]>
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. | |
12 | * | |
13 | * 23 Apr 2001: Harald Welte <laforge@gnumonks.org> | |
14 | * - new API and handling of conntrack/nat helpers | |
15 | * - now capable of multiple expectations for one master | |
16 | * 16 Jul 2002: Harald Welte <laforge@gnumonks.org> | |
17 | * - add usage/reference counts to ip_conntrack_expect | |
18 | * - export ip_conntrack[_expect]_{find_get,put} functions | |
19 | * 16 Dec 2003: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp> | |
20 | * - generalize L3 protocol denendent part. | |
21 | * 23 Mar 2004: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp> | |
22 | * - add support various size of conntrack structures. | |
dc808fe2 HW |
23 | * 26 Jan 2006: Harald Welte <laforge@netfilter.org> |
24 | * - restructure nf_conn (introduce nf_conn_help) | |
25 | * - redesign 'features' how they were originally intended | |
b9f78f9f PNA |
26 | * 26 Feb 2006: Pablo Neira Ayuso <pablo@eurodev.net> |
27 | * - add support for L3 protocol module load on demand. | |
9fb9cbb1 YK |
28 | * |
29 | * Derived from net/ipv4/netfilter/ip_conntrack_core.c | |
30 | */ | |
31 | ||
9fb9cbb1 YK |
32 | #include <linux/types.h> |
33 | #include <linux/netfilter.h> | |
34 | #include <linux/module.h> | |
35 | #include <linux/skbuff.h> | |
36 | #include <linux/proc_fs.h> | |
37 | #include <linux/vmalloc.h> | |
38 | #include <linux/stddef.h> | |
39 | #include <linux/slab.h> | |
40 | #include <linux/random.h> | |
41 | #include <linux/jhash.h> | |
42 | #include <linux/err.h> | |
43 | #include <linux/percpu.h> | |
44 | #include <linux/moduleparam.h> | |
45 | #include <linux/notifier.h> | |
46 | #include <linux/kernel.h> | |
47 | #include <linux/netdevice.h> | |
48 | #include <linux/socket.h> | |
49 | ||
9fb9cbb1 YK |
50 | #include <net/netfilter/nf_conntrack.h> |
51 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
605dcad6 | 52 | #include <net/netfilter/nf_conntrack_l4proto.h> |
77ab9cff | 53 | #include <net/netfilter/nf_conntrack_expect.h> |
9fb9cbb1 YK |
54 | #include <net/netfilter/nf_conntrack_helper.h> |
55 | #include <net/netfilter/nf_conntrack_core.h> | |
9fb9cbb1 | 56 | |
dc808fe2 | 57 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 YK |
58 | |
59 | #if 0 | |
60 | #define DEBUGP printk | |
61 | #else | |
62 | #define DEBUGP(format, args...) | |
63 | #endif | |
64 | ||
65 | DEFINE_RWLOCK(nf_conntrack_lock); | |
66 | ||
67 | /* nf_conntrack_standalone needs this */ | |
68 | atomic_t nf_conntrack_count = ATOMIC_INIT(0); | |
a999e683 | 69 | EXPORT_SYMBOL_GPL(nf_conntrack_count); |
9fb9cbb1 YK |
70 | |
71 | void (*nf_conntrack_destroyed)(struct nf_conn *conntrack) = NULL; | |
e2b7606c | 72 | unsigned int nf_conntrack_htable_size __read_mostly; |
94aec08e | 73 | int nf_conntrack_max __read_mostly; |
a999e683 | 74 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
1192e403 | 75 | struct list_head *nf_conntrack_hash __read_mostly; |
e2b7606c | 76 | struct nf_conn nf_conntrack_untracked __read_mostly; |
94aec08e | 77 | unsigned int nf_ct_log_invalid __read_mostly; |
7e5d03bb | 78 | LIST_HEAD(unconfirmed); |
1192e403 | 79 | static int nf_conntrack_vmalloc __read_mostly; |
9fb9cbb1 | 80 | |
4e3882f7 | 81 | static unsigned int nf_conntrack_next_id; |
77ab9cff | 82 | |
9fb9cbb1 YK |
83 | DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); |
84 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); | |
85 | ||
86 | /* | |
87 | * This scheme offers various size of "struct nf_conn" dependent on | |
88 | * features(helper, nat, ...) | |
89 | */ | |
90 | ||
91 | #define NF_CT_FEATURES_NAMELEN 256 | |
92 | static struct { | |
93 | /* name of slab cache. printed in /proc/slabinfo */ | |
94 | char *name; | |
95 | ||
96 | /* size of slab cache */ | |
97 | size_t size; | |
98 | ||
99 | /* slab cache pointer */ | |
100 | kmem_cache_t *cachep; | |
101 | ||
102 | /* allocated slab cache + modules which uses this slab cache */ | |
103 | int use; | |
104 | ||
9fb9cbb1 YK |
105 | } nf_ct_cache[NF_CT_F_NUM]; |
106 | ||
107 | /* protect members of nf_ct_cache except of "use" */ | |
108 | DEFINE_RWLOCK(nf_ct_cache_lock); | |
109 | ||
110 | /* This avoids calling kmem_cache_create() with same name simultaneously */ | |
57b47a53 | 111 | static DEFINE_MUTEX(nf_ct_cache_mutex); |
9fb9cbb1 | 112 | |
9fb9cbb1 YK |
113 | static int nf_conntrack_hash_rnd_initted; |
114 | static unsigned int nf_conntrack_hash_rnd; | |
115 | ||
116 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, | |
117 | unsigned int size, unsigned int rnd) | |
118 | { | |
119 | unsigned int a, b; | |
120 | a = jhash((void *)tuple->src.u3.all, sizeof(tuple->src.u3.all), | |
121 | ((tuple->src.l3num) << 16) | tuple->dst.protonum); | |
122 | b = jhash((void *)tuple->dst.u3.all, sizeof(tuple->dst.u3.all), | |
123 | (tuple->src.u.all << 16) | tuple->dst.u.all); | |
124 | ||
125 | return jhash_2words(a, b, rnd) % size; | |
126 | } | |
127 | ||
128 | static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) | |
129 | { | |
130 | return __hash_conntrack(tuple, nf_conntrack_htable_size, | |
131 | nf_conntrack_hash_rnd); | |
132 | } | |
133 | ||
9fb9cbb1 | 134 | int nf_conntrack_register_cache(u_int32_t features, const char *name, |
dc808fe2 | 135 | size_t size) |
9fb9cbb1 YK |
136 | { |
137 | int ret = 0; | |
138 | char *cache_name; | |
139 | kmem_cache_t *cachep; | |
140 | ||
141 | DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n", | |
142 | features, name, size); | |
143 | ||
144 | if (features < NF_CT_F_BASIC || features >= NF_CT_F_NUM) { | |
145 | DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n", | |
146 | features); | |
147 | return -EINVAL; | |
148 | } | |
149 | ||
57b47a53 | 150 | mutex_lock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
151 | |
152 | write_lock_bh(&nf_ct_cache_lock); | |
153 | /* e.g: multiple helpers are loaded */ | |
154 | if (nf_ct_cache[features].use > 0) { | |
155 | DEBUGP("nf_conntrack_register_cache: already resisterd.\n"); | |
