Commit | Line | Data |
---|---|---|
9fb9cbb1 YK |
1 | /* Connection state tracking for netfilter. This is separated from, |
2 | but required by, the NAT layer; it can also be used by an iptables | |
3 | extension. */ | |
4 | ||
5 | /* (C) 1999-2001 Paul `Rusty' Russell | |
dc808fe2 | 6 | * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org> |
9fb9cbb1 YK |
7 | * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org> |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
9fb9cbb1 YK |
12 | */ |
13 | ||
9fb9cbb1 YK |
14 | #include <linux/types.h> |
15 | #include <linux/netfilter.h> | |
16 | #include <linux/module.h> | |
d43c36dc | 17 | #include <linux/sched.h> |
9fb9cbb1 YK |
18 | #include <linux/skbuff.h> |
19 | #include <linux/proc_fs.h> | |
20 | #include <linux/vmalloc.h> | |
21 | #include <linux/stddef.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/random.h> | |
24 | #include <linux/jhash.h> | |
25 | #include <linux/err.h> | |
26 | #include <linux/percpu.h> | |
27 | #include <linux/moduleparam.h> | |
28 | #include <linux/notifier.h> | |
29 | #include <linux/kernel.h> | |
30 | #include <linux/netdevice.h> | |
31 | #include <linux/socket.h> | |
d7fe0f24 | 32 | #include <linux/mm.h> |
d696c7bd | 33 | #include <linux/nsproxy.h> |
ea781f19 | 34 | #include <linux/rculist_nulls.h> |
9fb9cbb1 | 35 | |
9fb9cbb1 YK |
36 | #include <net/netfilter/nf_conntrack.h> |
37 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
605dcad6 | 38 | #include <net/netfilter/nf_conntrack_l4proto.h> |
77ab9cff | 39 | #include <net/netfilter/nf_conntrack_expect.h> |
9fb9cbb1 YK |
40 | #include <net/netfilter/nf_conntrack_helper.h> |
41 | #include <net/netfilter/nf_conntrack_core.h> | |
ecfab2c9 | 42 | #include <net/netfilter/nf_conntrack_extend.h> |
58401572 | 43 | #include <net/netfilter/nf_conntrack_acct.h> |
a0891aa6 | 44 | #include <net/netfilter/nf_conntrack_ecache.h> |
5d0aa2cc | 45 | #include <net/netfilter/nf_conntrack_zones.h> |
a992ca2a | 46 | #include <net/netfilter/nf_conntrack_timestamp.h> |
dd705072 | 47 | #include <net/netfilter/nf_conntrack_timeout.h> |
e6a7d3c0 | 48 | #include <net/netfilter/nf_nat.h> |
e17b666a | 49 | #include <net/netfilter/nf_nat_core.h> |
9fb9cbb1 | 50 | |
dc808fe2 | 51 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 52 | |
e17b666a PM |
53 | int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, |
54 | enum nf_nat_manip_type manip, | |
39938324 | 55 | const struct nlattr *attr) __read_mostly; |
e6a7d3c0 PNA |
56 | EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); |
57 | ||
f8ba1aff | 58 | DEFINE_SPINLOCK(nf_conntrack_lock); |
13b18339 | 59 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); |
9fb9cbb1 | 60 | |
e2b7606c | 61 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
62 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
63 | ||
e478075c | 64 | unsigned int nf_conntrack_max __read_mostly; |
a999e683 | 65 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 66 | |
b3c5163f ED |
67 | DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked); |
68 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked); | |
13b18339 | 69 | |
f682cefa | 70 | unsigned int nf_conntrack_hash_rnd __read_mostly; |
4d4e61c6 | 71 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd); |
9fb9cbb1 | 72 | |
99f07e91 | 73 | static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone) |
9fb9cbb1 | 74 | { |
0794935e | 75 | unsigned int n; |
0794935e PM |
76 | |
77 | /* The direction must be ignored, so we hash everything up to the | |
78 | * destination ports (which is a multiple of 4) and treat the last | |
79 | * three bytes manually. | |
80 | */ | |
81 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
99f07e91 CG |
82 | return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^ |
83 | (((__force __u16)tuple->dst.u.all << 16) | | |
84 | tuple->dst.protonum)); | |
85 | } | |
86 | ||
87 | static u32 __hash_bucket(u32 hash, unsigned int size) | |
88 | { | |
89 | return ((u64)hash * size) >> 32; | |
90 | } | |
91 | ||
92 | static u32 hash_bucket(u32 hash, const struct net *net) | |
93 | { | |
94 | return __hash_bucket(hash, net->ct.htable_size); | |
95 | } | |
0794935e | 96 | |
99f07e91 CG |
97 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, |
98 | u16 zone, unsigned int size) | |
99 | { | |
100 | return __hash_bucket(hash_conntrack_raw(tuple, zone), size); | |
9fb9cbb1 YK |
101 | } |
102 | ||
5d0aa2cc | 103 | static inline u_int32_t hash_conntrack(const struct net *net, u16 zone, |
d696c7bd | 104 | const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 | 105 | { |
99f07e91 | 106 | return __hash_conntrack(tuple, zone, net->ct.htable_size); |
9fb9cbb1 YK |
107 | } |
108 | ||
5f2b4c90 | 109 | bool |
9fb9cbb1 YK |
110 | nf_ct_get_tuple(const struct sk_buff *skb, |
111 | unsigned int nhoff, | |
112 | unsigned int dataoff, | |
113 | u_int16_t l3num, | |
114 | u_int8_t protonum, | |
115 | struct nf_conntrack_tuple *tuple, | |
116 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 117 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 118 | { |
443a70d5 | 119 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
120 | |
121 | tuple->src.l3num = l3num; | |
122 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 123 | return false; |
9fb9cbb1 YK |
124 | |
125 | tuple->dst.protonum = protonum; | |
126 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
127 | ||
605dcad6 | 128 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 129 | } |
13b18339 | 130 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 131 | |
