Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
6 | * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org> | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or modify | |
9 | * it under the terms of the GNU General Public License version 2 as | |
10 | * published by the Free Software Foundation. | |
11 | * | |
12 | * 23 Apr 2001: Harald Welte <laforge@gnumonks.org> | |
13 | * - new API and handling of conntrack/nat helpers | |
14 | * - now capable of multiple expectations for one master | |
15 | * 16 Jul 2002: Harald Welte <laforge@gnumonks.org> | |
16 | * - add usage/reference counts to ip_conntrack_expect | |
17 | * - export ip_conntrack[_expect]_{find_get,put} functions | |
18 | * */ | |
19 | ||
20 | #include <linux/config.h> | |
21 | #include <linux/types.h> | |
22 | #include <linux/icmp.h> | |
23 | #include <linux/ip.h> | |
24 | #include <linux/netfilter.h> | |
25 | #include <linux/netfilter_ipv4.h> | |
26 | #include <linux/module.h> | |
27 | #include <linux/skbuff.h> | |
28 | #include <linux/proc_fs.h> | |
29 | #include <linux/vmalloc.h> | |
30 | #include <net/checksum.h> | |
31 | #include <net/ip.h> | |
32 | #include <linux/stddef.h> | |
33 | #include <linux/sysctl.h> | |
34 | #include <linux/slab.h> | |
35 | #include <linux/random.h> | |
36 | #include <linux/jhash.h> | |
37 | #include <linux/err.h> | |
38 | #include <linux/percpu.h> | |
39 | #include <linux/moduleparam.h> | |
ac3247ba | 40 | #include <linux/notifier.h> |
1da177e4 | 41 | |
e45b1be8 | 42 | /* ip_conntrack_lock protects the main hash table, protocol/helper/expected |
1da177e4 | 43 | registrations, conntrack timers*/ |
e45b1be8 PM |
44 | #define ASSERT_READ_LOCK(x) |
45 | #define ASSERT_WRITE_LOCK(x) | |
1da177e4 LT |
46 | |
47 | #include <linux/netfilter_ipv4/ip_conntrack.h> | |
48 | #include <linux/netfilter_ipv4/ip_conntrack_protocol.h> | |
49 | #include <linux/netfilter_ipv4/ip_conntrack_helper.h> | |
50 | #include <linux/netfilter_ipv4/ip_conntrack_core.h> | |
51 | #include <linux/netfilter_ipv4/listhelp.h> | |
52 | ||
080774a2 | 53 | #define IP_CONNTRACK_VERSION "2.3" |
1da177e4 LT |
54 | |
55 | #if 0 | |
56 | #define DEBUGP printk | |
57 | #else | |
58 | #define DEBUGP(format, args...) | |
59 | #endif | |
60 | ||
e45b1be8 | 61 | DEFINE_RWLOCK(ip_conntrack_lock); |
1da177e4 LT |
62 | |
63 | /* ip_conntrack_standalone needs this */ | |
64 | atomic_t ip_conntrack_count = ATOMIC_INIT(0); | |
65 | ||
66 | void (*ip_conntrack_destroyed)(struct ip_conntrack *conntrack) = NULL; | |
67 | LIST_HEAD(ip_conntrack_expect_list); | |
68 | struct ip_conntrack_protocol *ip_ct_protos[MAX_IP_CT_PROTO]; | |
69 | static LIST_HEAD(helpers); | |
70 | unsigned int ip_conntrack_htable_size = 0; | |
71 | int ip_conntrack_max; | |
72 | struct list_head *ip_conntrack_hash; | |
ba89966c ED |
73 | static kmem_cache_t *ip_conntrack_cachep __read_mostly; |
74 | static kmem_cache_t *ip_conntrack_expect_cachep __read_mostly; | |
1da177e4 LT |
75 | struct ip_conntrack ip_conntrack_untracked; |
76 | unsigned int ip_ct_log_invalid; | |
77 | static LIST_HEAD(unconfirmed); | |
78 | static int ip_conntrack_vmalloc; | |
79 | ||
080774a2 HW |
80 | static unsigned int ip_conntrack_next_id = 1; |
81 | static unsigned int ip_conntrack_expect_next_id = 1; | |
ac3247ba HW |
82 | #ifdef CONFIG_IP_NF_CONNTRACK_EVENTS |
83 | struct notifier_block *ip_conntrack_chain; | |
84 | struct notifier_block *ip_conntrack_expect_chain; | |
85 | ||
86 | DEFINE_PER_CPU(struct ip_conntrack_ecache, ip_conntrack_ecache); | |
87 | ||
a86888b9 PM |
88 | /* deliver cached events and clear cache entry - must be called with locally |
89 | * disabled softirqs */ | |
90 | static inline void | |
91 | __ip_ct_deliver_cached_events(struct ip_conntrack_ecache *ecache) | |
ac3247ba | 92 | { |
a86888b9 | 93 | DEBUGP("ecache: delivering events for %p\n", ecache->ct); |
ac3247ba HW |
94 | if (is_confirmed(ecache->ct) && !is_dying(ecache->ct) && ecache->events) |
95 | notifier_call_chain(&ip_conntrack_chain, ecache->events, | |
96 | ecache->ct); | |
97 | ecache->events = 0; | |
a86888b9 PM |
98 | ip_conntrack_put(ecache->ct); |
99 | ecache->ct = NULL; | |
ac3247ba HW |
100 | } |
101 | ||
102 | /* Deliver all cached events for a particular conntrack. This is called | |
103 | * by code prior to async packet handling or freeing the skb */ | |
a86888b9 | 104 | void ip_ct_deliver_cached_events(const struct ip_conntrack *ct) |
ac3247ba | 105 | { |
a86888b9 PM |
106 | struct ip_conntrack_ecache *ecache; |
107 | ||
108 | local_bh_disable(); | |
109 | ecache = &__get_cpu_var(ip_conntrack_ecache); | |
110 | if (ecache->ct == ct) | |
111 | __ip_ct_deliver_cached_events(ecache); | |
112 | local_bh_enable(); | |
113 | } | |
ac3247ba | 114 | |
a86888b9 PM |
115 | void __ip_ct_event_cache_init(struct ip_conntrack *ct) |
116 | { | |
117 | struct ip_conntrack_ecache *ecache; | |
ac3247ba | 118 | |
a86888b9 PM |
119 | /* take care of delivering potentially old events */ |
120 | ecache = &__get_cpu_var(ip_conntrack_ecache); | |
121 | BUG_ON(ecache->ct == ct); | |
122 | if (ecache->ct) | |
123 | __ip_ct_deliver_cached_events(ecache); | |
124 | /* initialize for this conntrack/packet */ | |
125 | ecache->ct = ct; | |
126 | nf_conntrack_get(&ct->ct_general); | |
ac3247ba HW |
127 | } |
128 | ||
a86888b9 PM |
129 | /* flush the event cache - touches other CPU's data and must not be called while |
130 | * packets are still passing through the code */ | |
131 | static void ip_ct_event_cache_flush(void) | |
ac3247ba | 132 | { |
a86888b9 PM |
133 | struct ip_conntrack_ecache *ecache; |
134 | int cpu; | |
ac3247ba | 135 | |
a86888b9 PM |
136 | for_each_cpu(cpu) { |
137 | ecache = &per_cpu(ip_conntrack_ecache, cpu); | |
138 | if (ecache->ct) | |
ac3247ba | 139 | ip_conntrack_put(ecache->ct); |
ac3247ba HW |
140 | } |
141 | } | |
a86888b9 PM |
142 | #else |
143 | static inline void ip_ct_event_cache_flush(void) {} | |
ac3247ba HW |
144 | #endif /* CONFIG_IP_NF_CONNTRACK_EVENTS */ |
145 | ||
1da177e4 LT |
146 | DEFINE_PER_CPU(struct ip_conntrack_stat, ip_conntrack_stat); |
147 | ||
1da177e4 LT |
148 | static int ip_conntrack_hash_rnd_initted; |
149 | static unsigned int ip_conntrack_hash_rnd; | |
150 | ||
151 | static u_int32_t | |
152 | hash_conntrack(const struct ip_conntrack_tuple *tuple) | |
153 | { | |
154 | #if 0 | |
155 | dump_tuple(tuple); | |
156 | #endif | |
157 | return (jhash_3words(tuple->src.ip, | |
158 | (tuple->dst.ip ^ tuple->dst.protonum), | |
159 | (tuple->src.u.all | (tuple->dst.u.all << 16)), | |
160 | ip_conntrack_hash_rnd) % ip_conntrack_htable_size); | |
161 | } | |
162 | ||
163 | int | |
164 | ip_ct_get_tuple(const struct iphdr *iph, | |
165 | const struct sk_buff *skb, | |
166 | unsigned int dataoff, | |
167 | struct ip_conntrack_tuple *tuple, | |
168 | const struct ip_conntrack_protocol *protocol) | |
169 | { | |
170 | /* Never happen */ | |
171 | if (iph->frag_off & htons(IP_OFFSET)) { | |
172 | printk("ip_conntrack_core: Frag of proto %u.\n", | |
173 | iph->protocol); | |
174 | return 0; | |
175 | } | |
176 | ||
177 | tuple->src.ip = iph->saddr; | |
178 | tuple->dst.ip = iph->daddr; | |
