]>
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 YK |
42 | |
43 | #if 0 | |
44 | #define DEBUGP printk | |
45 | #else | |
46 | #define DEBUGP(format, args...) | |
47 | #endif | |
48 | ||
49 | DEFINE_RWLOCK(nf_conntrack_lock); | |
13b18339 | 50 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 YK |
51 | |
52 | /* nf_conntrack_standalone needs this */ | |
53 | atomic_t nf_conntrack_count = ATOMIC_INIT(0); | |
a999e683 | 54 | EXPORT_SYMBOL_GPL(nf_conntrack_count); |
9fb9cbb1 | 55 | |
13b18339 PM |
56 | void (*nf_conntrack_destroyed)(struct nf_conn *conntrack); |
57 | EXPORT_SYMBOL_GPL(nf_conntrack_destroyed); | |
58 | ||
e2b7606c | 59 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
60 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
61 | ||
94aec08e | 62 | int nf_conntrack_max __read_mostly; |
a999e683 | 63 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 64 | |
1192e403 | 65 | struct list_head *nf_conntrack_hash __read_mostly; |
13b18339 PM |
66 | EXPORT_SYMBOL_GPL(nf_conntrack_hash); |
67 | ||
e2b7606c | 68 | struct nf_conn nf_conntrack_untracked __read_mostly; |
13b18339 PM |
69 | EXPORT_SYMBOL_GPL(nf_conntrack_untracked); |
70 | ||
94aec08e | 71 | unsigned int nf_ct_log_invalid __read_mostly; |
7e5d03bb | 72 | LIST_HEAD(unconfirmed); |
1192e403 | 73 | static int nf_conntrack_vmalloc __read_mostly; |
9fb9cbb1 | 74 | |
4e3882f7 | 75 | static unsigned int nf_conntrack_next_id; |
77ab9cff | 76 | |
9fb9cbb1 YK |
77 | DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); |
78 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); | |
79 | ||
80 | /* | |
81 | * This scheme offers various size of "struct nf_conn" dependent on | |
82 | * features(helper, nat, ...) | |
83 | */ | |
84 | ||
85 | #define NF_CT_FEATURES_NAMELEN 256 | |
86 | static struct { | |
87 | /* name of slab cache. printed in /proc/slabinfo */ | |
88 | char *name; | |
89 | ||
90 | /* size of slab cache */ | |
91 | size_t size; | |
92 | ||
93 | /* slab cache pointer */ | |
e18b890b | 94 | struct kmem_cache *cachep; |
9fb9cbb1 YK |
95 | |
96 | /* allocated slab cache + modules which uses this slab cache */ | |
97 | int use; | |
98 | ||
9fb9cbb1 YK |
99 | } nf_ct_cache[NF_CT_F_NUM]; |
100 | ||
101 | /* protect members of nf_ct_cache except of "use" */ | |
102 | DEFINE_RWLOCK(nf_ct_cache_lock); | |
103 | ||
104 | /* This avoids calling kmem_cache_create() with same name simultaneously */ | |
57b47a53 | 105 | static DEFINE_MUTEX(nf_ct_cache_mutex); |
9fb9cbb1 | 106 | |
9fb9cbb1 YK |
107 | static int nf_conntrack_hash_rnd_initted; |
108 | static unsigned int nf_conntrack_hash_rnd; | |
109 | ||
110 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
111 | unsigned int size, unsigned int rnd) | |
112 | { | |
113 | unsigned int a, b; | |
9b887909 SF |
114 | |
115 | a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all), | |
116 | (tuple->src.l3num << 16) | tuple->dst.protonum); | |
117 | b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all), | |
118 | (tuple->src.u.all << 16) | tuple->dst.u.all); | |
9fb9cbb1 YK |
119 | |
120 | return jhash_2words(a, b, rnd) % size; | |
121 | } | |
122 | ||
123 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
124 | { | |
125 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
126 | nf_conntrack_hash_rnd); | |
127 | } | |
128 | ||
9fb9cbb1 | 129 | int nf_conntrack_register_cache(u_int32_t features, const char *name, |
dc808fe2 | 130 | size_t size) |
9fb9cbb1 YK |
131 | { |
132 | int ret = 0; | |
133 | char *cache_name; | |
e18b890b | 134 | struct kmem_cache *cachep; |
9fb9cbb1 YK |
135 | |
136 | DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n", | |
137 | features, name, size); | |
138 | ||
139 | if (features < NF_CT_F_BASIC || features >= NF_CT_F_NUM) { | |
140 | DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n", | |
141 | features); | |
142 | return -EINVAL; | |
143 | } | |
144 | ||
57b47a53 | 145 | mutex_lock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
146 | |
147 | write_lock_bh(&nf_ct_cache_lock); | |
148 | /* e.g: multiple helpers are loaded */ | |
149 | if (nf_ct_cache[features].use > 0) { | |
150 | DEBUGP("nf_conntrack_register_cache: already resisterd.\n"); | |
151 | if ((!strncmp(nf_ct_cache[features].name, name, | |
152 | NF_CT_FEATURES_NAMELEN)) | |
dc808fe2 | 153 | && nf_ct_cache[features].size == size) { |
9fb9cbb1 YK |
154 | DEBUGP("nf_conntrack_register_cache: reusing.\n"); |
155 | nf_ct_cache[features].use++; | |
156 | ret = 0; | |
157 | } else | |
158 | ret = -EBUSY; | |
159 | ||
160 | write_unlock_bh(&nf_ct_cache_lock); | |
57b47a53 | 161 | mutex_unlock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
162 | return ret; |
163 | } | |
164 | write_unlock_bh(&nf_ct_cache_lock); | |
165 | ||
166 | /* | |
167 | * The memory space for name of slab cache must be alive until | |
168 | * cache is destroyed. | |
169 | */ | |
170 | cache_name = kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN, GFP_ATOMIC); | |
171 | if (cache_name == NULL) { | |
172 | DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n"); | |
173 | ret = -ENOMEM; | |
174 | goto out_up_mutex; | |
175 | } | |
176 | ||
177 | if (strlcpy(cache_name, name, NF_CT_FEATURES_NAMELEN) | |
178 | >= NF_CT_FEATURES_NAMELEN) { | |
179 | printk("nf_conntrack_register_cache: name too long\n"); | |
180 | ret = -EINVAL; | |
181 | goto out_free_name; | |
182 | } | |
183 | ||
184 | cachep = kmem_cache_create(cache_name, size, 0, 0, | |
185 | NULL, NULL); | |
186 | if (!cachep) { | |
187 | printk("nf_conntrack_register_cache: Can't create slab cache " | |
188 | "for the features = 0x%x\n", features); | |
189 | ret = -ENOMEM; | |
190 | goto out_free_name; | |
191 | } | |
192 | ||
193 | write_lock_bh(&nf_ct_cache_lock); | |
194 | nf_ct_cache[features].use = 1; | |
195 | nf_ct_cache[features].size = size; | |
9fb9cbb1 YK |
196 | nf_ct_cache[features].cachep = cachep; |
197 | nf_ct_cache[features].name = cache_name; | |
198 | write_unlock_bh(&nf_ct_cache_lock); | |
199 | ||
200 | goto out_up_mutex; | |
201 | ||
202 | out_free_name: | |
203 | kfree(cache_name); | |
204 | out_up_mutex: | |