156 | if ((!strncmp(nf_ct_cache[features].name, name, | |
157 | NF_CT_FEATURES_NAMELEN)) | |
dc808fe2 | 158 | && nf_ct_cache[features].size == size) { |
9fb9cbb1 YK |
159 | DEBUGP("nf_conntrack_register_cache: reusing.\n"); |
160 | nf_ct_cache[features].use++; | |
161 | ret = 0; | |
162 | } else | |
163 | ret = -EBUSY; | |
164 | ||
165 | write_unlock_bh(&nf_ct_cache_lock); | |
57b47a53 | 166 | mutex_unlock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
167 | return ret; |
168 | } | |
169 | write_unlock_bh(&nf_ct_cache_lock); | |
170 | ||
171 | /* | |
172 | * The memory space for name of slab cache must be alive until | |
173 | * cache is destroyed. | |
174 | */ | |
175 | cache_name = kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN, GFP_ATOMIC); | |
176 | if (cache_name == NULL) { | |
177 | DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n"); | |
178 | ret = -ENOMEM; | |
179 | goto out_up_mutex; | |
180 | } | |
181 | ||
182 | if (strlcpy(cache_name, name, NF_CT_FEATURES_NAMELEN) | |
183 | >= NF_CT_FEATURES_NAMELEN) { | |
184 | printk("nf_conntrack_register_cache: name too long\n"); | |
185 | ret = -EINVAL; | |
186 | goto out_free_name; | |
187 | } | |
188 | ||
189 | cachep = kmem_cache_create(cache_name, size, 0, 0, | |
190 | NULL, NULL); | |
191 | if (!cachep) { | |
192 | printk("nf_conntrack_register_cache: Can't create slab cache " | |
193 | "for the features = 0x%x\n", features); | |
194 | ret = -ENOMEM; | |
195 | goto out_free_name; | |
196 | } | |
197 | ||
198 | write_lock_bh(&nf_ct_cache_lock); | |
199 | nf_ct_cache[features].use = 1; | |
200 | nf_ct_cache[features].size = size; | |
9fb9cbb1 YK |
201 | nf_ct_cache[features].cachep = cachep; |
202 | nf_ct_cache[features].name = cache_name; | |
203 | write_unlock_bh(&nf_ct_cache_lock); | |
204 | ||
205 | goto out_up_mutex; | |
206 | ||
207 | out_free_name: | |
208 | kfree(cache_name); | |
209 | out_up_mutex: | |
57b47a53 | 210 | mutex_unlock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
211 | return ret; |
212 | } | |
213 | ||
214 | /* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */ | |
215 | void nf_conntrack_unregister_cache(u_int32_t features) | |
216 | { | |
217 | kmem_cache_t *cachep; | |
218 | char *name; | |
219 | ||
220 | /* | |
221 | * This assures that kmem_cache_create() isn't called before destroying | |
222 | * slab cache. | |
223 | */ | |
224 | DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features); | |
57b47a53 | 225 | mutex_lock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
226 | |
227 | write_lock_bh(&nf_ct_cache_lock); | |
228 | if (--nf_ct_cache[features].use > 0) { | |
229 | write_unlock_bh(&nf_ct_cache_lock); | |
57b47a53 | 230 | mutex_unlock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
231 | return; |
232 | } | |
233 | cachep = nf_ct_cache[features].cachep; | |
234 | name = nf_ct_cache[features].name; | |
235 | nf_ct_cache[features].cachep = NULL; | |
236 | nf_ct_cache[features].name = NULL; | |
9fb9cbb1 YK |
237 | nf_ct_cache[features].size = 0; |
238 | write_unlock_bh(&nf_ct_cache_lock); | |
239 | ||
240 | synchronize_net(); | |
241 | ||
242 | kmem_cache_destroy(cachep); | |
243 | kfree(name); | |
244 | ||
57b47a53 | 245 | mutex_unlock(&nf_ct_cache_mutex); |
9fb9cbb1 YK |
246 | } |
247 | ||
248 | int | |
249 | nf_ct_get_tuple(const struct sk_buff *skb, | |
250 | unsigned int nhoff, | |
251 | unsigned int dataoff, | |
252 | u_int16_t l3num, | |
253 | u_int8_t protonum, | |
254 | struct nf_conntrack_tuple *tuple, | |
255 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 256 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 YK |
257 | { |
258 | NF_CT_TUPLE_U_BLANK(tuple); | |
259 | ||
260 | tuple->src.l3num = l3num; | |
261 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
262 | return 0; | |
263 | ||
264 | tuple->dst.protonum = protonum; | |
265 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
266 | ||
605dcad6 | 267 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 YK |
268 | } |
269 | ||
270 | int | |
271 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, | |
272 | const struct nf_conntrack_tuple *orig, | |
273 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 274 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 YK |
275 | { |
276 | NF_CT_TUPLE_U_BLANK(inverse); | |
277 | ||
278 | inverse->src.l3num = orig->src.l3num; | |
279 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
280 | return 0; | |
281 | ||
282 | inverse->dst.dir = !orig->dst.dir; | |
283 | ||
284 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 285 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 YK |
286 | } |
287 | ||
9fb9cbb1 YK |
288 | static void |
289 | clean_from_lists(struct nf_conn *ct) | |
290 | { | |
9fb9cbb1 | 291 | DEBUGP("clean_from_lists(%p)\n", ct); |
df0933dc PM |
292 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); |
293 | list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list); | |
9fb9cbb1 YK |
294 | |
295 | /* Destroy all pending expectations */ | |
c1d10adb | 296 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
297 | } |
298 | ||
299 | static void | |
300 | destroy_conntrack(struct nf_conntrack *nfct) | |
301 | { | |
302 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
f09943fe | 303 | struct nf_conn_help *help = nfct_help(ct); |
9fb9cbb1 | 304 | struct nf_conntrack_l3proto *l3proto; |
605dcad6 | 305 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
306 | |
307 | DEBUGP("destroy_conntrack(%p)\n", ct); | |
308 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); | |
309 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
310 | ||
311 | nf_conntrack_event(IPCT_DESTROY, ct); | |
312 | set_bit(IPS_DYING_BIT, &ct->status); | |
313 | ||
f09943fe PM |
314 | if (help && help->helper && help->helper->destroy) |
315 | help->helper->destroy(ct); | |
316 | ||
9fb9cbb1 YK |
317 | /* To make sure we don't get any weird locking issues here: |
318 | * destroy_conntrack() MUST NOT be called with a write lock | |