179 | tuple->dst.protonum = iph->protocol; | |
180 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
181 | ||
182 | return protocol->pkt_to_tuple(skb, dataoff, tuple); | |
183 | } | |
184 | ||
185 | int | |
186 | ip_ct_invert_tuple(struct ip_conntrack_tuple *inverse, | |
187 | const struct ip_conntrack_tuple *orig, | |
188 | const struct ip_conntrack_protocol *protocol) | |
189 | { | |
190 | inverse->src.ip = orig->dst.ip; | |
191 | inverse->dst.ip = orig->src.ip; | |
192 | inverse->dst.protonum = orig->dst.protonum; | |
193 | inverse->dst.dir = !orig->dst.dir; | |
194 | ||
195 | return protocol->invert_tuple(inverse, orig); | |
196 | } | |
197 | ||
198 | ||
199 | /* ip_conntrack_expect helper functions */ | |
1da177e4 LT |
200 | static void unlink_expect(struct ip_conntrack_expect *exp) |
201 | { | |
e45b1be8 | 202 | ASSERT_WRITE_LOCK(&ip_conntrack_lock); |
4acdbdbe | 203 | IP_NF_ASSERT(!timer_pending(&exp->timeout)); |
1da177e4 | 204 | list_del(&exp->list); |
4acdbdbe | 205 | CONNTRACK_STAT_INC(expect_delete); |
1da177e4 | 206 | exp->master->expecting--; |
37012f7f | 207 | ip_conntrack_expect_put(exp); |
1da177e4 LT |
208 | } |
209 | ||
080774a2 HW |
210 | void __ip_ct_expect_unlink_destroy(struct ip_conntrack_expect *exp) |
211 | { | |
212 | unlink_expect(exp); | |
213 | ip_conntrack_expect_put(exp); | |
214 | } | |
215 | ||
1da177e4 LT |
216 | static void expectation_timed_out(unsigned long ul_expect) |
217 | { | |
218 | struct ip_conntrack_expect *exp = (void *)ul_expect; | |
219 | ||
e45b1be8 | 220 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 | 221 | unlink_expect(exp); |
e45b1be8 | 222 | write_unlock_bh(&ip_conntrack_lock); |
4acdbdbe | 223 | ip_conntrack_expect_put(exp); |
1da177e4 LT |
224 | } |
225 | ||
080774a2 HW |
226 | struct ip_conntrack_expect * |
227 | __ip_conntrack_expect_find(const struct ip_conntrack_tuple *tuple) | |
228 | { | |
229 | struct ip_conntrack_expect *i; | |
230 | ||
231 | list_for_each_entry(i, &ip_conntrack_expect_list, list) { | |
232 | if (ip_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask)) { | |
233 | atomic_inc(&i->use); | |
234 | return i; | |
235 | } | |
236 | } | |
237 | return NULL; | |
238 | } | |
239 | ||
240 | /* Just find a expectation corresponding to a tuple. */ | |
241 | struct ip_conntrack_expect * | |
242 | ip_conntrack_expect_find_get(const struct ip_conntrack_tuple *tuple) | |
243 | { | |
244 | struct ip_conntrack_expect *i; | |
245 | ||
246 | read_lock_bh(&ip_conntrack_lock); | |
247 | i = __ip_conntrack_expect_find(tuple); | |
248 | read_unlock_bh(&ip_conntrack_lock); | |
249 | ||
250 | return i; | |
251 | } | |
252 | ||
1da177e4 LT |
253 | /* If an expectation for this connection is found, it gets delete from |
254 | * global list then returned. */ | |
255 | static struct ip_conntrack_expect * | |
256 | find_expectation(const struct ip_conntrack_tuple *tuple) | |
257 | { | |
258 | struct ip_conntrack_expect *i; | |
259 | ||
260 | list_for_each_entry(i, &ip_conntrack_expect_list, list) { | |
261 | /* If master is not in hash table yet (ie. packet hasn't left | |
262 | this machine yet), how can other end know about expected? | |
263 | Hence these are not the droids you are looking for (if | |
264 | master ct never got confirmed, we'd hold a reference to it | |
265 | and weird things would happen to future packets). */ | |
266 | if (ip_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) | |
2248bcfc PM |
267 | && is_confirmed(i->master)) { |
268 | if (i->flags & IP_CT_EXPECT_PERMANENT) { | |
269 | atomic_inc(&i->use); | |
270 | return i; | |
271 | } else if (del_timer(&i->timeout)) { | |
272 | unlink_expect(i); | |
273 | return i; | |
274 | } | |
1da177e4 LT |
275 | } |
276 | } | |
277 | return NULL; | |
278 | } | |
279 | ||
280 | /* delete all expectations for this conntrack */ | |
080774a2 | 281 | void ip_ct_remove_expectations(struct ip_conntrack *ct) |
1da177e4 LT |
282 | { |
283 | struct ip_conntrack_expect *i, *tmp; | |
284 | ||
285 | /* Optimization: most connection never expect any others. */ | |
286 | if (ct->expecting == 0) | |
287 | return; | |
288 | ||
289 | list_for_each_entry_safe(i, tmp, &ip_conntrack_expect_list, list) { | |
290 | if (i->master == ct && del_timer(&i->timeout)) { | |
291 | unlink_expect(i); | |
4acdbdbe | 292 | ip_conntrack_expect_put(i); |
1da177e4 LT |
293 | } |
294 | } | |
295 | } | |
296 | ||
297 | static void | |
298 | clean_from_lists(struct ip_conntrack *ct) | |
299 | { | |
300 | unsigned int ho, hr; | |
301 | ||
302 | DEBUGP("clean_from_lists(%p)\n", ct); | |
e45b1be8 | 303 | ASSERT_WRITE_LOCK(&ip_conntrack_lock); |
1da177e4 LT |
304 | |
305 | ho = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
306 | hr = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
307 | LIST_DELETE(&ip_conntrack_hash[ho], &ct->tuplehash[IP_CT_DIR_ORIGINAL]); | |
308 | LIST_DELETE(&ip_conntrack_hash[hr], &ct->tuplehash[IP_CT_DIR_REPLY]); | |
309 | ||
310 | /* Destroy all pending expectations */ | |
080774a2 | 311 | ip_ct_remove_expectations(ct); |
1da177e4 LT |
312 | } |
313 | ||
314 | static void | |
315 | destroy_conntrack(struct nf_conntrack *nfct) | |
316 | { | |
317 | struct ip_conntrack *ct = (struct ip_conntrack *)nfct; | |
318 | struct ip_conntrack_protocol *proto; | |
319 | ||
320 | DEBUGP("destroy_conntrack(%p)\n", ct); | |
321 | IP_NF_ASSERT(atomic_read(&nfct->use) == 0); | |
322 | IP_NF_ASSERT(!timer_pending(&ct->timeout)); | |
323 | ||
14a50bba | 324 | ip_conntrack_event(IPCT_DESTROY, ct); |
ac3247ba HW |
325 | set_bit(IPS_DYING_BIT, &ct->status); |
326 | ||
1da177e4 LT |
327 | /* To make sure we don't get any weird locking issues here: |
328 | * destroy_conntrack() MUST NOT be called with a write lock | |
329 | * to ip_conntrack_lock!!! -HW */ | |
080774a2 | 330 | proto = __ip_conntrack_proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); |
1da177e4 LT |
331 | if (proto && proto->destroy) |
332 | proto->destroy(ct); | |
333 | ||
334 | if (ip_conntrack_destroyed) | |
335 | ip_conntrack_destroyed(ct); | |
336 | ||
e45b1be8 | 337 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
338 | /* Expectations will have been removed in clean_from_lists, |
339 | * except TFTP can create an expectation on the first packet, | |
340 | * before connection is in the list, so we need to clean here, | |
341 | * too. */ | |
080774a2 | 342 | ip_ct_remove_expectations(ct); |
1da177e4 LT |
343 | |
344 | /* We overload first tuple to link into unconfirmed list. */ | |
345 | if (!is_confirmed(ct)) { | |
346 | BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list)); | |
347 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); | |
348 | } | |
349 | ||
350 | CONNTRACK_STAT_INC(delete); | |
e45b1be8 | 351 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
352 | |
353 | if (ct->master) | |
354 | ip_conntrack_put(ct->master); | |
355 | ||
356 | DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct); | |
080774a2 | 357 | ip_conntrack_free(ct); |
1da177e4 LT |
358 | } |
359 | ||
360 | static void death_by_timeout(unsigned long ul_conntrack) | |
361 | { | |
362 | struct ip_conntrack *ct = (void *)ul_conntrack; | |