57b47a53 | 205 | mutex_unlock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
206 | return ret; |
207 | } | |
13b18339 | 208 | EXPORT_SYMBOL_GPL(nf_conntrack_register_cache); |
9fb9cbb1 YK |
209 | |
210 | /* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */ | |
211 | void nf_conntrack_unregister_cache(u_int32_t features) | |
212 | { | |
e18b890b | 213 | struct kmem_cache *cachep; |
9fb9cbb1 YK |
214 | char *name; |
215 | ||
216 | /* | |
217 | * This assures that kmem_cache_create() isn't called before destroying | |
218 | * slab cache. | |
219 | */ | |
220 | DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features); | |
57b47a53 | 221 | mutex_lock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
222 | |
223 | write_lock_bh(&nf_ct_cache_lock); | |
224 | if (--nf_ct_cache[features].use > 0) { | |
225 | write_unlock_bh(&nf_ct_cache_lock); | |
57b47a53 | 226 | mutex_unlock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
227 | return; |
228 | } | |
229 | cachep = nf_ct_cache[features].cachep; | |
230 | name = nf_ct_cache[features].name; | |
231 | nf_ct_cache[features].cachep = NULL; | |
232 | nf_ct_cache[features].name = NULL; | |
9fb9cbb1 YK |
233 | nf_ct_cache[features].size = 0; |
234 | write_unlock_bh(&nf_ct_cache_lock); | |
235 | ||
236 | synchronize_net(); | |
237 | ||
238 | kmem_cache_destroy(cachep); | |
239 | kfree(name); | |
240 | ||
57b47a53 | 241 | mutex_unlock(&nf_ct_cache_mutex); |
9fb9cbb1 | 242 | } |
13b18339 | 243 | EXPORT_SYMBOL_GPL(nf_conntrack_unregister_cache); |
9fb9cbb1 YK |
244 | |
245 | int | |
246 | nf_ct_get_tuple(const struct sk_buff *skb, | |
247 | unsigned int nhoff, | |
248 | unsigned int dataoff, | |
249 | u_int16_t l3num, | |
250 | u_int8_t protonum, | |
251 | struct nf_conntrack_tuple *tuple, | |
252 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 253 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 YK |
254 | { |
255 | NF_CT_TUPLE_U_BLANK(tuple); | |
256 | ||
257 | tuple->src.l3num = l3num; | |
258 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
259 | return 0; | |
260 | ||
261 | tuple->dst.protonum = protonum; | |
262 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
263 | ||
605dcad6 | 264 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 265 | } |
13b18339 | 266 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 YK |
267 | |
268 | int | |
269 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, | |
270 | const struct nf_conntrack_tuple *orig, | |
271 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 272 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 YK |
273 | { |
274 | NF_CT_TUPLE_U_BLANK(inverse); | |
275 | ||
276 | inverse->src.l3num = orig->src.l3num; | |
277 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
278 | return 0; | |
279 | ||
280 | inverse->dst.dir = !orig->dst.dir; | |
281 | ||
282 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 283 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 284 | } |
13b18339 | 285 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 286 | |
9fb9cbb1 YK |
287 | static void |
288 | clean_from_lists(struct nf_conn *ct) | |
289 | { | |
9fb9cbb1 | 290 | DEBUGP("clean_from_lists(%p)\n", ct); |
df0933dc PM |
291 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); |
292 | list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list); | |
9fb9cbb1 YK |
293 | |
294 | /* Destroy all pending expectations */ | |
c1d10adb | 295 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
296 | } |
297 | ||
298 | static void | |
299 | destroy_conntrack(struct nf_conntrack *nfct) | |
300 | { | |
301 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 302 | struct nf_conntrack_l4proto *l4proto; |
982d9a9c | 303 | typeof(nf_conntrack_destroyed) destroyed; |
9fb9cbb1 YK |
304 | |
305 | DEBUGP("destroy_conntrack(%p)\n", ct); | |
306 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); | |
307 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
308 | ||
309 | nf_conntrack_event(IPCT_DESTROY, ct); | |
310 | set_bit(IPS_DYING_BIT, &ct->status); | |
311 | ||
312 | /* To make sure we don't get any weird locking issues here: | |
313 | * destroy_conntrack() MUST NOT be called with a write lock | |
314 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 315 | rcu_read_lock(); |
923f4902 PM |
316 | l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, |
317 | ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); | |
605dcad6 MJ |
318 | if (l4proto && l4proto->destroy) |
319 | l4proto->destroy(ct); | |
9fb9cbb1 | 320 | |
ecfab2c9 YK |
321 | nf_ct_ext_destroy(ct); |
322 | ||
982d9a9c PM |
323 | destroyed = rcu_dereference(nf_conntrack_destroyed); |
324 | if (destroyed) | |
325 | destroyed(ct); | |
326 | ||
327 | rcu_read_unlock(); | |
9fb9cbb1 YK |
328 | |
329 | write_lock_bh(&nf_conntrack_lock); | |
330 | /* Expectations will have been removed in clean_from_lists, | |
331 | * except TFTP can create an expectation on the first packet, | |
332 | * before connection is in the list, so we need to clean here, | |
333 | * too. */ | |
c1d10adb | 334 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
335 | |
336 | /* We overload first tuple to link into unconfirmed list. */ | |
337 | if (!nf_ct_is_confirmed(ct)) { | |
338 | BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list)); | |
339 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); | |
340 | } | |
341 | ||
342 | NF_CT_STAT_INC(delete); | |
343 | write_unlock_bh(&nf_conntrack_lock); | |
344 | ||
345 | if (ct->master) | |
346 | nf_ct_put(ct->master); | |
347 | ||
348 | DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct); | |
349 | nf_conntrack_free(ct); | |
350 | } | |
351 | ||
352 | static void death_by_timeout(unsigned long ul_conntrack) | |
353 | { | |
354 | struct nf_conn *ct = (void *)ul_conntrack; | |
5397e97d | 355 | struct nf_conn_help *help = nfct_help(ct); |
3c158f7f | 356 | struct nf_conntrack_helper *helper; |
5397e97d | 357 | |
3c158f7f PM |
358 | if (help) { |
359 | rcu_read_lock(); | |
360 | helper = rcu_dereference(help->helper); | |
361 | if (helper && helper->destroy) | |
362 | helper->destroy(ct); | |
