f205c5e0 | 1025 | get_order(sizeof(struct hlist_head) |
9fb9cbb1 | 1026 | * size)); |
601e68e1 | 1027 | if (!hash) { |
9fb9cbb1 YK |
1028 | *vmalloced = 1; |
1029 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
f205c5e0 | 1030 | hash = vmalloc(sizeof(struct hlist_head) * size); |
9fb9cbb1 YK |
1031 | } |
1032 | ||
1033 | if (hash) | |
601e68e1 | 1034 | for (i = 0; i < size; i++) |
f205c5e0 | 1035 | INIT_HLIST_HEAD(&hash[i]); |
9fb9cbb1 YK |
1036 | |
1037 | return hash; | |
1038 | } | |
ac565e5f | 1039 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1040 | |
fae718dd | 1041 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1042 | { |
96eb24d7 SH |
1043 | int i, bucket, vmalloced, old_vmalloced; |
1044 | unsigned int hashsize, old_size; | |
9fb9cbb1 | 1045 | int rnd; |
f205c5e0 | 1046 | struct hlist_head *hash, *old_hash; |
9fb9cbb1 YK |
1047 | struct nf_conntrack_tuple_hash *h; |
1048 | ||
1049 | /* On boot, we can set this without any fancy locking. */ | |
1050 | if (!nf_conntrack_htable_size) | |
1051 | return param_set_uint(val, kp); | |
1052 | ||
96eb24d7 | 1053 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1054 | if (!hashsize) |
1055 | return -EINVAL; | |
1056 | ||
ac565e5f | 1057 | hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); |
9fb9cbb1 YK |
1058 | if (!hash) |
1059 | return -ENOMEM; | |
1060 | ||
1061 | /* We have to rehahs for the new table anyway, so we also can | |
1062 | * use a newrandom seed */ | |
1063 | get_random_bytes(&rnd, 4); | |
1064 | ||
76507f69 PM |
1065 | /* Lookups in the old hash might happen in parallel, which means we |
1066 | * might get false negatives during connection lookup. New connections | |
1067 | * created because of a false negative won't make it into the hash | |
1068 | * though since that required taking the lock. | |
1069 | */ | |
f8ba1aff | 1070 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1071 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
f205c5e0 PM |
1072 | while (!hlist_empty(&nf_conntrack_hash[i])) { |
1073 | h = hlist_entry(nf_conntrack_hash[i].first, | |
1074 | struct nf_conntrack_tuple_hash, hnode); | |
76507f69 | 1075 | hlist_del_rcu(&h->hnode); |
9fb9cbb1 | 1076 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); |
f205c5e0 | 1077 | hlist_add_head(&h->hnode, &hash[bucket]); |
9fb9cbb1 YK |
1078 | } |
1079 | } | |
1080 | old_size = nf_conntrack_htable_size; | |
1081 | old_vmalloced = nf_conntrack_vmalloc; | |
1082 | old_hash = nf_conntrack_hash; | |
1083 | ||
1084 | nf_conntrack_htable_size = hashsize; | |
1085 | nf_conntrack_vmalloc = vmalloced; | |
1086 | nf_conntrack_hash = hash; | |
1087 | nf_conntrack_hash_rnd = rnd; | |
f8ba1aff | 1088 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1089 | |
ac565e5f | 1090 | nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); |
9fb9cbb1 YK |
1091 | return 0; |
1092 | } | |
fae718dd | 1093 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1094 | |
fae718dd | 1095 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1096 | &nf_conntrack_htable_size, 0600); |
1097 | ||
1098 | int __init nf_conntrack_init(void) | |
1099 | { | |
f205c5e0 | 1100 | int max_factor = 8; |
9fb9cbb1 YK |
1101 | int ret; |
1102 | ||
1103 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1104 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1105 | if (!nf_conntrack_htable_size) { |
1106 | nf_conntrack_htable_size | |
1107 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
f205c5e0 | 1108 | / sizeof(struct hlist_head)); |
9fb9cbb1 | 1109 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1110 | nf_conntrack_htable_size = 16384; |
1111 | if (nf_conntrack_htable_size < 32) | |
1112 | nf_conntrack_htable_size = 32; | |
1113 | ||
1114 | /* Use a max. factor of four by default to get the same max as | |
1115 | * with the old struct list_heads. When a table size is given | |
1116 | * we use the old value of 8 to avoid reducing the max. | |
1117 | * entries. */ | |
1118 | max_factor = 4; | |
9fb9cbb1 | 1119 | } |
ac565e5f PM |
1120 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, |
1121 | &nf_conntrack_vmalloc); | |
9fb9cbb1 YK |
1122 | if (!nf_conntrack_hash) { |
1123 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1124 | goto err_out; | |
1125 | } | |
1126 | ||
f205c5e0 | 1127 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 PM |
1128 | |
1129 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1130 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1131 | nf_conntrack_max); | |
1132 | ||
dacd2a1a YK |
1133 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1134 | sizeof(struct nf_conn), | |
20c2df83 | 1135 | 0, 0, NULL); |
dacd2a1a | 1136 | if (!nf_conntrack_cachep) { |
9fb9cbb1 YK |
1137 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); |
1138 | goto err_free_hash; | |
1139 | } | |
1140 | ||
e9c1b084 PM |
1141 | ret = nf_conntrack_proto_init(); |
1142 | if (ret < 0) | |
9fb9cbb1 | 1143 | goto err_free_conntrack_slab; |
9fb9cbb1 | 1144 | |
e9c1b084 | 1145 | ret = nf_conntrack_expect_init(); |
933a41e7 | 1146 | if (ret < 0) |
e9c1b084 | 1147 | goto out_fini_proto; |
933a41e7 | 1148 | |
ceceae1b YK |
1149 | ret = nf_conntrack_helper_init(); |
1150 | if (ret < 0) | |
e9c1b084 | 1151 | goto out_fini_expect; |
ceceae1b | 1152 | |
7d3cdc6b | 1153 | /* For use by REJECT target */ |
b334aadc | 1154 | rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); |
de6e05c4 | 1155 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); |
7d3cdc6b | 1156 | |
9fb9cbb1 YK |
1157 | /* Set up fake conntrack: |
1158 | - to never be deleted, not in any hashes */ | |
1159 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); | |
1160 | /* - and look it like as a confirmed connection */ | |
1161 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1162 | ||
1163 | return ret; | |
1164 | ||
e9c1b084 PM |
1165 | out_fini_expect: |
1166 | nf_conntrack_expect_fini(); | |
ceceae1b YK |
1167 | out_fini_proto: |
1168 | nf_conntrack_proto_fini(); | |
9fb9cbb1 | 1169 | err_free_conntrack_slab: |
dacd2a1a | 1170 | kmem_cache_destroy(nf_conntrack_cachep); |
9fb9cbb1 | 1171 | err_free_hash: |
ac565e5f PM |
1172 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, |
1173 | nf_conntrack_htable_size); | |
9fb9cbb1 YK |
1174 | err_out: |
1175 | return -ENOMEM; | |
1176 | } |