363 | ||
e45b1be8 | 364 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
365 | /* Inside lock so preempt is disabled on module removal path. |
366 | * Otherwise we can get spurious warnings. */ | |
367 | CONNTRACK_STAT_INC(delete_list); | |
368 | clean_from_lists(ct); | |
e45b1be8 | 369 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
370 | ip_conntrack_put(ct); |
371 | } | |
372 | ||
373 | static inline int | |
374 | conntrack_tuple_cmp(const struct ip_conntrack_tuple_hash *i, | |
375 | const struct ip_conntrack_tuple *tuple, | |
376 | const struct ip_conntrack *ignored_conntrack) | |
377 | { | |
e45b1be8 | 378 | ASSERT_READ_LOCK(&ip_conntrack_lock); |
1da177e4 LT |
379 | return tuplehash_to_ctrack(i) != ignored_conntrack |
380 | && ip_ct_tuple_equal(tuple, &i->tuple); | |
381 | } | |
382 | ||
080774a2 | 383 | struct ip_conntrack_tuple_hash * |
1da177e4 LT |
384 | __ip_conntrack_find(const struct ip_conntrack_tuple *tuple, |
385 | const struct ip_conntrack *ignored_conntrack) | |
386 | { | |
387 | struct ip_conntrack_tuple_hash *h; | |
388 | unsigned int hash = hash_conntrack(tuple); | |
389 | ||
e45b1be8 | 390 | ASSERT_READ_LOCK(&ip_conntrack_lock); |
1da177e4 LT |
391 | list_for_each_entry(h, &ip_conntrack_hash[hash], list) { |
392 | if (conntrack_tuple_cmp(h, tuple, ignored_conntrack)) { | |
393 | CONNTRACK_STAT_INC(found); | |
394 | return h; | |
395 | } | |
396 | CONNTRACK_STAT_INC(searched); | |
397 | } | |
398 | ||
399 | return NULL; | |
400 | } | |
401 | ||
402 | /* Find a connection corresponding to a tuple. */ | |
403 | struct ip_conntrack_tuple_hash * | |
404 | ip_conntrack_find_get(const struct ip_conntrack_tuple *tuple, | |
405 | const struct ip_conntrack *ignored_conntrack) | |
406 | { | |
407 | struct ip_conntrack_tuple_hash *h; | |
408 | ||
e45b1be8 | 409 | read_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
410 | h = __ip_conntrack_find(tuple, ignored_conntrack); |
411 | if (h) | |
412 | atomic_inc(&tuplehash_to_ctrack(h)->ct_general.use); | |
e45b1be8 | 413 | read_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
414 | |
415 | return h; | |
416 | } | |
417 | ||
080774a2 HW |
418 | static void __ip_conntrack_hash_insert(struct ip_conntrack *ct, |
419 | unsigned int hash, | |
420 | unsigned int repl_hash) | |
421 | { | |
422 | ct->id = ++ip_conntrack_next_id; | |
423 | list_prepend(&ip_conntrack_hash[hash], | |
424 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].list); | |
425 | list_prepend(&ip_conntrack_hash[repl_hash], | |
426 | &ct->tuplehash[IP_CT_DIR_REPLY].list); | |
427 | } | |
428 | ||
429 | void ip_conntrack_hash_insert(struct ip_conntrack *ct) | |
430 | { | |
431 | unsigned int hash, repl_hash; | |
432 | ||
433 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
434 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
435 | ||
436 | write_lock_bh(&ip_conntrack_lock); | |
437 | __ip_conntrack_hash_insert(ct, hash, repl_hash); | |
438 | write_unlock_bh(&ip_conntrack_lock); | |
439 | } | |
440 | ||
1da177e4 LT |
441 | /* Confirm a connection given skb; places it in hash table */ |
442 | int | |
443 | __ip_conntrack_confirm(struct sk_buff **pskb) | |
444 | { | |
445 | unsigned int hash, repl_hash; | |
446 | struct ip_conntrack *ct; | |
447 | enum ip_conntrack_info ctinfo; | |
448 | ||
449 | ct = ip_conntrack_get(*pskb, &ctinfo); | |
450 | ||
451 | /* ipt_REJECT uses ip_conntrack_attach to attach related | |
452 | ICMP/TCP RST packets in other direction. Actual packet | |
453 | which created connection will be IP_CT_NEW or for an | |
454 | expected connection, IP_CT_RELATED. */ | |
455 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
456 | return NF_ACCEPT; | |
457 | ||
458 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
459 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
460 | ||
461 | /* We're not in hash table, and we refuse to set up related | |
462 | connections for unconfirmed conns. But packet copies and | |
463 | REJECT will give spurious warnings here. */ | |
464 | /* IP_NF_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
465 | ||
466 | /* No external references means noone else could have | |
467 | confirmed us. */ | |
468 | IP_NF_ASSERT(!is_confirmed(ct)); | |
469 | DEBUGP("Confirming conntrack %p\n", ct); | |
470 | ||
e45b1be8 | 471 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
472 | |
473 | /* See if there's one in the list already, including reverse: | |
474 | NAT could have grabbed it without realizing, since we're | |
475 | not in the hash. If there is, we lost race. */ | |
476 | if (!LIST_FIND(&ip_conntrack_hash[hash], | |
477 | conntrack_tuple_cmp, | |
478 | struct ip_conntrack_tuple_hash *, | |
479 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, NULL) | |
480 | && !LIST_FIND(&ip_conntrack_hash[repl_hash], | |
481 | conntrack_tuple_cmp, | |
482 | struct ip_conntrack_tuple_hash *, | |
483 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple, NULL)) { | |
484 | /* Remove from unconfirmed list */ | |
485 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); | |
486 | ||
080774a2 | 487 | __ip_conntrack_hash_insert(ct, hash, repl_hash); |
1da177e4 LT |
488 | /* Timer relative to confirmation time, not original |
489 | setting time, otherwise we'd get timer wrap in | |
490 | weird delay cases. */ | |
491 | ct->timeout.expires += jiffies; | |
492 | add_timer(&ct->timeout); | |
493 | atomic_inc(&ct->ct_general.use); | |
494 | set_bit(IPS_CONFIRMED_BIT, &ct->status); | |
495 | CONNTRACK_STAT_INC(insert); | |
e45b1be8 | 496 | write_unlock_bh(&ip_conntrack_lock); |
ac3247ba HW |
497 | if (ct->helper) |
498 | ip_conntrack_event_cache(IPCT_HELPER, *pskb); | |
499 | #ifdef CONFIG_IP_NF_NAT_NEEDED | |
500 | if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || | |
501 | test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) | |
502 | ip_conntrack_event_cache(IPCT_NATINFO, *pskb); | |
503 | #endif | |
504 | ip_conntrack_event_cache(master_ct(ct) ? | |
505 | IPCT_RELATED : IPCT_NEW, *pskb); | |
506 | ||
1da177e4 LT |
507 | return NF_ACCEPT; |
508 | } | |
509 | ||
510 | CONNTRACK_STAT_INC(insert_failed); | |
e45b1be8 | 511 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
512 | |
513 | return NF_DROP; | |
514 | } | |
515 | ||
516 | /* Returns true if a connection correspondings to the tuple (required | |
517 | for NAT). */ | |
518 | int | |
519 | ip_conntrack_tuple_taken(const struct ip_conntrack_tuple *tuple, | |
520 | const struct ip_conntrack *ignored_conntrack) | |
521 | { | |
522 | struct ip_conntrack_tuple_hash *h; | |
523 | ||
e45b1be8 | 524 | read_lock_bh(&ip_conntrack_lock); |
1da177e4 | 525 | h = __ip_conntrack_find(tuple, ignored_conntrack); |
e45b1be8 | 526 | read_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
527 | |
528 | return h != NULL; | |
529 | } | |
530 | ||
531 | /* There's a small race here where we may free a just-assured | |
532 | connection. Too bad: we're in trouble anyway. */ | |
533 | static inline int unreplied(const struct ip_conntrack_tuple_hash *i) | |
534 | { | |
535 | return !(test_bit(IPS_ASSURED_BIT, &tuplehash_to_ctrack(i)->status)); | |
536 | } | |