363 | rcu_read_unlock(); | |
364 | } | |
9fb9cbb1 YK |
365 | |
366 | write_lock_bh(&nf_conntrack_lock); | |
367 | /* Inside lock so preempt is disabled on module removal path. | |
368 | * Otherwise we can get spurious warnings. */ | |
369 | NF_CT_STAT_INC(delete_list); | |
370 | clean_from_lists(ct); | |
371 | write_unlock_bh(&nf_conntrack_lock); | |
372 | nf_ct_put(ct); | |
373 | } | |
374 | ||
c1d10adb | 375 | struct nf_conntrack_tuple_hash * |
9fb9cbb1 YK |
376 | __nf_conntrack_find(const struct nf_conntrack_tuple *tuple, |
377 | const struct nf_conn *ignored_conntrack) | |
378 | { | |
379 | struct nf_conntrack_tuple_hash *h; | |
380 | unsigned int hash = hash_conntrack(tuple); | |
381 | ||
9fb9cbb1 | 382 | list_for_each_entry(h, &nf_conntrack_hash[hash], list) { |
df0933dc PM |
383 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
384 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
9fb9cbb1 YK |
385 | NF_CT_STAT_INC(found); |
386 | return h; | |
387 | } | |
388 | NF_CT_STAT_INC(searched); | |
389 | } | |
390 | ||
391 | return NULL; | |
392 | } | |
13b18339 | 393 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
394 | |
395 | /* Find a connection corresponding to a tuple. */ | |
396 | struct nf_conntrack_tuple_hash * | |
397 | nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple, | |
398 | const struct nf_conn *ignored_conntrack) | |
399 | { | |
400 | struct nf_conntrack_tuple_hash *h; | |
401 | ||
402 | read_lock_bh(&nf_conntrack_lock); | |
403 | h = __nf_conntrack_find(tuple, ignored_conntrack); | |
404 | if (h) | |
405 | atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use); | |
406 | read_unlock_bh(&nf_conntrack_lock); | |
407 | ||
408 | return h; | |
409 | } | |
13b18339 | 410 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 411 | |
c1d10adb PNA |
412 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
413 | unsigned int hash, | |
601e68e1 | 414 | unsigned int repl_hash) |
c1d10adb PNA |
415 | { |
416 | ct->id = ++nf_conntrack_next_id; | |
df0933dc PM |
417 | list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list, |
418 | &nf_conntrack_hash[hash]); | |
419 | list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list, | |
420 | &nf_conntrack_hash[repl_hash]); | |
c1d10adb PNA |
421 | } |
422 | ||
423 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
424 | { | |
425 | unsigned int hash, repl_hash; | |
426 | ||
427 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
428 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
429 | ||
430 | write_lock_bh(&nf_conntrack_lock); | |
431 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
432 | write_unlock_bh(&nf_conntrack_lock); | |
433 | } | |
13b18339 | 434 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); |
c1d10adb | 435 | |
9fb9cbb1 YK |
436 | /* Confirm a connection given skb; places it in hash table */ |
437 | int | |
438 | __nf_conntrack_confirm(struct sk_buff **pskb) | |
439 | { | |
440 | unsigned int hash, repl_hash; | |
df0933dc | 441 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 442 | struct nf_conn *ct; |
df0933dc | 443 | struct nf_conn_help *help; |
9fb9cbb1 YK |
444 | enum ip_conntrack_info ctinfo; |
445 | ||
446 | ct = nf_ct_get(*pskb, &ctinfo); | |
447 | ||
448 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
449 | ICMP/TCP RST packets in other direction. Actual packet | |
450 | which created connection will be IP_CT_NEW or for an | |
451 | expected connection, IP_CT_RELATED. */ | |
452 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
453 | return NF_ACCEPT; | |
454 | ||
455 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
456 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
457 | ||
458 | /* We're not in hash table, and we refuse to set up related | |
459 | connections for unconfirmed conns. But packet copies and | |
460 | REJECT will give spurious warnings here. */ | |
461 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
462 | ||
463 | /* No external references means noone else could have | |
464 | confirmed us. */ | |
465 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
466 | DEBUGP("Confirming conntrack %p\n", ct); | |
467 | ||
468 | write_lock_bh(&nf_conntrack_lock); | |
469 | ||
470 | /* See if there's one in the list already, including reverse: | |
471 | NAT could have grabbed it without realizing, since we're | |
472 | not in the hash. If there is, we lost race. */ | |
df0933dc PM |
473 | list_for_each_entry(h, &nf_conntrack_hash[hash], list) |
474 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, | |
475 | &h->tuple)) | |
476 | goto out; | |
477 | list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list) | |
478 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, | |
479 | &h->tuple)) | |
480 | goto out; | |
9fb9cbb1 | 481 | |
df0933dc PM |
482 | /* Remove from unconfirmed list */ |
483 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); | |
484 | ||
485 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
486 | /* Timer relative to confirmation time, not original | |
487 | setting time, otherwise we'd get timer wrap in | |
488 | weird delay cases. */ | |
489 | ct->timeout.expires += jiffies; | |
490 | add_timer(&ct->timeout); | |
491 | atomic_inc(&ct->ct_general.use); | |
492 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
493 | NF_CT_STAT_INC(insert); | |
494 | write_unlock_bh(&nf_conntrack_lock); | |
495 | help = nfct_help(ct); | |
496 | if (help && help->helper) | |
497 | nf_conntrack_event_cache(IPCT_HELPER, *pskb); | |
9fb9cbb1 | 498 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
499 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
500 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
501 | nf_conntrack_event_cache(IPCT_NATINFO, *pskb); | |
9fb9cbb1 | 502 | #endif |
df0933dc PM |
503 | nf_conntrack_event_cache(master_ct(ct) ? |
504 | IPCT_RELATED : IPCT_NEW, *pskb); | |
505 | return NF_ACCEPT; | |
9fb9cbb1 | 506 | |
df0933dc | 507 | out: |
9fb9cbb1 YK |
508 | NF_CT_STAT_INC(insert_failed); |
509 | write_unlock_bh(&nf_conntrack_lock); | |
510 | return NF_DROP; | |
511 | } | |