319 | * to nf_conntrack_lock!!! -HW */ | |
c1d10adb | 320 | l3proto = __nf_ct_l3proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num); |
9fb9cbb1 YK |
321 | if (l3proto && l3proto->destroy) |
322 | l3proto->destroy(ct); | |
323 | ||
605dcad6 MJ |
324 | l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); |
325 | if (l4proto && l4proto->destroy) | |
326 | l4proto->destroy(ct); | |
9fb9cbb1 YK |
327 | |
328 | if (nf_conntrack_destroyed) | |
329 | nf_conntrack_destroyed(ct); | |
330 | ||
331 | write_lock_bh(&nf_conntrack_lock); | |
332 | /* Expectations will have been removed in clean_from_lists, | |
333 | * except TFTP can create an expectation on the first packet, | |
334 | * before connection is in the list, so we need to clean here, | |
335 | * too. */ | |
c1d10adb | 336 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
337 | |
338 | /* We overload first tuple to link into unconfirmed list. */ | |
339 | if (!nf_ct_is_confirmed(ct)) { | |
340 | BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list)); | |
341 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); | |
342 | } | |
343 | ||
344 | NF_CT_STAT_INC(delete); | |
345 | write_unlock_bh(&nf_conntrack_lock); | |
346 | ||
347 | if (ct->master) | |
348 | nf_ct_put(ct->master); | |
349 | ||
350 | DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct); | |
351 | nf_conntrack_free(ct); | |
352 | } | |
353 | ||
354 | static void death_by_timeout(unsigned long ul_conntrack) | |
355 | { | |
356 | struct nf_conn *ct = (void *)ul_conntrack; | |
357 | ||
358 | write_lock_bh(&nf_conntrack_lock); | |
359 | /* Inside lock so preempt is disabled on module removal path. | |
360 | * Otherwise we can get spurious warnings. */ | |
361 | NF_CT_STAT_INC(delete_list); | |
362 | clean_from_lists(ct); | |
363 | write_unlock_bh(&nf_conntrack_lock); | |
364 | nf_ct_put(ct); | |
365 | } | |
366 | ||
c1d10adb | 367 | struct nf_conntrack_tuple_hash * |
9fb9cbb1 YK |
368 | __nf_conntrack_find(const struct nf_conntrack_tuple *tuple, |
369 | const struct nf_conn *ignored_conntrack) | |
370 | { | |
371 | struct nf_conntrack_tuple_hash *h; | |
372 | unsigned int hash = hash_conntrack(tuple); | |
373 | ||
9fb9cbb1 | 374 | list_for_each_entry(h, &nf_conntrack_hash[hash], list) { |
df0933dc PM |
375 | if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && |
376 | nf_ct_tuple_equal(tuple, &h->tuple)) { | |
9fb9cbb1 YK |
377 | NF_CT_STAT_INC(found); |
378 | return h; | |
379 | } | |
380 | NF_CT_STAT_INC(searched); | |
381 | } | |
382 | ||
383 | return NULL; | |
384 | } | |
385 | ||
386 | /* Find a connection corresponding to a tuple. */ | |
387 | struct nf_conntrack_tuple_hash * | |
388 | nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple, | |
389 | const struct nf_conn *ignored_conntrack) | |
390 | { | |
391 | struct nf_conntrack_tuple_hash *h; | |
392 | ||
393 | read_lock_bh(&nf_conntrack_lock); | |
394 | h = __nf_conntrack_find(tuple, ignored_conntrack); | |
395 | if (h) | |
396 | atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use); | |
397 | read_unlock_bh(&nf_conntrack_lock); | |
398 | ||
399 | return h; | |
400 | } | |
401 | ||
c1d10adb PNA |
402 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
403 | unsigned int hash, | |
404 | unsigned int repl_hash) | |
405 | { | |
406 | ct->id = ++nf_conntrack_next_id; | |
df0933dc PM |
407 | list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list, |
408 | &nf_conntrack_hash[hash]); | |
409 | list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list, | |
410 | &nf_conntrack_hash[repl_hash]); | |
c1d10adb PNA |
411 | } |
412 | ||
413 | void nf_conntrack_hash_insert(struct nf_conn *ct) | |
414 | { | |
415 | unsigned int hash, repl_hash; | |
416 | ||
417 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
418 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
419 | ||
420 | write_lock_bh(&nf_conntrack_lock); | |
421 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
422 | write_unlock_bh(&nf_conntrack_lock); | |
423 | } | |
424 | ||
9fb9cbb1 YK |
425 | /* Confirm a connection given skb; places it in hash table */ |
426 | int | |
427 | __nf_conntrack_confirm(struct sk_buff **pskb) | |
428 | { | |
429 | unsigned int hash, repl_hash; | |
df0933dc | 430 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 431 | struct nf_conn *ct; |
df0933dc | 432 | struct nf_conn_help *help; |
9fb9cbb1 YK |
433 | enum ip_conntrack_info ctinfo; |
434 | ||
435 | ct = nf_ct_get(*pskb, &ctinfo); | |
436 | ||
437 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
438 | ICMP/TCP RST packets in other direction. Actual packet | |
439 | which created connection will be IP_CT_NEW or for an | |
440 | expected connection, IP_CT_RELATED. */ | |
441 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
442 | return NF_ACCEPT; | |
443 | ||
444 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
445 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
446 | ||
447 | /* We're not in hash table, and we refuse to set up related | |
448 | connections for unconfirmed conns. But packet copies and | |
449 | REJECT will give spurious warnings here. */ | |
450 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
451 | ||
452 | /* No external references means noone else could have | |
453 | confirmed us. */ | |
454 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
455 | DEBUGP("Confirming conntrack %p\n", ct); | |
456 | ||
457 | write_lock_bh(&nf_conntrack_lock); | |
458 | ||
459 | /* See if there's one in the list already, including reverse: | |
460 | NAT could have grabbed it without realizing, since we're | |
461 | not in the hash. If there is, we lost race. */ | |
df0933dc PM |
462 | list_for_each_entry(h, &nf_conntrack_hash[hash], list) |
463 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, | |
464 | &h->tuple)) | |
465 | goto out; | |
466 | list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list) | |
467 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, | |
468 | &h->tuple)) | |
469 | goto out; | |