537 | ||
538 | static int early_drop(struct list_head *chain) | |
539 | { | |
540 | /* Traverse backwards: gives us oldest, which is roughly LRU */ | |
541 | struct ip_conntrack_tuple_hash *h; | |
542 | struct ip_conntrack *ct = NULL; | |
543 | int dropped = 0; | |
544 | ||
e45b1be8 | 545 | read_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
546 | h = LIST_FIND_B(chain, unreplied, struct ip_conntrack_tuple_hash *); |
547 | if (h) { | |
548 | ct = tuplehash_to_ctrack(h); | |
549 | atomic_inc(&ct->ct_general.use); | |
550 | } | |
e45b1be8 | 551 | read_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
552 | |
553 | if (!ct) | |
554 | return dropped; | |
555 | ||
556 | if (del_timer(&ct->timeout)) { | |
557 | death_by_timeout((unsigned long)ct); | |
558 | dropped = 1; | |
559 | CONNTRACK_STAT_INC(early_drop); | |
560 | } | |
561 | ip_conntrack_put(ct); | |
562 | return dropped; | |
563 | } | |
564 | ||
565 | static inline int helper_cmp(const struct ip_conntrack_helper *i, | |
566 | const struct ip_conntrack_tuple *rtuple) | |
567 | { | |
568 | return ip_ct_tuple_mask_cmp(rtuple, &i->tuple, &i->mask); | |
569 | } | |
570 | ||
080774a2 HW |
571 | static struct ip_conntrack_helper * |
572 | __ip_conntrack_helper_find( const struct ip_conntrack_tuple *tuple) | |
1da177e4 LT |
573 | { |
574 | return LIST_FIND(&helpers, helper_cmp, | |
575 | struct ip_conntrack_helper *, | |
576 | tuple); | |
577 | } | |
578 | ||
080774a2 HW |
579 | struct ip_conntrack_helper * |
580 | ip_conntrack_helper_find_get( const struct ip_conntrack_tuple *tuple) | |
581 | { | |
582 | struct ip_conntrack_helper *helper; | |
583 | ||
584 | /* need ip_conntrack_lock to assure that helper exists until | |
585 | * try_module_get() is called */ | |
586 | read_lock_bh(&ip_conntrack_lock); | |
587 | ||
588 | helper = __ip_conntrack_helper_find(tuple); | |
589 | if (helper) { | |
590 | /* need to increase module usage count to assure helper will | |
591 | * not go away while the caller is e.g. busy putting a | |
592 | * conntrack in the hash that uses the helper */ | |
593 | if (!try_module_get(helper->me)) | |
594 | helper = NULL; | |
595 | } | |
596 | ||
597 | read_unlock_bh(&ip_conntrack_lock); | |
598 | ||
599 | return helper; | |
600 | } | |
601 | ||
602 | void ip_conntrack_helper_put(struct ip_conntrack_helper *helper) | |
603 | { | |
604 | module_put(helper->me); | |
605 | } | |
606 | ||
607 | struct ip_conntrack_protocol * | |
608 | __ip_conntrack_proto_find(u_int8_t protocol) | |
609 | { | |
610 | return ip_ct_protos[protocol]; | |
611 | } | |
612 | ||
613 | /* this is guaranteed to always return a valid protocol helper, since | |
614 | * it falls back to generic_protocol */ | |
615 | struct ip_conntrack_protocol * | |
616 | ip_conntrack_proto_find_get(u_int8_t protocol) | |
617 | { | |
618 | struct ip_conntrack_protocol *p; | |
619 | ||
620 | preempt_disable(); | |
621 | p = __ip_conntrack_proto_find(protocol); | |
622 | if (p) { | |
623 | if (!try_module_get(p->me)) | |
624 | p = &ip_conntrack_generic_protocol; | |
625 | } | |
626 | preempt_enable(); | |
627 | ||
628 | return p; | |
629 | } | |
630 | ||
631 | void ip_conntrack_proto_put(struct ip_conntrack_protocol *p) | |
632 | { | |
633 | module_put(p->me); | |
634 | } | |
635 | ||
636 | struct ip_conntrack *ip_conntrack_alloc(struct ip_conntrack_tuple *orig, | |
637 | struct ip_conntrack_tuple *repl) | |
1da177e4 LT |
638 | { |
639 | struct ip_conntrack *conntrack; | |
1da177e4 LT |
640 | |
641 | if (!ip_conntrack_hash_rnd_initted) { | |
642 | get_random_bytes(&ip_conntrack_hash_rnd, 4); | |
643 | ip_conntrack_hash_rnd_initted = 1; | |
644 | } | |
645 | ||
1da177e4 LT |
646 | if (ip_conntrack_max |
647 | && atomic_read(&ip_conntrack_count) >= ip_conntrack_max) { | |
080774a2 | 648 | unsigned int hash = hash_conntrack(orig); |
1da177e4 LT |
649 | /* Try dropping from this hash chain. */ |
650 | if (!early_drop(&ip_conntrack_hash[hash])) { | |
651 | if (net_ratelimit()) | |
652 | printk(KERN_WARNING | |
653 | "ip_conntrack: table full, dropping" | |
654 | " packet.\n"); | |
655 | return ERR_PTR(-ENOMEM); | |
656 | } | |
657 | } | |
658 | ||
1da177e4 LT |
659 | conntrack = kmem_cache_alloc(ip_conntrack_cachep, GFP_ATOMIC); |
660 | if (!conntrack) { | |
661 | DEBUGP("Can't allocate conntrack.\n"); | |
7663f188 | 662 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
663 | } |
664 | ||
665 | memset(conntrack, 0, sizeof(*conntrack)); | |
666 | atomic_set(&conntrack->ct_general.use, 1); | |
667 | conntrack->ct_general.destroy = destroy_conntrack; | |
080774a2 HW |
668 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
669 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; | |
1da177e4 LT |
670 | /* Don't set timer yet: wait for confirmation */ |
671 | init_timer(&conntrack->timeout); | |
672 | conntrack->timeout.data = (unsigned long)conntrack; | |
673 | conntrack->timeout.function = death_by_timeout; | |
674 | ||
080774a2 HW |
675 | atomic_inc(&ip_conntrack_count); |
676 | ||
677 | return conntrack; | |
678 | } | |
679 | ||
680 | void | |
681 | ip_conntrack_free(struct ip_conntrack *conntrack) | |
682 | { | |
683 | atomic_dec(&ip_conntrack_count); | |
684 | kmem_cache_free(ip_conntrack_cachep, conntrack); | |
685 | } | |
686 | ||
687 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
688 | * failed due to stress. Otherwise it really is unclassifiable */ | |
689 | static struct ip_conntrack_tuple_hash * | |
690 | init_conntrack(struct ip_conntrack_tuple *tuple, | |
691 | struct ip_conntrack_protocol *protocol, | |
692 | struct sk_buff *skb) | |
693 | { | |
694 | struct ip_conntrack *conntrack; | |
695 | struct ip_conntrack_tuple repl_tuple; | |
696 | struct ip_conntrack_expect *exp; | |
697 | ||
698 | if (!ip_ct_invert_tuple(&repl_tuple, tuple, protocol)) { | |
699 | DEBUGP("Can't invert tuple.\n"); | |
700 | return NULL; | |
701 | } | |
702 | ||
7663f188 YK |
703 | conntrack = ip_conntrack_alloc(tuple, &repl_tuple); |
704 | if (conntrack == NULL || IS_ERR(conntrack)) | |
705 | return (struct ip_conntrack_tuple_hash *)conntrack; | |
080774a2 HW |
706 | |
707 | if (!protocol->new(conntrack, skb)) { | |
708 | ip_conntrack_free(conntrack); | |
709 | return NULL; | |
710 | } | |
711 | ||
e45b1be8 | 712 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
713 | exp = find_expectation(tuple); |
714 | ||
715 | if (exp) { | |
716 | DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n", | |
717 | conntrack, exp); | |
718 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
719 | __set_bit(IPS_EXPECTED_BIT, &conntrack->status); | |
720 | conntrack->master = exp->master; | |
7cee432a | 721 | #ifdef CONFIG_IP_NF_CONNTRACK_MARK |
1da177e4 | 722 | conntrack->mark = exp->master->mark; |
1f494c0e HW |
723 | #endif |
724 | #if defined(CONFIG_IP_NF_TARGET_MASQUERADE) || \ | |
725 | defined(CONFIG_IP_NF_TARGET_MASQUERADE_MODULE) | |
726 | /* this is ugly, but there is no other place where to put it */ | |
727 | conntrack->nat.masq_index = exp->master->nat.masq_index; | |
1da177e4 LT |
728 | #endif |
729 | nf_conntrack_get(&conntrack->master->ct_general); | |
730 | CONNTRACK_STAT_INC(expect_new); | |