13b18339 | 512 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
513 | |
514 | /* Returns true if a connection correspondings to the tuple (required | |
515 | for NAT). */ | |
516 | int | |
517 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
518 | const struct nf_conn *ignored_conntrack) | |
519 | { | |
520 | struct nf_conntrack_tuple_hash *h; | |
521 | ||
522 | read_lock_bh(&nf_conntrack_lock); | |
523 | h = __nf_conntrack_find(tuple, ignored_conntrack); | |
524 | read_unlock_bh(&nf_conntrack_lock); | |
525 | ||
526 | return h != NULL; | |
527 | } | |
13b18339 | 528 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 YK |
529 | |
530 | /* There's a small race here where we may free a just-assured | |
531 | connection. Too bad: we're in trouble anyway. */ | |
9fb9cbb1 YK |
532 | static int early_drop(struct list_head *chain) |
533 | { | |
534 | /* Traverse backwards: gives us oldest, which is roughly LRU */ | |
535 | struct nf_conntrack_tuple_hash *h; | |
df0933dc | 536 | struct nf_conn *ct = NULL, *tmp; |
9fb9cbb1 YK |
537 | int dropped = 0; |
538 | ||
539 | read_lock_bh(&nf_conntrack_lock); | |
df0933dc PM |
540 | list_for_each_entry_reverse(h, chain, list) { |
541 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
542 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) { | |
543 | ct = tmp; | |
544 | atomic_inc(&ct->ct_general.use); | |
545 | break; | |
546 | } | |
9fb9cbb1 YK |
547 | } |
548 | read_unlock_bh(&nf_conntrack_lock); | |
549 | ||
550 | if (!ct) | |
551 | return dropped; | |
552 | ||
553 | if (del_timer(&ct->timeout)) { | |
554 | death_by_timeout((unsigned long)ct); | |
555 | dropped = 1; | |
c0e912d7 | 556 | NF_CT_STAT_INC_ATOMIC(early_drop); |
9fb9cbb1 YK |
557 | } |
558 | nf_ct_put(ct); | |
559 | return dropped; | |
560 | } | |
561 | ||
9fb9cbb1 YK |
562 | static struct nf_conn * |
563 | __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, | |
564 | const struct nf_conntrack_tuple *repl, | |
9457d851 PM |
565 | const struct nf_conntrack_l3proto *l3proto, |
566 | u_int32_t features) | |
9fb9cbb1 YK |
567 | { |
568 | struct nf_conn *conntrack = NULL; | |
9fb9cbb1 | 569 | |
dc808fe2 | 570 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
571 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
572 | nf_conntrack_hash_rnd_initted = 1; | |
573 | } | |
574 | ||
5251e2d2 PNA |
575 | /* We don't want any race condition at early drop stage */ |
576 | atomic_inc(&nf_conntrack_count); | |
577 | ||
9fb9cbb1 | 578 | if (nf_conntrack_max |
5251e2d2 | 579 | && atomic_read(&nf_conntrack_count) > nf_conntrack_max) { |
9fb9cbb1 YK |
580 | unsigned int hash = hash_conntrack(orig); |
581 | /* Try dropping from this hash chain. */ | |
582 | if (!early_drop(&nf_conntrack_hash[hash])) { | |
5251e2d2 | 583 | atomic_dec(&nf_conntrack_count); |
9fb9cbb1 YK |
584 | if (net_ratelimit()) |
585 | printk(KERN_WARNING | |
586 | "nf_conntrack: table full, dropping" | |
587 | " packet.\n"); | |
588 | return ERR_PTR(-ENOMEM); | |
589 | } | |
590 | } | |
591 | ||
592 | /* find features needed by this conntrack. */ | |
9457d851 | 593 | features |= l3proto->get_features(orig); |
dc808fe2 | 594 | |
9fb9cbb1 YK |
595 | DEBUGP("nf_conntrack_alloc: features=0x%x\n", features); |
596 | ||
597 | read_lock_bh(&nf_ct_cache_lock); | |
598 | ||
dc808fe2 | 599 | if (unlikely(!nf_ct_cache[features].use)) { |
9fb9cbb1 YK |
600 | DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n", |
601 | features); | |
602 | goto out; | |
603 | } | |
604 | ||
605 | conntrack = kmem_cache_alloc(nf_ct_cache[features].cachep, GFP_ATOMIC); | |
606 | if (conntrack == NULL) { | |
607 | DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n"); | |
608 | goto out; | |
609 | } | |
610 | ||
611 | memset(conntrack, 0, nf_ct_cache[features].size); | |
612 | conntrack->features = features; | |
9fb9cbb1 | 613 | atomic_set(&conntrack->ct_general.use, 1); |
9fb9cbb1 YK |
614 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
615 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
616 | /* Don't set timer yet: wait for confirmation */ | |
e6f689db PM |
617 | setup_timer(&conntrack->timeout, death_by_timeout, |
618 | (unsigned long)conntrack); | |
5251e2d2 | 619 | read_unlock_bh(&nf_ct_cache_lock); |
9fb9cbb1 | 620 | |
5251e2d2 | 621 | return conntrack; |
9fb9cbb1 YK |
622 | out: |
623 | read_unlock_bh(&nf_ct_cache_lock); | |
5251e2d2 | 624 | atomic_dec(&nf_conntrack_count); |
9fb9cbb1 YK |
625 | return conntrack; |
626 | } | |
627 | ||
628 | struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, | |
629 | const struct nf_conntrack_tuple *repl) | |
630 | { | |
631 | struct nf_conntrack_l3proto *l3proto; | |
923f4902 | 632 | struct nf_conn *ct; |
9fb9cbb1 | 633 | |
923f4902 | 634 | rcu_read_lock(); |
c1d10adb | 635 | l3proto = __nf_ct_l3proto_find(orig->src.l3num); |
923f4902 PM |
636 | ct = __nf_conntrack_alloc(orig, repl, l3proto, 0); |
637 | rcu_read_unlock(); | |
638 | ||
639 | return ct; | |
9fb9cbb1 | 640 | } |
13b18339 | 641 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 YK |
642 | |
643 | void nf_conntrack_free(struct nf_conn *conntrack) | |
644 | { | |
645 | u_int32_t features = conntrack->features; | |
646 | NF_CT_ASSERT(features >= NF_CT_F_BASIC && features < NF_CT_F_NUM); | |
ecfab2c9 | 647 | nf_ct_ext_free(conntrack); |
9fb9cbb1 YK |
648 | DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features, |
649 | conntrack); | |
650 | kmem_cache_free(nf_ct_cache[features].cachep, conntrack); | |
651 | atomic_dec(&nf_conntrack_count); | |
652 | } | |
13b18339 | 653 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
654 | |
655 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
656 | failed due to stress. Otherwise it really is unclassifiable. */ | |
657 | static struct nf_conntrack_tuple_hash * | |
658 | init_conntrack(const struct nf_conntrack_tuple *tuple, | |
659 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 660 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
661 | struct sk_buff *skb, |
662 | unsigned int dataoff) | |
663 | { | |
664 | struct nf_conn *conntrack; | |
3c158f7f | 665 | struct nf_conn_help *help; |