9fb9cbb1 | 470 | |
df0933dc PM |
471 | /* Remove from unconfirmed list */ |
472 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); | |
473 | ||
474 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
475 | /* Timer relative to confirmation time, not original | |
476 | setting time, otherwise we'd get timer wrap in | |
477 | weird delay cases. */ | |
478 | ct->timeout.expires += jiffies; | |
479 | add_timer(&ct->timeout); | |
480 | atomic_inc(&ct->ct_general.use); | |
481 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
482 | NF_CT_STAT_INC(insert); | |
483 | write_unlock_bh(&nf_conntrack_lock); | |
484 | help = nfct_help(ct); | |
485 | if (help && help->helper) | |
486 | nf_conntrack_event_cache(IPCT_HELPER, *pskb); | |
9fb9cbb1 | 487 | #ifdef CONFIG_NF_NAT_NEEDED |
df0933dc PM |
488 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || |
489 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
490 | nf_conntrack_event_cache(IPCT_NATINFO, *pskb); | |
9fb9cbb1 | 491 | #endif |
df0933dc PM |
492 | nf_conntrack_event_cache(master_ct(ct) ? |
493 | IPCT_RELATED : IPCT_NEW, *pskb); | |
494 | return NF_ACCEPT; | |
9fb9cbb1 | 495 | |
df0933dc | 496 | out: |
9fb9cbb1 YK |
497 | NF_CT_STAT_INC(insert_failed); |
498 | write_unlock_bh(&nf_conntrack_lock); | |
499 | return NF_DROP; | |
500 | } | |
501 | ||
502 | /* Returns true if a connection correspondings to the tuple (required | |
503 | for NAT). */ | |
504 | int | |
505 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
506 | const struct nf_conn *ignored_conntrack) | |
507 | { | |
508 | struct nf_conntrack_tuple_hash *h; | |
509 | ||
510 | read_lock_bh(&nf_conntrack_lock); | |
511 | h = __nf_conntrack_find(tuple, ignored_conntrack); | |
512 | read_unlock_bh(&nf_conntrack_lock); | |
513 | ||
514 | return h != NULL; | |
515 | } | |
516 | ||
517 | /* There's a small race here where we may free a just-assured | |
518 | connection. Too bad: we're in trouble anyway. */ | |
9fb9cbb1 YK |
519 | static int early_drop(struct list_head *chain) |
520 | { | |
521 | /* Traverse backwards: gives us oldest, which is roughly LRU */ | |
522 | struct nf_conntrack_tuple_hash *h; | |
df0933dc | 523 | struct nf_conn *ct = NULL, *tmp; |
9fb9cbb1 YK |
524 | int dropped = 0; |
525 | ||
526 | read_lock_bh(&nf_conntrack_lock); | |
df0933dc PM |
527 | list_for_each_entry_reverse(h, chain, list) { |
528 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
529 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) { | |
530 | ct = tmp; | |
531 | atomic_inc(&ct->ct_general.use); | |
532 | break; | |
533 | } | |
9fb9cbb1 YK |
534 | } |
535 | read_unlock_bh(&nf_conntrack_lock); | |
536 | ||
537 | if (!ct) | |
538 | return dropped; | |
539 | ||
540 | if (del_timer(&ct->timeout)) { | |
541 | death_by_timeout((unsigned long)ct); | |
542 | dropped = 1; | |
543 | NF_CT_STAT_INC(early_drop); | |
544 | } | |
545 | nf_ct_put(ct); | |
546 | return dropped; | |
547 | } | |
548 | ||
9fb9cbb1 YK |
549 | static struct nf_conn * |
550 | __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, | |
551 | const struct nf_conntrack_tuple *repl, | |
9457d851 PM |
552 | const struct nf_conntrack_l3proto *l3proto, |
553 | u_int32_t features) | |
9fb9cbb1 YK |
554 | { |
555 | struct nf_conn *conntrack = NULL; | |
dc808fe2 | 556 | struct nf_conntrack_helper *helper; |
9fb9cbb1 | 557 | |
dc808fe2 | 558 | if (unlikely(!nf_conntrack_hash_rnd_initted)) { |
9fb9cbb1 YK |
559 | get_random_bytes(&nf_conntrack_hash_rnd, 4); |
560 | nf_conntrack_hash_rnd_initted = 1; | |
561 | } | |
562 | ||
5251e2d2 PNA |
563 | /* We don't want any race condition at early drop stage */ |
564 | atomic_inc(&nf_conntrack_count); | |
565 | ||
9fb9cbb1 | 566 | if (nf_conntrack_max |
5251e2d2 | 567 | && atomic_read(&nf_conntrack_count) > nf_conntrack_max) { |
9fb9cbb1 YK |
568 | unsigned int hash = hash_conntrack(orig); |
569 | /* Try dropping from this hash chain. */ | |
570 | if (!early_drop(&nf_conntrack_hash[hash])) { | |
5251e2d2 | 571 | atomic_dec(&nf_conntrack_count); |
9fb9cbb1 YK |
572 | if (net_ratelimit()) |
573 | printk(KERN_WARNING | |
574 | "nf_conntrack: table full, dropping" | |
575 | " packet.\n"); | |
576 | return ERR_PTR(-ENOMEM); | |
577 | } | |
578 | } | |
579 | ||
580 | /* find features needed by this conntrack. */ | |
9457d851 | 581 | features |= l3proto->get_features(orig); |
dc808fe2 HW |
582 | |
583 | /* FIXME: protect helper list per RCU */ | |
9fb9cbb1 | 584 | read_lock_bh(&nf_conntrack_lock); |
dc808fe2 | 585 | helper = __nf_ct_helper_find(repl); |
5b1158e9 JK |
586 | /* NAT might want to assign a helper later */ |
587 | if (helper || features & NF_CT_F_NAT) | |
9fb9cbb1 YK |
588 | features |= NF_CT_F_HELP; |
589 | read_unlock_bh(&nf_conntrack_lock); | |
590 | ||
591 | DEBUGP("nf_conntrack_alloc: features=0x%x\n", features); | |
592 | ||
593 | read_lock_bh(&nf_ct_cache_lock); | |
594 | ||
dc808fe2 | 595 | if (unlikely(!nf_ct_cache[features].use)) { |
9fb9cbb1 YK |
596 | DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n", |
597 | features); | |
598 | goto out; | |
599 | } | |
600 | ||
601 | conntrack = kmem_cache_alloc(nf_ct_cache[features].cachep, GFP_ATOMIC); | |
602 | if (conntrack == NULL) { | |
603 | DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n"); | |
604 | goto out; | |
605 | } | |
606 | ||
607 | memset(conntrack, 0, nf_ct_cache[features].size); | |
608 | conntrack->features = features; | |
9fb9cbb1 YK |
609 | atomic_set(&conntrack->ct_general.use, 1); |
610 | conntrack->ct_general.destroy = destroy_conntrack; | |
611 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; | |
612 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
613 | /* Don't set timer yet: wait for confirmation */ | |
614 | init_timer(&conntrack->timeout); | |
615 | conntrack->timeout.data = (unsigned long)conntrack; | |
616 | conntrack->timeout.function = death_by_timeout; | |
5251e2d2 | 617 | read_unlock_bh(&nf_ct_cache_lock); |
9fb9cbb1 | 618 | |
5251e2d2 | 619 | return conntrack; |
9fb9cbb1 YK |
620 | out: |