731 | } else { | |
080774a2 | 732 | conntrack->helper = __ip_conntrack_helper_find(&repl_tuple); |
1da177e4 LT |
733 | |
734 | CONNTRACK_STAT_INC(new); | |
735 | } | |
736 | ||
737 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
738 | list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed); | |
739 | ||
e45b1be8 | 740 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
741 | |
742 | if (exp) { | |
743 | if (exp->expectfn) | |
744 | exp->expectfn(conntrack, exp); | |
4acdbdbe | 745 | ip_conntrack_expect_put(exp); |
1da177e4 LT |
746 | } |
747 | ||
748 | return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; | |
749 | } | |
750 | ||
751 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
752 | static inline struct ip_conntrack * | |
753 | resolve_normal_ct(struct sk_buff *skb, | |
754 | struct ip_conntrack_protocol *proto, | |
755 | int *set_reply, | |
756 | unsigned int hooknum, | |
757 | enum ip_conntrack_info *ctinfo) | |
758 | { | |
759 | struct ip_conntrack_tuple tuple; | |
760 | struct ip_conntrack_tuple_hash *h; | |
761 | struct ip_conntrack *ct; | |
762 | ||
763 | IP_NF_ASSERT((skb->nh.iph->frag_off & htons(IP_OFFSET)) == 0); | |
764 | ||
765 | if (!ip_ct_get_tuple(skb->nh.iph, skb, skb->nh.iph->ihl*4, | |
766 | &tuple,proto)) | |
767 | return NULL; | |
768 | ||
769 | /* look for tuple match */ | |
770 | h = ip_conntrack_find_get(&tuple, NULL); | |
771 | if (!h) { | |
772 | h = init_conntrack(&tuple, proto, skb); | |
773 | if (!h) | |
774 | return NULL; | |
775 | if (IS_ERR(h)) | |
776 | return (void *)h; | |
777 | } | |
778 | ct = tuplehash_to_ctrack(h); | |
779 | ||
780 | /* It exists; we have (non-exclusive) reference. */ | |
781 | if (DIRECTION(h) == IP_CT_DIR_REPLY) { | |
782 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; | |
783 | /* Please set reply bit if this packet OK */ | |
784 | *set_reply = 1; | |
785 | } else { | |
786 | /* Once we've had two way comms, always ESTABLISHED. */ | |
787 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
788 | DEBUGP("ip_conntrack_in: normal packet for %p\n", | |
789 | ct); | |
790 | *ctinfo = IP_CT_ESTABLISHED; | |
791 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
792 | DEBUGP("ip_conntrack_in: related packet for %p\n", | |
793 | ct); | |
794 | *ctinfo = IP_CT_RELATED; | |
795 | } else { | |
796 | DEBUGP("ip_conntrack_in: new packet for %p\n", | |
797 | ct); | |
798 | *ctinfo = IP_CT_NEW; | |
799 | } | |
800 | *set_reply = 0; | |
801 | } | |
802 | skb->nfct = &ct->ct_general; | |
803 | skb->nfctinfo = *ctinfo; | |
804 | return ct; | |
805 | } | |
806 | ||
807 | /* Netfilter hook itself. */ | |
808 | unsigned int ip_conntrack_in(unsigned int hooknum, | |
809 | struct sk_buff **pskb, | |
810 | const struct net_device *in, | |
811 | const struct net_device *out, | |
812 | int (*okfn)(struct sk_buff *)) | |
813 | { | |
814 | struct ip_conntrack *ct; | |
815 | enum ip_conntrack_info ctinfo; | |
816 | struct ip_conntrack_protocol *proto; | |
ac3247ba | 817 | int set_reply = 0; |
1da177e4 LT |
818 | int ret; |
819 | ||
820 | /* Previously seen (loopback or untracked)? Ignore. */ | |
821 | if ((*pskb)->nfct) { | |
822 | CONNTRACK_STAT_INC(ignore); | |
823 | return NF_ACCEPT; | |
824 | } | |
825 | ||
826 | /* Never happen */ | |
827 | if ((*pskb)->nh.iph->frag_off & htons(IP_OFFSET)) { | |
828 | if (net_ratelimit()) { | |
829 | printk(KERN_ERR "ip_conntrack_in: Frag of proto %u (hook=%u)\n", | |
830 | (*pskb)->nh.iph->protocol, hooknum); | |
831 | } | |
832 | return NF_DROP; | |
833 | } | |
834 | ||
1da177e4 LT |
835 | /* Doesn't cover locally-generated broadcast, so not worth it. */ |
836 | #if 0 | |
837 | /* Ignore broadcast: no `connection'. */ | |
838 | if ((*pskb)->pkt_type == PACKET_BROADCAST) { | |
839 | printk("Broadcast packet!\n"); | |
840 | return NF_ACCEPT; | |
841 | } else if (((*pskb)->nh.iph->daddr & htonl(0x000000FF)) | |
842 | == htonl(0x000000FF)) { | |
843 | printk("Should bcast: %u.%u.%u.%u->%u.%u.%u.%u (sk=%p, ptype=%u)\n", | |
844 | NIPQUAD((*pskb)->nh.iph->saddr), | |
845 | NIPQUAD((*pskb)->nh.iph->daddr), | |
846 | (*pskb)->sk, (*pskb)->pkt_type); | |
847 | } | |
848 | #endif | |
849 | ||
080774a2 | 850 | proto = __ip_conntrack_proto_find((*pskb)->nh.iph->protocol); |
1da177e4 LT |
851 | |
852 | /* It may be an special packet, error, unclean... | |
853 | * inverse of the return code tells to the netfilter | |
854 | * core what to do with the packet. */ | |
855 | if (proto->error != NULL | |
856 | && (ret = proto->error(*pskb, &ctinfo, hooknum)) <= 0) { | |
857 | CONNTRACK_STAT_INC(error); | |
858 | CONNTRACK_STAT_INC(invalid); | |
859 | return -ret; | |
860 | } | |
861 | ||
862 | if (!(ct = resolve_normal_ct(*pskb, proto,&set_reply,hooknum,&ctinfo))) { | |
863 | /* Not valid part of a connection */ | |
864 | CONNTRACK_STAT_INC(invalid); | |
865 | return NF_ACCEPT; | |
866 | } | |
867 | ||
868 | if (IS_ERR(ct)) { | |
869 | /* Too stressed to deal. */ | |
870 | CONNTRACK_STAT_INC(drop); | |
871 | return NF_DROP; | |
872 | } | |
873 | ||
874 | IP_NF_ASSERT((*pskb)->nfct); | |
875 | ||
876 | ret = proto->packet(ct, *pskb, ctinfo); | |
877 | if (ret < 0) { | |
878 | /* Invalid: inverse of the return code tells | |
879 | * the netfilter core what to do*/ | |
880 | nf_conntrack_put((*pskb)->nfct); | |
881 | (*pskb)->nfct = NULL; | |
882 | CONNTRACK_STAT_INC(invalid); | |
883 | return -ret; | |
884 | } | |
885 | ||
ac3247ba HW |
886 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) |
887 | ip_conntrack_event_cache(IPCT_STATUS, *pskb); | |
1da177e4 LT |
888 | |
889 | return ret; | |
890 | } | |
891 | ||
892 | int invert_tuplepr(struct ip_conntrack_tuple *inverse, | |
893 | const struct ip_conntrack_tuple *orig) | |
894 | { | |
895 | return ip_ct_invert_tuple(inverse, orig, | |
080774a2 | 896 | __ip_conntrack_proto_find(orig->dst.protonum)); |
1da177e4 LT |
897 | } |
898 | ||
899 | /* Would two expected things clash? */ | |
900 | static inline int expect_clash(const struct ip_conntrack_expect *a, | |
901 | const struct ip_conntrack_expect *b) | |
902 | { | |
903 | /* Part covered by intersection of masks must be unequal, | |
904 | otherwise they clash */ | |
905 | struct ip_conntrack_tuple intersect_mask | |
906 | = { { a->mask.src.ip & b->mask.src.ip, | |
907 | { a->mask.src.u.all & b->mask.src.u.all } }, | |
908 | { a->mask.dst.ip & b->mask.dst.ip, | |
909 | { a->mask.dst.u.all & b->mask.dst.u.all }, | |
910 | a->mask.dst.protonum & b->mask.dst.protonum } }; | |
911 | ||
912 | return ip_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask); | |
913 | } | |
914 | ||
915 | static inline int expect_matches(const struct ip_conntrack_expect *a, | |
916 | const struct ip_conntrack_expect *b) | |
917 | { | |
918 | return a->master == b->master | |
919 | && ip_ct_tuple_equal(&a->tuple, &b->tuple) | |
920 | && ip_ct_tuple_equal(&a->mask, &b->mask); | |
921 | } | |
922 | ||
923 | /* Generally a bad idea to call this: could have matched already. */ | |
924 | void ip_conntrack_unexpect_related(struct ip_conntrack_expect *exp) | |
925 | { | |
926 | struct ip_conntrack_expect *i; | |
927 | ||