9fb9cbb1 YK |
666 | struct nf_conntrack_tuple repl_tuple; |
667 | struct nf_conntrack_expect *exp; | |
9457d851 | 668 | u_int32_t features = 0; |
9fb9cbb1 | 669 | |
605dcad6 | 670 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
9fb9cbb1 YK |
671 | DEBUGP("Can't invert tuple.\n"); |
672 | return NULL; | |
673 | } | |
674 | ||
9457d851 | 675 | conntrack = __nf_conntrack_alloc(tuple, &repl_tuple, l3proto, features); |
9fb9cbb1 YK |
676 | if (conntrack == NULL || IS_ERR(conntrack)) { |
677 | DEBUGP("Can't allocate conntrack.\n"); | |
678 | return (struct nf_conntrack_tuple_hash *)conntrack; | |
679 | } | |
680 | ||
605dcad6 | 681 | if (!l4proto->new(conntrack, skb, dataoff)) { |
9fb9cbb1 YK |
682 | nf_conntrack_free(conntrack); |
683 | DEBUGP("init conntrack: can't track with proto module\n"); | |
684 | return NULL; | |
685 | } | |
686 | ||
687 | write_lock_bh(&nf_conntrack_lock); | |
688 | exp = find_expectation(tuple); | |
9fb9cbb1 YK |
689 | if (exp) { |
690 | DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n", | |
691 | conntrack, exp); | |
692 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
693 | __set_bit(IPS_EXPECTED_BIT, &conntrack->status); | |
694 | conntrack->master = exp->master; | |
ceceae1b YK |
695 | if (exp->helper) { |
696 | help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER, | |
697 | GFP_ATOMIC); | |
698 | if (help) | |
699 | rcu_assign_pointer(help->helper, exp->helper); | |
700 | else | |
701 | DEBUGP("failed to add helper extension area"); | |
702 | } | |
703 | ||
9fb9cbb1 YK |
704 | #ifdef CONFIG_NF_CONNTRACK_MARK |
705 | conntrack->mark = exp->master->mark; | |
7c9728c3 JM |
706 | #endif |
707 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
708 | conntrack->secmark = exp->master->secmark; | |
9fb9cbb1 YK |
709 | #endif |
710 | nf_conntrack_get(&conntrack->master->ct_general); | |
711 | NF_CT_STAT_INC(expect_new); | |
22e7410b | 712 | } else { |
ceceae1b YK |
713 | struct nf_conntrack_helper *helper; |
714 | ||
715 | helper = __nf_ct_helper_find(&repl_tuple); | |
716 | if (helper) { | |
717 | help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER, | |
718 | GFP_ATOMIC); | |
719 | if (help) | |
720 | /* not in hash table yet, so not strictly | |
721 | necessary */ | |
722 | rcu_assign_pointer(help->helper, helper); | |
723 | else | |
724 | DEBUGP("failed to add helper extension area"); | |
3c158f7f | 725 | } |
9fb9cbb1 | 726 | NF_CT_STAT_INC(new); |
22e7410b | 727 | } |
9fb9cbb1 YK |
728 | |
729 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
730 | list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed); | |
731 | ||
732 | write_unlock_bh(&nf_conntrack_lock); | |
733 | ||
734 | if (exp) { | |
735 | if (exp->expectfn) | |
736 | exp->expectfn(conntrack, exp); | |
737 | nf_conntrack_expect_put(exp); | |
738 | } | |
739 | ||
740 | return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; | |
741 | } | |
742 | ||
743 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
744 | static inline struct nf_conn * | |
745 | resolve_normal_ct(struct sk_buff *skb, | |
746 | unsigned int dataoff, | |
747 | u_int16_t l3num, | |
748 | u_int8_t protonum, | |
749 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 750 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
751 | int *set_reply, |
752 | enum ip_conntrack_info *ctinfo) | |
753 | { | |
754 | struct nf_conntrack_tuple tuple; | |
755 | struct nf_conntrack_tuple_hash *h; | |
756 | struct nf_conn *ct; | |
757 | ||
bbe735e4 | 758 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 759 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 760 | l4proto)) { |
9fb9cbb1 YK |
761 | DEBUGP("resolve_normal_ct: Can't get tuple\n"); |
762 | return NULL; | |
763 | } | |
764 | ||
765 | /* look for tuple match */ | |
766 | h = nf_conntrack_find_get(&tuple, NULL); | |
767 | if (!h) { | |
605dcad6 | 768 | h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
769 | if (!h) |
770 | return NULL; | |
771 | if (IS_ERR(h)) | |
772 | return (void *)h; | |
773 | } | |
774 | ct = nf_ct_tuplehash_to_ctrack(h); | |
775 | ||
776 | /* It exists; we have (non-exclusive) reference. */ | |
777 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
778 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
779 | /* Please set reply bit if this packet OK */ | |
780 | *set_reply = 1; | |
781 | } else { | |
782 | /* Once we've had two way comms, always ESTABLISHED. */ | |
783 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
784 | DEBUGP("nf_conntrack_in: normal packet for %p\n", ct); | |
785 | *ctinfo = IP_CT_ESTABLISHED; | |
786 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
787 | DEBUGP("nf_conntrack_in: related packet for %p\n", ct); | |
788 | *ctinfo = IP_CT_RELATED; | |
789 | } else { | |
790 | DEBUGP("nf_conntrack_in: new packet for %p\n", ct); | |
791 | *ctinfo = IP_CT_NEW; | |
792 | } | |
793 | *set_reply = 0; | |
794 | } | |
795 | skb->nfct = &ct->ct_general; | |
796 | skb->nfctinfo = *ctinfo; | |
797 | return ct; | |
798 | } | |
799 | ||
800 | unsigned int | |
801 | nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb) | |
802 | { | |
803 | struct nf_conn *ct; | |
804 | enum ip_conntrack_info ctinfo; | |
805 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 806 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
807 | unsigned int dataoff; |
808 | u_int8_t protonum; | |
809 | int set_reply = 0; | |
810 | int ret; | |
811 | ||
812 | /* Previously seen (loopback or untracked)? Ignore. */ | |
813 | if ((*pskb)->nfct) { | |
c0e912d7 | 814 | NF_CT_STAT_INC_ATOMIC(ignore); |
9fb9cbb1 YK |
815 | return NF_ACCEPT; |
816 | } | |
817 | ||
923f4902 | 818 | /* rcu_read_lock()ed by nf_hook_slow */ |
c1d10adb | 819 | l3proto = __nf_ct_l3proto_find((u_int16_t)pf); |
923f4902 | 820 | |
9fb9cbb1 YK |
821 | if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) { |
822 | DEBUGP("not prepared to track yet or error occured\n"); | |
823 | return -ret; | |
824 | } | |
825 | ||
605dcad6 | 826 | l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); |
9fb9cbb1 YK |
827 | |
828 | /* It may be an special packet, error, unclean... | |