621 | read_unlock_bh(&nf_ct_cache_lock); | |
5251e2d2 | 622 | atomic_dec(&nf_conntrack_count); |
9fb9cbb1 YK |
623 | return conntrack; |
624 | } | |
625 | ||
626 | struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, | |
627 | const struct nf_conntrack_tuple *repl) | |
628 | { | |
629 | struct nf_conntrack_l3proto *l3proto; | |
630 | ||
c1d10adb | 631 | l3proto = __nf_ct_l3proto_find(orig->src.l3num); |
9457d851 | 632 | return __nf_conntrack_alloc(orig, repl, l3proto, 0); |
9fb9cbb1 YK |
633 | } |
634 | ||
635 | void nf_conntrack_free(struct nf_conn *conntrack) | |
636 | { | |
637 | u_int32_t features = conntrack->features; | |
638 | NF_CT_ASSERT(features >= NF_CT_F_BASIC && features < NF_CT_F_NUM); | |
639 | DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features, | |
640 | conntrack); | |
641 | kmem_cache_free(nf_ct_cache[features].cachep, conntrack); | |
642 | atomic_dec(&nf_conntrack_count); | |
643 | } | |
644 | ||
645 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
646 | failed due to stress. Otherwise it really is unclassifiable. */ | |
647 | static struct nf_conntrack_tuple_hash * | |
648 | init_conntrack(const struct nf_conntrack_tuple *tuple, | |
649 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 650 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
651 | struct sk_buff *skb, |
652 | unsigned int dataoff) | |
653 | { | |
654 | struct nf_conn *conntrack; | |
655 | struct nf_conntrack_tuple repl_tuple; | |
656 | struct nf_conntrack_expect *exp; | |
9457d851 | 657 | u_int32_t features = 0; |
9fb9cbb1 | 658 | |
605dcad6 | 659 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
9fb9cbb1 YK |
660 | DEBUGP("Can't invert tuple.\n"); |
661 | return NULL; | |
662 | } | |
663 | ||
9457d851 PM |
664 | read_lock_bh(&nf_conntrack_lock); |
665 | exp = __nf_conntrack_expect_find(tuple); | |
666 | if (exp && exp->helper) | |
667 | features = NF_CT_F_HELP; | |
668 | read_unlock_bh(&nf_conntrack_lock); | |
669 | ||
670 | conntrack = __nf_conntrack_alloc(tuple, &repl_tuple, l3proto, features); | |
9fb9cbb1 YK |
671 | if (conntrack == NULL || IS_ERR(conntrack)) { |
672 | DEBUGP("Can't allocate conntrack.\n"); | |
673 | return (struct nf_conntrack_tuple_hash *)conntrack; | |
674 | } | |
675 | ||
605dcad6 | 676 | if (!l4proto->new(conntrack, skb, dataoff)) { |
9fb9cbb1 YK |
677 | nf_conntrack_free(conntrack); |
678 | DEBUGP("init conntrack: can't track with proto module\n"); | |
679 | return NULL; | |
680 | } | |
681 | ||
682 | write_lock_bh(&nf_conntrack_lock); | |
683 | exp = find_expectation(tuple); | |
684 | ||
685 | if (exp) { | |
686 | DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n", | |
687 | conntrack, exp); | |
688 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
689 | __set_bit(IPS_EXPECTED_BIT, &conntrack->status); | |
690 | conntrack->master = exp->master; | |
9457d851 PM |
691 | if (exp->helper) |
692 | nfct_help(conntrack)->helper = exp->helper; | |
9fb9cbb1 YK |
693 | #ifdef CONFIG_NF_CONNTRACK_MARK |
694 | conntrack->mark = exp->master->mark; | |
7c9728c3 JM |
695 | #endif |
696 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
697 | conntrack->secmark = exp->master->secmark; | |
9fb9cbb1 YK |
698 | #endif |
699 | nf_conntrack_get(&conntrack->master->ct_general); | |
700 | NF_CT_STAT_INC(expect_new); | |
22e7410b YK |
701 | } else { |
702 | struct nf_conn_help *help = nfct_help(conntrack); | |
703 | ||
704 | if (help) | |
705 | help->helper = __nf_ct_helper_find(&repl_tuple); | |
9fb9cbb1 | 706 | NF_CT_STAT_INC(new); |
22e7410b | 707 | } |
9fb9cbb1 YK |
708 | |
709 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
710 | list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed); | |
711 | ||
712 | write_unlock_bh(&nf_conntrack_lock); | |
713 | ||
714 | if (exp) { | |
715 | if (exp->expectfn) | |
716 | exp->expectfn(conntrack, exp); | |
717 | nf_conntrack_expect_put(exp); | |
718 | } | |
719 | ||
720 | return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; | |
721 | } | |
722 | ||
723 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
724 | static inline struct nf_conn * | |
725 | resolve_normal_ct(struct sk_buff *skb, | |
726 | unsigned int dataoff, | |
727 | u_int16_t l3num, | |
728 | u_int8_t protonum, | |
729 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 730 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 YK |
731 | int *set_reply, |
732 | enum ip_conntrack_info *ctinfo) | |
733 | { | |
734 | struct nf_conntrack_tuple tuple; | |
735 | struct nf_conntrack_tuple_hash *h; | |
736 | struct nf_conn *ct; | |
737 | ||
738 | if (!nf_ct_get_tuple(skb, (unsigned int)(skb->nh.raw - skb->data), | |
739 | dataoff, l3num, protonum, &tuple, l3proto, | |
605dcad6 | 740 | l4proto)) { |
9fb9cbb1 YK |
741 | DEBUGP("resolve_normal_ct: Can't get tuple\n"); |
742 | return NULL; | |
743 | } | |
744 | ||
745 | /* look for tuple match */ | |
746 | h = nf_conntrack_find_get(&tuple, NULL); | |
747 | if (!h) { | |
605dcad6 | 748 | h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); |
9fb9cbb1 YK |
749 | if (!h) |
750 | return NULL; | |
751 | if (IS_ERR(h)) | |
752 | return (void *)h; | |
753 | } | |
754 | ct = nf_ct_tuplehash_to_ctrack(h); | |
755 | ||
756 | /* It exists; we have (non-exclusive) reference. */ | |
757 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
758 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
759 | /* Please set reply bit if this packet OK */ | |
760 | *set_reply = 1; | |
761 | } else { | |
762 | /* Once we've had two way comms, always ESTABLISHED. */ | |
763 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
764 | DEBUGP("nf_conntrack_in: normal packet for %p\n", ct); | |
765 | *ctinfo = IP_CT_ESTABLISHED; | |
766 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
767 | DEBUGP("nf_conntrack_in: related packet for %p\n", ct); | |
768 | *ctinfo = IP_CT_RELATED; | |
769 | } else { | |
770 | DEBUGP("nf_conntrack_in: new packet for %p\n", ct); | |
771 | *ctinfo = IP_CT_NEW; | |
772 | } | |