e45b1be8 | 928 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
929 | /* choose the the oldest expectation to evict */ |
930 | list_for_each_entry_reverse(i, &ip_conntrack_expect_list, list) { | |
931 | if (expect_matches(i, exp) && del_timer(&i->timeout)) { | |
932 | unlink_expect(i); | |
e45b1be8 | 933 | write_unlock_bh(&ip_conntrack_lock); |
4acdbdbe | 934 | ip_conntrack_expect_put(i); |
1da177e4 LT |
935 | return; |
936 | } | |
937 | } | |
e45b1be8 | 938 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
939 | } |
940 | ||
4acdbdbe | 941 | struct ip_conntrack_expect *ip_conntrack_expect_alloc(struct ip_conntrack *me) |
1da177e4 LT |
942 | { |
943 | struct ip_conntrack_expect *new; | |
944 | ||
945 | new = kmem_cache_alloc(ip_conntrack_expect_cachep, GFP_ATOMIC); | |
946 | if (!new) { | |
947 | DEBUGP("expect_related: OOM allocating expect\n"); | |
948 | return NULL; | |
949 | } | |
4acdbdbe RR |
950 | new->master = me; |
951 | atomic_inc(&new->master->ct_general.use); | |
952 | atomic_set(&new->use, 1); | |
1da177e4 LT |
953 | return new; |
954 | } | |
955 | ||
4acdbdbe | 956 | void ip_conntrack_expect_put(struct ip_conntrack_expect *exp) |
1da177e4 | 957 | { |
4acdbdbe RR |
958 | if (atomic_dec_and_test(&exp->use)) { |
959 | ip_conntrack_put(exp->master); | |
960 | kmem_cache_free(ip_conntrack_expect_cachep, exp); | |
961 | } | |
1da177e4 LT |
962 | } |
963 | ||
964 | static void ip_conntrack_expect_insert(struct ip_conntrack_expect *exp) | |
965 | { | |
4acdbdbe | 966 | atomic_inc(&exp->use); |
1da177e4 LT |
967 | exp->master->expecting++; |
968 | list_add(&exp->list, &ip_conntrack_expect_list); | |
969 | ||
1d3cdb41 PO |
970 | init_timer(&exp->timeout); |
971 | exp->timeout.data = (unsigned long)exp; | |
972 | exp->timeout.function = expectation_timed_out; | |
973 | exp->timeout.expires = jiffies + exp->master->helper->timeout * HZ; | |
974 | add_timer(&exp->timeout); | |
1da177e4 | 975 | |
080774a2 HW |
976 | exp->id = ++ip_conntrack_expect_next_id; |
977 | atomic_inc(&exp->use); | |
1da177e4 LT |
978 | CONNTRACK_STAT_INC(expect_create); |
979 | } | |
980 | ||
981 | /* Race with expectations being used means we could have none to find; OK. */ | |
982 | static void evict_oldest_expect(struct ip_conntrack *master) | |
983 | { | |
984 | struct ip_conntrack_expect *i; | |
985 | ||
986 | list_for_each_entry_reverse(i, &ip_conntrack_expect_list, list) { | |
987 | if (i->master == master) { | |
988 | if (del_timer(&i->timeout)) { | |
989 | unlink_expect(i); | |
4acdbdbe | 990 | ip_conntrack_expect_put(i); |
1da177e4 LT |
991 | } |
992 | break; | |
993 | } | |
994 | } | |
995 | } | |
996 | ||
997 | static inline int refresh_timer(struct ip_conntrack_expect *i) | |
998 | { | |
999 | if (!del_timer(&i->timeout)) | |
1000 | return 0; | |
1001 | ||
1002 | i->timeout.expires = jiffies + i->master->helper->timeout*HZ; | |
1003 | add_timer(&i->timeout); | |
1004 | return 1; | |
1005 | } | |
1006 | ||
1007 | int ip_conntrack_expect_related(struct ip_conntrack_expect *expect) | |
1008 | { | |
1009 | struct ip_conntrack_expect *i; | |
1010 | int ret; | |
1011 | ||
1012 | DEBUGP("ip_conntrack_expect_related %p\n", related_to); | |
1013 | DEBUGP("tuple: "); DUMP_TUPLE(&expect->tuple); | |
1014 | DEBUGP("mask: "); DUMP_TUPLE(&expect->mask); | |
1015 | ||
e45b1be8 | 1016 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1017 | list_for_each_entry(i, &ip_conntrack_expect_list, list) { |
1018 | if (expect_matches(i, expect)) { | |
1019 | /* Refresh timer: if it's dying, ignore.. */ | |
1020 | if (refresh_timer(i)) { | |
1021 | ret = 0; | |
1da177e4 LT |
1022 | goto out; |
1023 | } | |
1024 | } else if (expect_clash(i, expect)) { | |
1025 | ret = -EBUSY; | |
1026 | goto out; | |
1027 | } | |
1028 | } | |
1029 | ||
1030 | /* Will be over limit? */ | |
1031 | if (expect->master->helper->max_expected && | |
1032 | expect->master->expecting >= expect->master->helper->max_expected) | |
1033 | evict_oldest_expect(expect->master); | |
1034 | ||
1035 | ip_conntrack_expect_insert(expect); | |
ac3247ba | 1036 | ip_conntrack_expect_event(IPEXP_NEW, expect); |
1da177e4 LT |
1037 | ret = 0; |
1038 | out: | |
e45b1be8 | 1039 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1040 | return ret; |
1041 | } | |
1042 | ||
1043 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is | |
1044 | implicitly racy: see __ip_conntrack_confirm */ | |
1045 | void ip_conntrack_alter_reply(struct ip_conntrack *conntrack, | |
1046 | const struct ip_conntrack_tuple *newreply) | |
1047 | { | |
e45b1be8 | 1048 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1049 | /* Should be unconfirmed, so not in hash table yet */ |
1050 | IP_NF_ASSERT(!is_confirmed(conntrack)); | |
1051 | ||
1052 | DEBUGP("Altering reply tuple of %p to ", conntrack); | |
1053 | DUMP_TUPLE(newreply); | |
1054 | ||
1055 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
1056 | if (!conntrack->master && conntrack->expecting == 0) | |
080774a2 | 1057 | conntrack->helper = __ip_conntrack_helper_find(newreply); |
e45b1be8 | 1058 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1059 | } |
1060 | ||
1061 | int ip_conntrack_helper_register(struct ip_conntrack_helper *me) | |
1062 | { | |
1063 | BUG_ON(me->timeout == 0); | |
e45b1be8 | 1064 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 | 1065 | list_prepend(&helpers, me); |
e45b1be8 | 1066 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1067 | |
1068 | return 0; | |
1069 | } | |
1070 | ||
080774a2 HW |
1071 | struct ip_conntrack_helper * |
1072 | __ip_conntrack_helper_find_byname(const char *name) | |
1073 | { | |
1074 | struct ip_conntrack_helper *h; | |
1075 | ||
1076 | list_for_each_entry(h, &helpers, list) { | |
1077 | if (!strcmp(h->name, name)) | |
1078 | return h; | |
1079 | } | |
1080 | ||
1081 | return NULL; | |
1082 | } | |
1083 | ||
1da177e4 LT |
1084 | static inline int unhelp(struct ip_conntrack_tuple_hash *i, |
1085 | const struct ip_conntrack_helper *me) | |
1086 | { | |
ac3247ba HW |
1087 | if (tuplehash_to_ctrack(i)->helper == me) { |
1088 | ip_conntrack_event(IPCT_HELPER, tuplehash_to_ctrack(i)); | |
1da177e4 | 1089 | tuplehash_to_ctrack(i)->helper = NULL; |
ac3247ba | 1090 | } |
1da177e4 LT |
1091 | return 0; |
1092 | } | |
1093 | ||
1094 | void ip_conntrack_helper_unregister(struct ip_conntrack_helper *me) | |
1095 | { | |
1096 | unsigned int i; | |
1097 | struct ip_conntrack_expect *exp, *tmp; | |
1098 | ||
1099 | /* Need write lock here, to delete helper. */ | |
e45b1be8 | 1100 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1101 | LIST_DELETE(&helpers, me); |
1102 | ||
1103 | /* Get rid of expectations */ | |
1104 | list_for_each_entry_safe(exp, tmp, &ip_conntrack_expect_list, list) { | |
1105 | if (exp->master->helper == me && del_timer(&exp->timeout)) { | |
1106 | unlink_expect(exp); | |
4acdbdbe | 1107 | ip_conntrack_expect_put(exp); |
1da177e4 LT |
1108 | } |
1109 | } | |
1110 | /* Get rid of expecteds, set helpers to NULL. */ | |
1111 | LIST_FIND_W(&unconfirmed, unhelp, struct ip_conntrack_tuple_hash*, me); | |
1112 | for (i = 0; i < ip_conntrack_htable_size; i++) | |