829 | * inverse of the return code tells to the netfilter | |
830 | * core what to do with the packet. */ | |
605dcad6 MJ |
831 | if (l4proto->error != NULL && |
832 | (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) { | |
c0e912d7 PM |
833 | NF_CT_STAT_INC_ATOMIC(error); |
834 | NF_CT_STAT_INC_ATOMIC(invalid); | |
9fb9cbb1 YK |
835 | return -ret; |
836 | } | |
837 | ||
605dcad6 | 838 | ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto, |
9fb9cbb1 YK |
839 | &set_reply, &ctinfo); |
840 | if (!ct) { | |
841 | /* Not valid part of a connection */ | |
c0e912d7 | 842 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
843 | return NF_ACCEPT; |
844 | } | |
845 | ||
846 | if (IS_ERR(ct)) { | |
847 | /* Too stressed to deal. */ | |
c0e912d7 | 848 | NF_CT_STAT_INC_ATOMIC(drop); |
9fb9cbb1 YK |
849 | return NF_DROP; |
850 | } | |
851 | ||
852 | NF_CT_ASSERT((*pskb)->nfct); | |
853 | ||
605dcad6 | 854 | ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
855 | if (ret < 0) { |
856 | /* Invalid: inverse of the return code tells | |
857 | * the netfilter core what to do */ | |
858 | DEBUGP("nf_conntrack_in: Can't track with proto module\n"); | |
859 | nf_conntrack_put((*pskb)->nfct); | |
860 | (*pskb)->nfct = NULL; | |
c0e912d7 | 861 | NF_CT_STAT_INC_ATOMIC(invalid); |
9fb9cbb1 YK |
862 | return -ret; |
863 | } | |
864 | ||
865 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
866 | nf_conntrack_event_cache(IPCT_STATUS, *pskb); | |
867 | ||
868 | return ret; | |
869 | } | |
13b18339 | 870 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 YK |
871 | |
872 | int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, | |
873 | const struct nf_conntrack_tuple *orig) | |
874 | { | |
923f4902 PM |
875 | int ret; |
876 | ||
877 | rcu_read_lock(); | |
878 | ret = nf_ct_invert_tuple(inverse, orig, | |
879 | __nf_ct_l3proto_find(orig->src.l3num), | |
880 | __nf_ct_l4proto_find(orig->src.l3num, | |
881 | orig->dst.protonum)); | |
882 | rcu_read_unlock(); | |
883 | return ret; | |
9fb9cbb1 | 884 | } |
13b18339 | 885 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 886 | |
5b1158e9 JK |
887 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
888 | implicitly racy: see __nf_conntrack_confirm */ | |
889 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
890 | const struct nf_conntrack_tuple *newreply) | |
891 | { | |
892 | struct nf_conn_help *help = nfct_help(ct); | |
ceceae1b | 893 | struct nf_conntrack_helper *helper; |
5b1158e9 JK |
894 | |
895 | write_lock_bh(&nf_conntrack_lock); | |
896 | /* Should be unconfirmed, so not in hash table yet */ | |
897 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
898 | ||
899 | DEBUGP("Altering reply tuple of %p to ", ct); | |
900 | NF_CT_DUMP_TUPLE(newreply); | |
901 | ||
902 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ceceae1b YK |
903 | if (ct->master || (help && help->expecting != 0)) |
904 | goto out; | |
905 | ||
906 | helper = __nf_ct_helper_find(newreply); | |
907 | if (helper == NULL) { | |
908 | if (help) | |
909 | rcu_assign_pointer(help->helper, NULL); | |
910 | goto out; | |
5d78a849 | 911 | } |
ceceae1b YK |
912 | |
913 | if (help == NULL) { | |
914 | help = nf_ct_ext_add(ct, NF_CT_EXT_HELPER, GFP_ATOMIC); | |
915 | if (help == NULL) { | |
916 | DEBUGP("failed to add helper extension area"); | |
917 | goto out; | |
918 | } | |
919 | } else { | |
920 | memset(&help->help, 0, sizeof(help->help)); | |
921 | } | |
922 | ||
923 | rcu_assign_pointer(help->helper, helper); | |
924 | out: | |
5b1158e9 JK |
925 | write_unlock_bh(&nf_conntrack_lock); |
926 | } | |
13b18339 | 927 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 928 | |
9fb9cbb1 YK |
929 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
930 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
931 | enum ip_conntrack_info ctinfo, | |
932 | const struct sk_buff *skb, | |
933 | unsigned long extra_jiffies, | |
934 | int do_acct) | |
935 | { | |
936 | int event = 0; | |
937 | ||
938 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
939 | NF_CT_ASSERT(skb); | |
940 | ||
941 | write_lock_bh(&nf_conntrack_lock); | |
942 | ||
997ae831 EL |
943 | /* Only update if this is not a fixed timeout */ |
944 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) { | |
945 | write_unlock_bh(&nf_conntrack_lock); | |
946 | return; | |
947 | } | |
948 | ||
9fb9cbb1 YK |
949 | /* If not in hash table, timer will not be active yet */ |
950 | if (!nf_ct_is_confirmed(ct)) { | |
951 | ct->timeout.expires = extra_jiffies; | |
952 | event = IPCT_REFRESH; | |
953 | } else { | |
be00c8e4 MJ |
954 | unsigned long newtime = jiffies + extra_jiffies; |
955 | ||
956 | /* Only update the timeout if the new timeout is at least | |
957 | HZ jiffies from the old timeout. Need del_timer for race | |
958 | avoidance (may already be dying). */ | |
959 | if (newtime - ct->timeout.expires >= HZ | |
960 | && del_timer(&ct->timeout)) { | |
961 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
962 | add_timer(&ct->timeout); |
963 | event = IPCT_REFRESH; | |
964 | } | |
965 | } | |
966 | ||
967 | #ifdef CONFIG_NF_CT_ACCT | |
968 | if (do_acct) { | |
969 | ct->counters[CTINFO2DIR(ctinfo)].packets++; | |
970 | ct->counters[CTINFO2DIR(ctinfo)].bytes += | |
bbe735e4 | 971 | skb->len - skb_network_offset(skb); |
3ffd5eeb MJ |
972 | |
973 | if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) | |
974 | || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) | |
975 | event |= IPCT_COUNTER_FILLING; | |
9fb9cbb1 YK |
976 | } |
977 | #endif | |
978 | ||
979 | write_unlock_bh(&nf_conntrack_lock); | |
980 | ||
981 | /* must be unlocked when calling event cache */ | |
982 | if (event) | |
983 | nf_conntrack_event_cache(event, skb); | |
984 | } | |
13b18339 | 985 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 986 | |
e281db5c | 987 | #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) |
c1d10adb PNA |
988 | |
989 | #include <linux/netfilter/nfnetlink.h> | |
990 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
991 | #include <linux/mutex.h> |
992 | ||
c1d10adb PNA |
993 | |