773 | *set_reply = 0; | |
774 | } | |
775 | skb->nfct = &ct->ct_general; | |
776 | skb->nfctinfo = *ctinfo; | |
777 | return ct; | |
778 | } | |
779 | ||
780 | unsigned int | |
781 | nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb) | |
782 | { | |
783 | struct nf_conn *ct; | |
784 | enum ip_conntrack_info ctinfo; | |
785 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 786 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 YK |
787 | unsigned int dataoff; |
788 | u_int8_t protonum; | |
789 | int set_reply = 0; | |
790 | int ret; | |
791 | ||
792 | /* Previously seen (loopback or untracked)? Ignore. */ | |
793 | if ((*pskb)->nfct) { | |
794 | NF_CT_STAT_INC(ignore); | |
795 | return NF_ACCEPT; | |
796 | } | |
797 | ||
c1d10adb | 798 | l3proto = __nf_ct_l3proto_find((u_int16_t)pf); |
9fb9cbb1 YK |
799 | if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) { |
800 | DEBUGP("not prepared to track yet or error occured\n"); | |
801 | return -ret; | |
802 | } | |
803 | ||
605dcad6 | 804 | l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); |
9fb9cbb1 YK |
805 | |
806 | /* It may be an special packet, error, unclean... | |
807 | * inverse of the return code tells to the netfilter | |
808 | * core what to do with the packet. */ | |
605dcad6 MJ |
809 | if (l4proto->error != NULL && |
810 | (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) { | |
9fb9cbb1 YK |
811 | NF_CT_STAT_INC(error); |
812 | NF_CT_STAT_INC(invalid); | |
813 | return -ret; | |
814 | } | |
815 | ||
605dcad6 | 816 | ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto, |
9fb9cbb1 YK |
817 | &set_reply, &ctinfo); |
818 | if (!ct) { | |
819 | /* Not valid part of a connection */ | |
820 | NF_CT_STAT_INC(invalid); | |
821 | return NF_ACCEPT; | |
822 | } | |
823 | ||
824 | if (IS_ERR(ct)) { | |
825 | /* Too stressed to deal. */ | |
826 | NF_CT_STAT_INC(drop); | |
827 | return NF_DROP; | |
828 | } | |
829 | ||
830 | NF_CT_ASSERT((*pskb)->nfct); | |
831 | ||
605dcad6 | 832 | ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum); |
9fb9cbb1 YK |
833 | if (ret < 0) { |
834 | /* Invalid: inverse of the return code tells | |
835 | * the netfilter core what to do */ | |
836 | DEBUGP("nf_conntrack_in: Can't track with proto module\n"); | |
837 | nf_conntrack_put((*pskb)->nfct); | |
838 | (*pskb)->nfct = NULL; | |
839 | NF_CT_STAT_INC(invalid); | |
840 | return -ret; | |
841 | } | |
842 | ||
843 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
844 | nf_conntrack_event_cache(IPCT_STATUS, *pskb); | |
845 | ||
846 | return ret; | |
847 | } | |
848 | ||
849 | int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, | |
850 | const struct nf_conntrack_tuple *orig) | |
851 | { | |
852 | return nf_ct_invert_tuple(inverse, orig, | |
c1d10adb | 853 | __nf_ct_l3proto_find(orig->src.l3num), |
605dcad6 | 854 | __nf_ct_l4proto_find(orig->src.l3num, |
c1d10adb | 855 | orig->dst.protonum)); |
9fb9cbb1 YK |
856 | } |
857 | ||
5b1158e9 JK |
858 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
859 | implicitly racy: see __nf_conntrack_confirm */ | |
860 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
861 | const struct nf_conntrack_tuple *newreply) | |
862 | { | |
863 | struct nf_conn_help *help = nfct_help(ct); | |
864 | ||
865 | write_lock_bh(&nf_conntrack_lock); | |
866 | /* Should be unconfirmed, so not in hash table yet */ | |
867 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
868 | ||
869 | DEBUGP("Altering reply tuple of %p to ", ct); | |
870 | NF_CT_DUMP_TUPLE(newreply); | |
871 | ||
872 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
873 | if (!ct->master && help && help->expecting == 0) | |
874 | help->helper = __nf_ct_helper_find(newreply); | |
875 | write_unlock_bh(&nf_conntrack_lock); | |
876 | } | |
877 | ||
9fb9cbb1 YK |
878 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
879 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
880 | enum ip_conntrack_info ctinfo, | |
881 | const struct sk_buff *skb, | |
882 | unsigned long extra_jiffies, | |
883 | int do_acct) | |
884 | { | |
885 | int event = 0; | |
886 | ||
887 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); | |
888 | NF_CT_ASSERT(skb); | |
889 | ||
890 | write_lock_bh(&nf_conntrack_lock); | |
891 | ||
997ae831 EL |
892 | /* Only update if this is not a fixed timeout */ |
893 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) { | |
894 | write_unlock_bh(&nf_conntrack_lock); | |
895 | return; | |
896 | } | |
897 | ||
9fb9cbb1 YK |
898 | /* If not in hash table, timer will not be active yet */ |
899 | if (!nf_ct_is_confirmed(ct)) { | |
900 | ct->timeout.expires = extra_jiffies; | |
901 | event = IPCT_REFRESH; | |
902 | } else { | |
be00c8e4 MJ |
903 | unsigned long newtime = jiffies + extra_jiffies; |
904 | ||
905 | /* Only update the timeout if the new timeout is at least | |
906 | HZ jiffies from the old timeout. Need del_timer for race | |
907 | avoidance (may already be dying). */ | |
908 | if (newtime - ct->timeout.expires >= HZ | |
909 | && del_timer(&ct->timeout)) { | |
910 | ct->timeout.expires = newtime; | |
9fb9cbb1 YK |
911 | add_timer(&ct->timeout); |
912 | event = IPCT_REFRESH; | |
913 | } | |
914 | } | |
915 | ||
916 | #ifdef CONFIG_NF_CT_ACCT | |
917 | if (do_acct) { | |
918 | ct->counters[CTINFO2DIR(ctinfo)].packets++; | |
919 | ct->counters[CTINFO2DIR(ctinfo)].bytes += | |
920 | skb->len - (unsigned int)(skb->nh.raw - skb->data); | |
3ffd5eeb MJ |
921 | |
922 | if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) | |
923 | || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) | |
924 | event |= IPCT_COUNTER_FILLING; | |
9fb9cbb1 YK |
925 | } |
926 | #endif | |
927 | ||
928 | write_unlock_bh(&nf_conntrack_lock); | |
929 | ||
930 | /* must be unlocked when calling event cache */ | |
931 | if (event) | |
932 | nf_conntrack_event_cache(event, skb); | |
933 | } | |
934 | ||
c1d10adb PNA |
935 | #if defined(CONFIG_NF_CT_NETLINK) || \ |
936 | defined(CONFIG_NF_CT_NETLINK_MODULE) | |
937 | ||