1113 | LIST_FIND_W(&ip_conntrack_hash[i], unhelp, | |
1114 | struct ip_conntrack_tuple_hash *, me); | |
e45b1be8 | 1115 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1116 | |
1117 | /* Someone could be still looking at the helper in a bh. */ | |
1118 | synchronize_net(); | |
1119 | } | |
1120 | ||
1121 | static inline void ct_add_counters(struct ip_conntrack *ct, | |
1122 | enum ip_conntrack_info ctinfo, | |
1123 | const struct sk_buff *skb) | |
1124 | { | |
1125 | #ifdef CONFIG_IP_NF_CT_ACCT | |
1126 | if (skb) { | |
1127 | ct->counters[CTINFO2DIR(ctinfo)].packets++; | |
1128 | ct->counters[CTINFO2DIR(ctinfo)].bytes += | |
1129 | ntohs(skb->nh.iph->tot_len); | |
1130 | } | |
1131 | #endif | |
1132 | } | |
1133 | ||
1134 | /* Refresh conntrack for this many jiffies and do accounting (if skb != NULL) */ | |
1135 | void ip_ct_refresh_acct(struct ip_conntrack *ct, | |
1136 | enum ip_conntrack_info ctinfo, | |
1137 | const struct sk_buff *skb, | |
1138 | unsigned long extra_jiffies) | |
1139 | { | |
1140 | IP_NF_ASSERT(ct->timeout.data == (unsigned long)ct); | |
1141 | ||
1142 | /* If not in hash table, timer will not be active yet */ | |
1143 | if (!is_confirmed(ct)) { | |
1144 | ct->timeout.expires = extra_jiffies; | |
1145 | ct_add_counters(ct, ctinfo, skb); | |
1146 | } else { | |
e45b1be8 | 1147 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1148 | /* Need del_timer for race avoidance (may already be dying). */ |
1149 | if (del_timer(&ct->timeout)) { | |
1150 | ct->timeout.expires = jiffies + extra_jiffies; | |
1151 | add_timer(&ct->timeout); | |
ac3247ba | 1152 | ip_conntrack_event_cache(IPCT_REFRESH, skb); |
1da177e4 LT |
1153 | } |
1154 | ct_add_counters(ct, ctinfo, skb); | |
e45b1be8 | 1155 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1156 | } |
1157 | } | |
1158 | ||
080774a2 HW |
1159 | #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \ |
1160 | defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE) | |
1161 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be | |
1162 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1163 | * or depend on ctnetlink */ | |
1164 | int ip_ct_port_tuple_to_nfattr(struct sk_buff *skb, | |
1165 | const struct ip_conntrack_tuple *tuple) | |
1166 | { | |
1167 | NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t), | |
1168 | &tuple->src.u.tcp.port); | |
1169 | NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t), | |
1170 | &tuple->dst.u.tcp.port); | |
1171 | return 0; | |
1172 | ||
1173 | nfattr_failure: | |
1174 | return -1; | |
1175 | } | |
1176 | ||
1177 | int ip_ct_port_nfattr_to_tuple(struct nfattr *tb[], | |
1178 | struct ip_conntrack_tuple *t) | |
1179 | { | |
1180 | if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1]) | |
1181 | return -EINVAL; | |
1182 | ||
1183 | t->src.u.tcp.port = | |
1184 | *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]); | |
1185 | t->dst.u.tcp.port = | |
1186 | *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]); | |
1187 | ||
1188 | return 0; | |
1189 | } | |
1190 | #endif | |
1191 | ||
1da177e4 LT |
1192 | /* Returns new sk_buff, or NULL */ |
1193 | struct sk_buff * | |
1194 | ip_ct_gather_frags(struct sk_buff *skb, u_int32_t user) | |
1195 | { | |
8be58932 | 1196 | skb_orphan(skb); |
1da177e4 LT |
1197 | |
1198 | local_bh_disable(); | |
1199 | skb = ip_defrag(skb, user); | |
1200 | local_bh_enable(); | |
1201 | ||
6869c4d8 | 1202 | if (skb) |
8be58932 | 1203 | ip_send_check(skb->nh.iph); |
1da177e4 LT |
1204 | return skb; |
1205 | } | |
1206 | ||
1207 | /* Used by ipt_REJECT. */ | |
1208 | static void ip_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) | |
1209 | { | |
1210 | struct ip_conntrack *ct; | |
1211 | enum ip_conntrack_info ctinfo; | |
1212 | ||
1213 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1214 | ct = ip_conntrack_get(skb, &ctinfo); | |
1215 | ||
1216 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
1217 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; | |
1218 | else | |
1219 | ctinfo = IP_CT_RELATED; | |
1220 | ||
1221 | /* Attach to new skbuff, and increment count */ | |
1222 | nskb->nfct = &ct->ct_general; | |
1223 | nskb->nfctinfo = ctinfo; | |
1224 | nf_conntrack_get(nskb->nfct); | |
1225 | } | |
1226 | ||
1227 | static inline int | |
1228 | do_iter(const struct ip_conntrack_tuple_hash *i, | |
1229 | int (*iter)(struct ip_conntrack *i, void *data), | |
1230 | void *data) | |
1231 | { | |
1232 | return iter(tuplehash_to_ctrack(i), data); | |
1233 | } | |
1234 | ||
1235 | /* Bring out ya dead! */ | |
1236 | static struct ip_conntrack_tuple_hash * | |
1237 | get_next_corpse(int (*iter)(struct ip_conntrack *i, void *data), | |
1238 | void *data, unsigned int *bucket) | |
1239 | { | |
1240 | struct ip_conntrack_tuple_hash *h = NULL; | |
1241 | ||
e45b1be8 | 1242 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1243 | for (; *bucket < ip_conntrack_htable_size; (*bucket)++) { |
1244 | h = LIST_FIND_W(&ip_conntrack_hash[*bucket], do_iter, | |
1245 | struct ip_conntrack_tuple_hash *, iter, data); | |
1246 | if (h) | |
1247 | break; | |
1248 | } | |
1249 | if (!h) | |
1250 | h = LIST_FIND_W(&unconfirmed, do_iter, | |
1251 | struct ip_conntrack_tuple_hash *, iter, data); | |
1252 | if (h) | |
1253 | atomic_inc(&tuplehash_to_ctrack(h)->ct_general.use); | |
e45b1be8 | 1254 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1255 | |
1256 | return h; | |
1257 | } | |
1258 | ||
1259 | void | |
1260 | ip_ct_iterate_cleanup(int (*iter)(struct ip_conntrack *i, void *), void *data) | |
1261 | { | |
1262 | struct ip_conntrack_tuple_hash *h; | |
1263 | unsigned int bucket = 0; | |
1264 | ||
1265 | while ((h = get_next_corpse(iter, data, &bucket)) != NULL) { | |
1266 | struct ip_conntrack *ct = tuplehash_to_ctrack(h); | |
1267 | /* Time to push up daises... */ | |
1268 | if (del_timer(&ct->timeout)) | |
1269 | death_by_timeout((unsigned long)ct); | |
1270 | /* ... else the timer will get him soon. */ | |
1271 | ||
1272 | ip_conntrack_put(ct); | |
1273 | } | |
1274 | } | |
1275 | ||
1276 | /* Fast function for those who don't want to parse /proc (and I don't | |
1277 | blame them). */ | |
1278 | /* Reversing the socket's dst/src point of view gives us the reply | |
1279 | mapping. */ | |
1280 | static int | |
1281 | getorigdst(struct sock *sk, int optval, void __user *user, int *len) | |
1282 | { | |
1283 | struct inet_sock *inet = inet_sk(sk); | |
1284 | struct ip_conntrack_tuple_hash *h; | |
1285 | struct ip_conntrack_tuple tuple; | |
1286 | ||
1287 | IP_CT_TUPLE_U_BLANK(&tuple); | |
1288 | tuple.src.ip = inet->rcv_saddr; | |
1289 | tuple.src.u.tcp.port = inet->sport; | |
1290 | tuple.dst.ip = inet->daddr; | |
1291 | tuple.dst.u.tcp.port = inet->dport; | |
1292 | tuple.dst.protonum = IPPROTO_TCP; | |
1293 | ||
1294 | /* We only do TCP at the moment: is there a better way? */ | |
1295 | if (strcmp(sk->sk_prot->name, "TCP")) { | |
1296 | DEBUGP("SO_ORIGINAL_DST: Not a TCP socket\n"); | |
1297 | return -ENOPROTOOPT; | |
1298 | } | |
1299 | ||
1300 | if ((unsigned int) *len < sizeof(struct sockaddr_in)) { | |
1301 | DEBUGP("SO_ORIGINAL_DST: len %u not %u\n", | |