994 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be | |
995 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
996 | * or depend on ctnetlink */ | |
997 | int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb, | |
998 | const struct nf_conntrack_tuple *tuple) | |
999 | { | |
1000 | NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t), | |
1001 | &tuple->src.u.tcp.port); | |
1002 | NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t), | |
1003 | &tuple->dst.u.tcp.port); | |
1004 | return 0; | |
1005 | ||
1006 | nfattr_failure: | |
1007 | return -1; | |
1008 | } | |
13b18339 | 1009 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr); |
c1d10adb PNA |
1010 | |
1011 | static const size_t cta_min_proto[CTA_PROTO_MAX] = { | |
1012 | [CTA_PROTO_SRC_PORT-1] = sizeof(u_int16_t), | |
1013 | [CTA_PROTO_DST_PORT-1] = sizeof(u_int16_t) | |
1014 | }; | |
1015 | ||
1016 | int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[], | |
1017 | struct nf_conntrack_tuple *t) | |
1018 | { | |
1019 | if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1]) | |
1020 | return -EINVAL; | |
1021 | ||
1022 | if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto)) | |
1023 | return -EINVAL; | |
1024 | ||
bff9a89b PM |
1025 | t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]); |
1026 | t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]); | |
c1d10adb PNA |
1027 | |
1028 | return 0; | |
1029 | } | |
13b18339 | 1030 | EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple); |
c1d10adb PNA |
1031 | #endif |
1032 | ||
9fb9cbb1 YK |
1033 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
1034 | void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) | |
1035 | { | |
1036 | struct nf_conn *ct; | |
1037 | enum ip_conntrack_info ctinfo; | |
1038 | ||
1039 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1040 | ct = nf_ct_get(skb, &ctinfo); | |
1041 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
1042 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
1043 | else | |
1044 | ctinfo = IP_CT_RELATED; | |
1045 | ||
1046 | /* Attach to new skbuff, and increment count */ | |
1047 | nskb->nfct = &ct->ct_general; | |
1048 | nskb->nfctinfo = ctinfo; | |
1049 | nf_conntrack_get(nskb->nfct); | |
1050 | } | |
13b18339 | 1051 | EXPORT_SYMBOL_GPL(__nf_conntrack_attach); |
9fb9cbb1 YK |
1052 | |
1053 | static inline int | |
1054 | do_iter(const struct nf_conntrack_tuple_hash *i, | |
1055 | int (*iter)(struct nf_conn *i, void *data), | |
1056 | void *data) | |
1057 | { | |
1058 | return iter(nf_ct_tuplehash_to_ctrack(i), data); | |
1059 | } | |
1060 | ||
1061 | /* Bring out ya dead! */ | |
df0933dc | 1062 | static struct nf_conn * |
9fb9cbb1 YK |
1063 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
1064 | void *data, unsigned int *bucket) | |
1065 | { | |
df0933dc PM |
1066 | struct nf_conntrack_tuple_hash *h; |
1067 | struct nf_conn *ct; | |
9fb9cbb1 YK |
1068 | |
1069 | write_lock_bh(&nf_conntrack_lock); | |
1070 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { | |
df0933dc PM |
1071 | list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) { |
1072 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1073 | if (iter(ct, data)) | |
1074 | goto found; | |
1075 | } | |
601e68e1 | 1076 | } |
df0933dc PM |
1077 | list_for_each_entry(h, &unconfirmed, list) { |
1078 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1079 | if (iter(ct, data)) | |
ec68e97d | 1080 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 1081 | } |
c073e3fa | 1082 | write_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
1083 | return NULL; |
1084 | found: | |
c073e3fa | 1085 | atomic_inc(&ct->ct_general.use); |
9fb9cbb1 | 1086 | write_unlock_bh(&nf_conntrack_lock); |
df0933dc | 1087 | return ct; |
9fb9cbb1 YK |
1088 | } |
1089 | ||
1090 | void | |
1091 | nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) | |
1092 | { | |
df0933dc | 1093 | struct nf_conn *ct; |
9fb9cbb1 YK |
1094 | unsigned int bucket = 0; |
1095 | ||
df0933dc | 1096 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
1097 | /* Time to push up daises... */ |
1098 | if (del_timer(&ct->timeout)) | |
1099 | death_by_timeout((unsigned long)ct); | |
1100 | /* ... else the timer will get him soon. */ | |
1101 | ||
1102 | nf_ct_put(ct); | |
1103 | } | |
1104 | } | |
13b18339 | 1105 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 YK |
1106 | |
1107 | static int kill_all(struct nf_conn *i, void *data) | |
1108 | { | |
1109 | return 1; | |
1110 | } | |
1111 | ||
1112 | static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size) | |
1113 | { | |
1114 | if (vmalloced) | |
1115 | vfree(hash); | |
1116 | else | |
601e68e1 | 1117 | free_pages((unsigned long)hash, |
9fb9cbb1 YK |
1118 | get_order(sizeof(struct list_head) * size)); |
1119 | } | |
1120 | ||
272491ef | 1121 | void nf_conntrack_flush(void) |
c1d10adb PNA |
1122 | { |
1123 | nf_ct_iterate_cleanup(kill_all, NULL); | |
1124 | } | |
13b18339 | 1125 | EXPORT_SYMBOL_GPL(nf_conntrack_flush); |
c1d10adb | 1126 | |
9fb9cbb1 YK |
1127 | /* Mishearing the voices in his head, our hero wonders how he's |
1128 | supposed to kill the mall. */ | |
1129 | void nf_conntrack_cleanup(void) | |
1130 | { | |
1131 | int i; | |
1132 | ||
c3a47ab3 | 1133 | rcu_assign_pointer(ip_ct_attach, NULL); |
7d3cdc6b | 1134 | |
9fb9cbb1 YK |
1135 | /* This makes sure all current packets have passed through |
1136 | netfilter framework. Roll on, two-stage module | |
1137 | delete... */ | |
1138 | synchronize_net(); | |
1139 | ||
1140 | nf_ct_event_cache_flush(); | |
1141 | i_see_dead_people: | |
c1d10adb | 1142 | nf_conntrack_flush(); |
9fb9cbb1 YK |
1143 | if (atomic_read(&nf_conntrack_count) != 0) { |
1144 | schedule(); | |
1145 | goto i_see_dead_people; | |
1146 | } | |
6636568c PM |
1147 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1148 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1149 | schedule(); | |
9fb9cbb1 | 1150 | |
de6e05c4 YK |
1151 | rcu_assign_pointer(nf_ct_destroy, NULL); |
1152 | ||
9fb9cbb1 YK |
1153 | for (i = 0; i < NF_CT_F_NUM; i++) { |
1154 | if (nf_ct_cache[i].use == 0) | |