938 | #include <linux/netfilter/nfnetlink.h> | |
939 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
940 | #include <linux/mutex.h> |
941 | ||
c1d10adb PNA |
942 | |
943 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be | |
944 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
945 | * or depend on ctnetlink */ | |
946 | int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb, | |
947 | const struct nf_conntrack_tuple *tuple) | |
948 | { | |
949 | NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t), | |
950 | &tuple->src.u.tcp.port); | |
951 | NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t), | |
952 | &tuple->dst.u.tcp.port); | |
953 | return 0; | |
954 | ||
955 | nfattr_failure: | |
956 | return -1; | |
957 | } | |
958 | ||
959 | static const size_t cta_min_proto[CTA_PROTO_MAX] = { | |
960 | [CTA_PROTO_SRC_PORT-1] = sizeof(u_int16_t), | |
961 | [CTA_PROTO_DST_PORT-1] = sizeof(u_int16_t) | |
962 | }; | |
963 | ||
964 | int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[], | |
965 | struct nf_conntrack_tuple *t) | |
966 | { | |
967 | if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1]) | |
968 | return -EINVAL; | |
969 | ||
970 | if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto)) | |
971 | return -EINVAL; | |
972 | ||
bff9a89b PM |
973 | t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]); |
974 | t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]); | |
c1d10adb PNA |
975 | |
976 | return 0; | |
977 | } | |
978 | #endif | |
979 | ||
9fb9cbb1 YK |
980 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
981 | void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) | |
982 | { | |
983 | struct nf_conn *ct; | |
984 | enum ip_conntrack_info ctinfo; | |
985 | ||
986 | /* This ICMP is in reverse direction to the packet which caused it */ | |
987 | ct = nf_ct_get(skb, &ctinfo); | |
988 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
989 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
990 | else | |
991 | ctinfo = IP_CT_RELATED; | |
992 | ||
993 | /* Attach to new skbuff, and increment count */ | |
994 | nskb->nfct = &ct->ct_general; | |
995 | nskb->nfctinfo = ctinfo; | |
996 | nf_conntrack_get(nskb->nfct); | |
997 | } | |
998 | ||
999 | static inline int | |
1000 | do_iter(const struct nf_conntrack_tuple_hash *i, | |
1001 | int (*iter)(struct nf_conn *i, void *data), | |
1002 | void *data) | |
1003 | { | |
1004 | return iter(nf_ct_tuplehash_to_ctrack(i), data); | |
1005 | } | |
1006 | ||
1007 | /* Bring out ya dead! */ | |
df0933dc | 1008 | static struct nf_conn * |
9fb9cbb1 YK |
1009 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
1010 | void *data, unsigned int *bucket) | |
1011 | { | |
df0933dc PM |
1012 | struct nf_conntrack_tuple_hash *h; |
1013 | struct nf_conn *ct; | |
9fb9cbb1 YK |
1014 | |
1015 | write_lock_bh(&nf_conntrack_lock); | |
1016 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { | |
df0933dc PM |
1017 | list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) { |
1018 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1019 | if (iter(ct, data)) | |
1020 | goto found; | |
1021 | } | |
9fb9cbb1 | 1022 | } |
df0933dc PM |
1023 | list_for_each_entry(h, &unconfirmed, list) { |
1024 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1025 | if (iter(ct, data)) | |
1026 | goto found; | |
1027 | } | |
c073e3fa | 1028 | write_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
1029 | return NULL; |
1030 | found: | |
c073e3fa | 1031 | atomic_inc(&ct->ct_general.use); |
9fb9cbb1 | 1032 | write_unlock_bh(&nf_conntrack_lock); |
df0933dc | 1033 | return ct; |
9fb9cbb1 YK |
1034 | } |
1035 | ||
1036 | void | |
1037 | nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) | |
1038 | { | |
df0933dc | 1039 | struct nf_conn *ct; |
9fb9cbb1 YK |
1040 | unsigned int bucket = 0; |
1041 | ||
df0933dc | 1042 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
1043 | /* Time to push up daises... */ |
1044 | if (del_timer(&ct->timeout)) | |
1045 | death_by_timeout((unsigned long)ct); | |
1046 | /* ... else the timer will get him soon. */ | |
1047 | ||
1048 | nf_ct_put(ct); | |
1049 | } | |
1050 | } | |
1051 | ||
1052 | static int kill_all(struct nf_conn *i, void *data) | |
1053 | { | |
1054 | return 1; | |
1055 | } | |
1056 | ||
1057 | static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size) | |
1058 | { | |
1059 | if (vmalloced) | |
1060 | vfree(hash); | |
1061 | else | |
1062 | free_pages((unsigned long)hash, | |
1063 | get_order(sizeof(struct list_head) * size)); | |
1064 | } | |
1065 | ||
c1d10adb PNA |
1066 | void nf_conntrack_flush() |
1067 | { | |
1068 | nf_ct_iterate_cleanup(kill_all, NULL); | |
1069 | } | |
1070 | ||
9fb9cbb1 YK |
1071 | /* Mishearing the voices in his head, our hero wonders how he's |
1072 | supposed to kill the mall. */ | |
1073 | void nf_conntrack_cleanup(void) | |
1074 | { | |
1075 | int i; | |
1076 | ||
7d3cdc6b YK |
1077 | ip_ct_attach = NULL; |
1078 | ||
9fb9cbb1 YK |
1079 | /* This makes sure all current packets have passed through |
1080 | netfilter framework. Roll on, two-stage module | |
1081 | delete... */ | |
1082 | synchronize_net(); | |
1083 | ||
1084 | nf_ct_event_cache_flush(); | |
1085 | i_see_dead_people: | |
c1d10adb | 1086 | nf_conntrack_flush(); |
9fb9cbb1 YK |
1087 | if (atomic_read(&nf_conntrack_count) != 0) { |
1088 | schedule(); | |
1089 | goto i_see_dead_people; | |
1090 | } | |
6636568c PM |
1091 | /* wait until all references to nf_conntrack_untracked are dropped */ |
1092 | while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) | |
1093 | schedule(); | |
9fb9cbb1 YK |
1094 | |
1095 | for (i = 0; i < NF_CT_F_NUM; i++) { | |
1096 | if (nf_ct_cache[i].use == 0) | |
1097 | continue; | |
1098 | ||
1099 | NF_CT_ASSERT(nf_ct_cache[i].use == 1); | |
1100 | nf_ct_cache[i].use = 1; | |
1101 | nf_conntrack_unregister_cache(i); | |
1102 | } | |
1103 | kmem_cache_destroy(nf_conntrack_expect_cachep); | |
1104 | free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, | |