1302 | *len, sizeof(struct sockaddr_in)); | |
1303 | return -EINVAL; | |
1304 | } | |
1305 | ||
1306 | h = ip_conntrack_find_get(&tuple, NULL); | |
1307 | if (h) { | |
1308 | struct sockaddr_in sin; | |
1309 | struct ip_conntrack *ct = tuplehash_to_ctrack(h); | |
1310 | ||
1311 | sin.sin_family = AF_INET; | |
1312 | sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL] | |
1313 | .tuple.dst.u.tcp.port; | |
1314 | sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL] | |
1315 | .tuple.dst.ip; | |
1316 | ||
1317 | DEBUGP("SO_ORIGINAL_DST: %u.%u.%u.%u %u\n", | |
1318 | NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port)); | |
1319 | ip_conntrack_put(ct); | |
1320 | if (copy_to_user(user, &sin, sizeof(sin)) != 0) | |
1321 | return -EFAULT; | |
1322 | else | |
1323 | return 0; | |
1324 | } | |
1325 | DEBUGP("SO_ORIGINAL_DST: Can't find %u.%u.%u.%u/%u-%u.%u.%u.%u/%u.\n", | |
1326 | NIPQUAD(tuple.src.ip), ntohs(tuple.src.u.tcp.port), | |
1327 | NIPQUAD(tuple.dst.ip), ntohs(tuple.dst.u.tcp.port)); | |
1328 | return -ENOENT; | |
1329 | } | |
1330 | ||
1331 | static struct nf_sockopt_ops so_getorigdst = { | |
1332 | .pf = PF_INET, | |
1333 | .get_optmin = SO_ORIGINAL_DST, | |
1334 | .get_optmax = SO_ORIGINAL_DST+1, | |
1335 | .get = &getorigdst, | |
1336 | }; | |
1337 | ||
1338 | static int kill_all(struct ip_conntrack *i, void *data) | |
1339 | { | |
1340 | return 1; | |
1341 | } | |
1342 | ||
1343 | static void free_conntrack_hash(void) | |
1344 | { | |
1345 | if (ip_conntrack_vmalloc) | |
1346 | vfree(ip_conntrack_hash); | |
1347 | else | |
1348 | free_pages((unsigned long)ip_conntrack_hash, | |
1349 | get_order(sizeof(struct list_head) | |
1350 | * ip_conntrack_htable_size)); | |
1351 | } | |
1352 | ||
080774a2 | 1353 | void ip_conntrack_flush() |
1da177e4 | 1354 | { |
1da177e4 LT |
1355 | /* This makes sure all current packets have passed through |
1356 | netfilter framework. Roll on, two-stage module | |
1357 | delete... */ | |
1358 | synchronize_net(); | |
080774a2 | 1359 | |
a86888b9 | 1360 | ip_ct_event_cache_flush(); |
1da177e4 LT |
1361 | i_see_dead_people: |
1362 | ip_ct_iterate_cleanup(kill_all, NULL); | |
1363 | if (atomic_read(&ip_conntrack_count) != 0) { | |
1364 | schedule(); | |
1365 | goto i_see_dead_people; | |
1366 | } | |
21f930e4 PM |
1367 | /* wait until all references to ip_conntrack_untracked are dropped */ |
1368 | while (atomic_read(&ip_conntrack_untracked.ct_general.use) > 1) | |
1369 | schedule(); | |
080774a2 | 1370 | } |
1da177e4 | 1371 | |
080774a2 HW |
1372 | /* Mishearing the voices in his head, our hero wonders how he's |
1373 | supposed to kill the mall. */ | |
1374 | void ip_conntrack_cleanup(void) | |
1375 | { | |
1376 | ip_ct_attach = NULL; | |
1377 | ip_conntrack_flush(); | |
1da177e4 LT |
1378 | kmem_cache_destroy(ip_conntrack_cachep); |
1379 | kmem_cache_destroy(ip_conntrack_expect_cachep); | |
1380 | free_conntrack_hash(); | |
1381 | nf_unregister_sockopt(&so_getorigdst); | |
1382 | } | |
1383 | ||
1384 | static int hashsize; | |
1385 | module_param(hashsize, int, 0400); | |
1386 | ||
1387 | int __init ip_conntrack_init(void) | |
1388 | { | |
1389 | unsigned int i; | |
1390 | int ret; | |
1391 | ||
1392 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
1393 | * machine has 256 buckets. >= 1GB machines have 8192 buckets. */ | |
1394 | if (hashsize) { | |
1395 | ip_conntrack_htable_size = hashsize; | |
1396 | } else { | |
1397 | ip_conntrack_htable_size | |
1398 | = (((num_physpages << PAGE_SHIFT) / 16384) | |
1399 | / sizeof(struct list_head)); | |
1400 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
1401 | ip_conntrack_htable_size = 8192; | |
1402 | if (ip_conntrack_htable_size < 16) | |
1403 | ip_conntrack_htable_size = 16; | |
1404 | } | |
1405 | ip_conntrack_max = 8 * ip_conntrack_htable_size; | |
1406 | ||
1407 | printk("ip_conntrack version %s (%u buckets, %d max)" | |
1408 | " - %Zd bytes per conntrack\n", IP_CONNTRACK_VERSION, | |
1409 | ip_conntrack_htable_size, ip_conntrack_max, | |
1410 | sizeof(struct ip_conntrack)); | |
1411 | ||
1412 | ret = nf_register_sockopt(&so_getorigdst); | |
1413 | if (ret != 0) { | |
1414 | printk(KERN_ERR "Unable to register netfilter socket option\n"); | |
1415 | return ret; | |
1416 | } | |
1417 | ||
1418 | /* AK: the hash table is twice as big than needed because it | |
1419 | uses list_head. it would be much nicer to caches to use a | |
1420 | single pointer list head here. */ | |
1421 | ip_conntrack_vmalloc = 0; | |
1422 | ip_conntrack_hash | |
1423 | =(void*)__get_free_pages(GFP_KERNEL, | |
1424 | get_order(sizeof(struct list_head) | |
1425 | *ip_conntrack_htable_size)); | |
1426 | if (!ip_conntrack_hash) { | |
1427 | ip_conntrack_vmalloc = 1; | |
1428 | printk(KERN_WARNING "ip_conntrack: falling back to vmalloc.\n"); | |
1429 | ip_conntrack_hash = vmalloc(sizeof(struct list_head) | |
1430 | * ip_conntrack_htable_size); | |
1431 | } | |
1432 | if (!ip_conntrack_hash) { | |
1433 | printk(KERN_ERR "Unable to create ip_conntrack_hash\n"); | |
1434 | goto err_unreg_sockopt; | |
1435 | } | |
1436 | ||
1437 | ip_conntrack_cachep = kmem_cache_create("ip_conntrack", | |
1438 | sizeof(struct ip_conntrack), 0, | |
1439 | 0, NULL, NULL); | |
1440 | if (!ip_conntrack_cachep) { | |
1441 | printk(KERN_ERR "Unable to create ip_conntrack slab cache\n"); | |
1442 | goto err_free_hash; | |
1443 | } | |
1444 | ||
1445 | ip_conntrack_expect_cachep = kmem_cache_create("ip_conntrack_expect", | |
1446 | sizeof(struct ip_conntrack_expect), | |
1447 | 0, 0, NULL, NULL); | |
1448 | if (!ip_conntrack_expect_cachep) { | |
1449 | printk(KERN_ERR "Unable to create ip_expect slab cache\n"); | |
1450 | goto err_free_conntrack_slab; | |
1451 | } | |
1452 | ||
1453 | /* Don't NEED lock here, but good form anyway. */ | |
e45b1be8 | 1454 | write_lock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1455 | for (i = 0; i < MAX_IP_CT_PROTO; i++) |
1456 | ip_ct_protos[i] = &ip_conntrack_generic_protocol; | |
1457 | /* Sew in builtin protocols. */ | |
1458 | ip_ct_protos[IPPROTO_TCP] = &ip_conntrack_protocol_tcp; | |
1459 | ip_ct_protos[IPPROTO_UDP] = &ip_conntrack_protocol_udp; | |
1460 | ip_ct_protos[IPPROTO_ICMP] = &ip_conntrack_protocol_icmp; | |
e45b1be8 | 1461 | write_unlock_bh(&ip_conntrack_lock); |
1da177e4 LT |
1462 | |
1463 | for (i = 0; i < ip_conntrack_htable_size; i++) | |
1464 | INIT_LIST_HEAD(&ip_conntrack_hash[i]); | |
1465 | ||
1466 | /* For use by ipt_REJECT */ | |
1467 | ip_ct_attach = ip_conntrack_attach; | |
1468 | ||
1469 | /* Set up fake conntrack: | |
1470 | - to never be deleted, not in any hashes */ | |
1471 | atomic_set(&ip_conntrack_untracked.ct_general.use, 1); | |
1472 | /* - and look it like as a confirmed connection */ | |
1473 | set_bit(IPS_CONFIRMED_BIT, &ip_conntrack_untracked.status); | |
1474 | ||
1475 | return ret; | |
1476 | ||
1477 | err_free_conntrack_slab: | |
1478 | kmem_cache_destroy(ip_conntrack_cachep); | |
1479 | err_free_hash: | |
1480 | free_conntrack_hash(); | |
1481 | err_unreg_sockopt: | |
1482 | nf_unregister_sockopt(&so_getorigdst); | |
1483 | ||
1484 | return -ENOMEM; | |
1485 | } |