1155 | continue; | |
1156 | ||
1157 | NF_CT_ASSERT(nf_ct_cache[i].use == 1); | |
1158 | nf_ct_cache[i].use = 1; | |
1159 | nf_conntrack_unregister_cache(i); | |
1160 | } | |
1161 | kmem_cache_destroy(nf_conntrack_expect_cachep); | |
1162 | free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, | |
1163 | nf_conntrack_htable_size); | |
5a6f294e | 1164 | |
ac5357eb | 1165 | nf_conntrack_proto_fini(); |
ceceae1b | 1166 | nf_conntrack_helper_fini(); |
9fb9cbb1 YK |
1167 | } |
1168 | ||
1169 | static struct list_head *alloc_hashtable(int size, int *vmalloced) | |
1170 | { | |
1171 | struct list_head *hash; | |
1172 | unsigned int i; | |
1173 | ||
601e68e1 YH |
1174 | *vmalloced = 0; |
1175 | hash = (void*)__get_free_pages(GFP_KERNEL, | |
9fb9cbb1 YK |
1176 | get_order(sizeof(struct list_head) |
1177 | * size)); | |
601e68e1 | 1178 | if (!hash) { |
9fb9cbb1 YK |
1179 | *vmalloced = 1; |
1180 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
1181 | hash = vmalloc(sizeof(struct list_head) * size); | |
1182 | } | |
1183 | ||
1184 | if (hash) | |
601e68e1 | 1185 | for (i = 0; i < size; i++) |
9fb9cbb1 YK |
1186 | INIT_LIST_HEAD(&hash[i]); |
1187 | ||
1188 | return hash; | |
1189 | } | |
1190 | ||
1191 | int set_hashsize(const char *val, struct kernel_param *kp) | |
1192 | { | |
1193 | int i, bucket, hashsize, vmalloced; | |
1194 | int old_vmalloced, old_size; | |
1195 | int rnd; | |
1196 | struct list_head *hash, *old_hash; | |
1197 | struct nf_conntrack_tuple_hash *h; | |
1198 | ||
1199 | /* On boot, we can set this without any fancy locking. */ | |
1200 | if (!nf_conntrack_htable_size) | |
1201 | return param_set_uint(val, kp); | |
1202 | ||
1203 | hashsize = simple_strtol(val, NULL, 0); | |
1204 | if (!hashsize) | |
1205 | return -EINVAL; | |
1206 | ||
1207 | hash = alloc_hashtable(hashsize, &vmalloced); | |
1208 | if (!hash) | |
1209 | return -ENOMEM; | |
1210 | ||
1211 | /* We have to rehahs for the new table anyway, so we also can | |
1212 | * use a newrandom seed */ | |
1213 | get_random_bytes(&rnd, 4); | |
1214 | ||
1215 | write_lock_bh(&nf_conntrack_lock); | |
1216 | for (i = 0; i < nf_conntrack_htable_size; i++) { | |
1217 | while (!list_empty(&nf_conntrack_hash[i])) { | |
1218 | h = list_entry(nf_conntrack_hash[i].next, | |
1219 | struct nf_conntrack_tuple_hash, list); | |
1220 | list_del(&h->list); | |
1221 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); | |
1222 | list_add_tail(&h->list, &hash[bucket]); | |
1223 | } | |
1224 | } | |
1225 | old_size = nf_conntrack_htable_size; | |
1226 | old_vmalloced = nf_conntrack_vmalloc; | |
1227 | old_hash = nf_conntrack_hash; | |
1228 | ||
1229 | nf_conntrack_htable_size = hashsize; | |
1230 | nf_conntrack_vmalloc = vmalloced; | |
1231 | nf_conntrack_hash = hash; | |
1232 | nf_conntrack_hash_rnd = rnd; | |
1233 | write_unlock_bh(&nf_conntrack_lock); | |
1234 | ||
1235 | free_conntrack_hash(old_hash, old_vmalloced, old_size); | |
1236 | return 0; | |
1237 | } | |
1238 | ||
1239 | module_param_call(hashsize, set_hashsize, param_get_uint, | |
1240 | &nf_conntrack_htable_size, 0600); | |
1241 | ||
1242 | int __init nf_conntrack_init(void) | |
1243 | { | |
9fb9cbb1 YK |
1244 | int ret; |
1245 | ||
1246 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
1247 | * machine has 256 buckets. >= 1GB machines have 8192 buckets. */ | |
1248 | if (!nf_conntrack_htable_size) { | |
1249 | nf_conntrack_htable_size | |
1250 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
1251 | / sizeof(struct list_head)); | |
1252 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
1253 | nf_conntrack_htable_size = 8192; | |
1254 | if (nf_conntrack_htable_size < 16) | |
1255 | nf_conntrack_htable_size = 16; | |
1256 | } | |
1257 | nf_conntrack_max = 8 * nf_conntrack_htable_size; | |
1258 | ||
1259 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1260 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1261 | nf_conntrack_max); | |
1262 | ||
1263 | nf_conntrack_hash = alloc_hashtable(nf_conntrack_htable_size, | |
1264 | &nf_conntrack_vmalloc); | |
1265 | if (!nf_conntrack_hash) { | |
1266 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1267 | goto err_out; | |
1268 | } | |
1269 | ||
1270 | ret = nf_conntrack_register_cache(NF_CT_F_BASIC, "nf_conntrack:basic", | |
dc808fe2 | 1271 | sizeof(struct nf_conn)); |
9fb9cbb1 YK |
1272 | if (ret < 0) { |
1273 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); | |
1274 | goto err_free_hash; | |
1275 | } | |
1276 | ||
1277 | nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect", | |
1278 | sizeof(struct nf_conntrack_expect), | |
1279 | 0, 0, NULL, NULL); | |
1280 | if (!nf_conntrack_expect_cachep) { | |
1281 | printk(KERN_ERR "Unable to create nf_expect slab cache\n"); | |
1282 | goto err_free_conntrack_slab; | |
1283 | } | |
1284 | ||
ac5357eb | 1285 | ret = nf_conntrack_proto_init(); |
933a41e7 PM |
1286 | if (ret < 0) |
1287 | goto out_free_expect_slab; | |
1288 | ||
ceceae1b YK |
1289 | ret = nf_conntrack_helper_init(); |
1290 | if (ret < 0) | |
1291 | goto out_fini_proto; | |
1292 | ||
7d3cdc6b | 1293 | /* For use by REJECT target */ |
c3a47ab3 | 1294 | rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach); |
de6e05c4 | 1295 | rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); |
7d3cdc6b | 1296 | |
9fb9cbb1 YK |
1297 | /* Set up fake conntrack: |
1298 | - to never be deleted, not in any hashes */ | |
1299 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); | |
1300 | /* - and look it like as a confirmed connection */ | |
1301 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1302 | ||
1303 | return ret; | |
1304 | ||
ceceae1b YK |
1305 | out_fini_proto: |
1306 | nf_conntrack_proto_fini(); | |
933a41e7 PM |
1307 | out_free_expect_slab: |
1308 | kmem_cache_destroy(nf_conntrack_expect_cachep); | |
9fb9cbb1 YK |
1309 | err_free_conntrack_slab: |
1310 | nf_conntrack_unregister_cache(NF_CT_F_BASIC); | |
1311 | err_free_hash: | |
1312 | free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, | |
1313 | nf_conntrack_htable_size); | |
1314 | err_out: | |
1315 | return -ENOMEM; | |
1316 | } |