1105 | nf_conntrack_htable_size); | |
5a6f294e | 1106 | |
933a41e7 PM |
1107 | nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_generic); |
1108 | ||
5a6f294e KK |
1109 | /* free l3proto protocol tables */ |
1110 | for (i = 0; i < PF_MAX; i++) | |
1111 | if (nf_ct_protos[i]) { | |
1112 | kfree(nf_ct_protos[i]); | |
1113 | nf_ct_protos[i] = NULL; | |
1114 | } | |
9fb9cbb1 YK |
1115 | } |
1116 | ||
1117 | static struct list_head *alloc_hashtable(int size, int *vmalloced) | |
1118 | { | |
1119 | struct list_head *hash; | |
1120 | unsigned int i; | |
1121 | ||
1122 | *vmalloced = 0; | |
1123 | hash = (void*)__get_free_pages(GFP_KERNEL, | |
1124 | get_order(sizeof(struct list_head) | |
1125 | * size)); | |
1126 | if (!hash) { | |
1127 | *vmalloced = 1; | |
1128 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); | |
1129 | hash = vmalloc(sizeof(struct list_head) * size); | |
1130 | } | |
1131 | ||
1132 | if (hash) | |
1133 | for (i = 0; i < size; i++) | |
1134 | INIT_LIST_HEAD(&hash[i]); | |
1135 | ||
1136 | return hash; | |
1137 | } | |
1138 | ||
1139 | int set_hashsize(const char *val, struct kernel_param *kp) | |
1140 | { | |
1141 | int i, bucket, hashsize, vmalloced; | |
1142 | int old_vmalloced, old_size; | |
1143 | int rnd; | |
1144 | struct list_head *hash, *old_hash; | |
1145 | struct nf_conntrack_tuple_hash *h; | |
1146 | ||
1147 | /* On boot, we can set this without any fancy locking. */ | |
1148 | if (!nf_conntrack_htable_size) | |
1149 | return param_set_uint(val, kp); | |
1150 | ||
1151 | hashsize = simple_strtol(val, NULL, 0); | |
1152 | if (!hashsize) | |
1153 | return -EINVAL; | |
1154 | ||
1155 | hash = alloc_hashtable(hashsize, &vmalloced); | |
1156 | if (!hash) | |
1157 | return -ENOMEM; | |
1158 | ||
1159 | /* We have to rehahs for the new table anyway, so we also can | |
1160 | * use a newrandom seed */ | |
1161 | get_random_bytes(&rnd, 4); | |
1162 | ||
1163 | write_lock_bh(&nf_conntrack_lock); | |
1164 | for (i = 0; i < nf_conntrack_htable_size; i++) { | |
1165 | while (!list_empty(&nf_conntrack_hash[i])) { | |
1166 | h = list_entry(nf_conntrack_hash[i].next, | |
1167 | struct nf_conntrack_tuple_hash, list); | |
1168 | list_del(&h->list); | |
1169 | bucket = __hash_conntrack(&h->tuple, hashsize, rnd); | |
1170 | list_add_tail(&h->list, &hash[bucket]); | |
1171 | } | |
1172 | } | |
1173 | old_size = nf_conntrack_htable_size; | |
1174 | old_vmalloced = nf_conntrack_vmalloc; | |
1175 | old_hash = nf_conntrack_hash; | |
1176 | ||
1177 | nf_conntrack_htable_size = hashsize; | |
1178 | nf_conntrack_vmalloc = vmalloced; | |
1179 | nf_conntrack_hash = hash; | |
1180 | nf_conntrack_hash_rnd = rnd; | |
1181 | write_unlock_bh(&nf_conntrack_lock); | |
1182 | ||
1183 | free_conntrack_hash(old_hash, old_vmalloced, old_size); | |
1184 | return 0; | |
1185 | } | |
1186 | ||
1187 | module_param_call(hashsize, set_hashsize, param_get_uint, | |
1188 | &nf_conntrack_htable_size, 0600); | |
1189 | ||
1190 | int __init nf_conntrack_init(void) | |
1191 | { | |
1192 | unsigned int i; | |
1193 | int ret; | |
1194 | ||
1195 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
1196 | * machine has 256 buckets. >= 1GB machines have 8192 buckets. */ | |
1197 | if (!nf_conntrack_htable_size) { | |
1198 | nf_conntrack_htable_size | |
1199 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
1200 | / sizeof(struct list_head)); | |
1201 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
1202 | nf_conntrack_htable_size = 8192; | |
1203 | if (nf_conntrack_htable_size < 16) | |
1204 | nf_conntrack_htable_size = 16; | |
1205 | } | |
1206 | nf_conntrack_max = 8 * nf_conntrack_htable_size; | |
1207 | ||
1208 | printk("nf_conntrack version %s (%u buckets, %d max)\n", | |
1209 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, | |
1210 | nf_conntrack_max); | |
1211 | ||
1212 | nf_conntrack_hash = alloc_hashtable(nf_conntrack_htable_size, | |
1213 | &nf_conntrack_vmalloc); | |
1214 | if (!nf_conntrack_hash) { | |
1215 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1216 | goto err_out; | |
1217 | } | |
1218 | ||
1219 | ret = nf_conntrack_register_cache(NF_CT_F_BASIC, "nf_conntrack:basic", | |
dc808fe2 | 1220 | sizeof(struct nf_conn)); |
9fb9cbb1 YK |
1221 | if (ret < 0) { |
1222 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); | |
1223 | goto err_free_hash; | |
1224 | } | |
1225 | ||
1226 | nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect", | |
1227 | sizeof(struct nf_conntrack_expect), | |
1228 | 0, 0, NULL, NULL); | |
1229 | if (!nf_conntrack_expect_cachep) { | |
1230 | printk(KERN_ERR "Unable to create nf_expect slab cache\n"); | |
1231 | goto err_free_conntrack_slab; | |
1232 | } | |
1233 | ||
933a41e7 PM |
1234 | ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_generic); |
1235 | if (ret < 0) | |
1236 | goto out_free_expect_slab; | |
1237 | ||
9fb9cbb1 YK |
1238 | /* Don't NEED lock here, but good form anyway. */ |
1239 | write_lock_bh(&nf_conntrack_lock); | |
ae5718fb | 1240 | for (i = 0; i < AF_MAX; i++) |
605dcad6 | 1241 | nf_ct_l3protos[i] = &nf_conntrack_l3proto_generic; |
9fb9cbb1 YK |
1242 | write_unlock_bh(&nf_conntrack_lock); |
1243 | ||
7d3cdc6b YK |
1244 | /* For use by REJECT target */ |
1245 | ip_ct_attach = __nf_conntrack_attach; | |
1246 | ||
9fb9cbb1 YK |
1247 | /* Set up fake conntrack: |
1248 | - to never be deleted, not in any hashes */ | |
1249 | atomic_set(&nf_conntrack_untracked.ct_general.use, 1); | |
1250 | /* - and look it like as a confirmed connection */ | |
1251 | set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); | |
1252 | ||
1253 | return ret; | |
1254 | ||
933a41e7 PM |
1255 | out_free_expect_slab: |
1256 | kmem_cache_destroy(nf_conntrack_expect_cachep); | |
9fb9cbb1 YK |
1257 | err_free_conntrack_slab: |
1258 | nf_conntrack_unregister_cache(NF_CT_F_BASIC); | |
1259 | err_free_hash: | |
1260 | free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, | |
1261 | nf_conntrack_htable_size); | |
1262 | err_out: | |
1263 | return -ENOMEM; | |
1264 | } |