5f2b4c90 JE |
132 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
133 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
134 | { |
135 | struct nf_conntrack_l3proto *l3proto; | |
136 | struct nf_conntrack_l4proto *l4proto; | |
137 | unsigned int protoff; | |
138 | u_int8_t protonum; | |
139 | int ret; | |
140 | ||
141 | rcu_read_lock(); | |
142 | ||
143 | l3proto = __nf_ct_l3proto_find(l3num); | |
144 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
145 | if (ret != NF_ACCEPT) { | |
146 | rcu_read_unlock(); | |
5f2b4c90 | 147 | return false; |
e2a3123f YK |
148 | } |
149 | ||
150 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
151 | ||
152 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
153 | l3proto, l4proto); | |
154 | ||
155 | rcu_read_unlock(); | |
156 | return ret; | |
157 | } | |
158 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
159 | ||
5f2b4c90 | 160 | bool |
9fb9cbb1 YK |
161 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
162 | const struct nf_conntrack_tuple *orig, | |
163 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 164 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 165 | { |
443a70d5 | 166 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
167 | |
168 | inverse->src.l3num = orig->src.l3num; | |
169 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 170 | return false; |
9fb9cbb1 YK |
171 | |
172 | inverse->dst.dir = !orig->dst.dir; | |
173 | ||
174 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 175 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 176 | } |
13b18339 | 177 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 178 | |
9fb9cbb1 YK |
179 | static void |
180 | clean_from_lists(struct nf_conn *ct) | |
181 | { | |
0d53778e | 182 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
183 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
184 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
185 | |
186 | /* Destroy all pending expectations */ | |
c1d10adb | 187 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
188 | } |
189 | ||
190 | static void | |
191 | destroy_conntrack(struct nf_conntrack *nfct) | |
192 | { | |
193 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
0d55af87 | 194 | struct net *net = nf_ct_net(ct); |
605dcad6 | 195 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 196 | |
0d53778e | 197 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
198 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
199 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
200 | ||
9fb9cbb1 YK |
201 | /* To make sure we don't get any weird locking issues here: |
202 | * destroy_conntrack() MUST NOT be called with a write lock | |
203 | * to nf_conntrack_lock!!! -HW */ | |
923f4902 | 204 | rcu_read_lock(); |
5e8fbe2a | 205 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
206 | if (l4proto && l4proto->destroy) |
207 | l4proto->destroy(ct); | |
9fb9cbb1 | 208 | |
982d9a9c | 209 | rcu_read_unlock(); |
9fb9cbb1 | 210 | |
f8ba1aff | 211 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
212 | /* Expectations will have been removed in clean_from_lists, |
213 | * except TFTP can create an expectation on the first packet, | |
214 | * before connection is in the list, so we need to clean here, | |
215 | * too. */ | |
c1d10adb | 216 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
217 | |
218 | /* We overload first tuple to link into unconfirmed list. */ | |
219 | if (!nf_ct_is_confirmed(ct)) { | |
ea781f19 ED |
220 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); |
221 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
9fb9cbb1 YK |
222 | } |
223 | ||
0d55af87 | 224 | NF_CT_STAT_INC(net, delete); |
f8ba1aff | 225 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
226 | |
227 | if (ct->master) | |
228 | nf_ct_put(ct->master); | |
229 | ||
0d53778e | 230 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
231 | nf_conntrack_free(ct); |
232 | } | |
233 | ||
dd7669a9 | 234 | void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 235 | { |
0d55af87 | 236 | struct net *net = nf_ct_net(ct); |
9fb9cbb1 | 237 | |
9858a3ae | 238 | nf_ct_helper_destroy(ct); |
f8ba1aff | 239 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
240 | /* Inside lock so preempt is disabled on module removal path. |
241 | * Otherwise we can get spurious warnings. */ | |
0d55af87 | 242 | NF_CT_STAT_INC(net, delete_list); |
9fb9cbb1 | 243 | clean_from_lists(ct); |
f8ba1aff | 244 | spin_unlock_bh(&nf_conntrack_lock); |
dd7669a9 PNA |
245 | } |
246 | EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists); | |
247 | ||
248 | static void death_by_event(unsigned long ul_conntrack) | |
249 | { | |
250 | struct nf_conn *ct = (void *)ul_conntrack; | |
251 | struct net *net = nf_ct_net(ct); | |
252 | ||
253 | if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) { | |
254 | /* bad luck, let's retry again */ | |
255 | ct->timeout.expires = jiffies + | |
256 | (random32() % net->ct.sysctl_events_retry_timeout); | |
257 | add_timer(&ct->timeout); | |
258 | return; | |
259 | } | |
260 | /* we've got the event delivered, now it's dying */ | |
261 | set_bit(IPS_DYING_BIT, &ct->status); | |
262 | spin_lock(&nf_conntrack_lock); | |
263 | hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
264 | spin_unlock(&nf_conntrack_lock); | |
265 | nf_ct_put(ct); | |
266 | } | |
267 | ||
268 | void nf_ct_insert_dying_list(struct nf_conn *ct) | |
269 | { | |
270 | struct net *net = nf_ct_net(ct); | |
271 | ||
272 | /* add this conntrack to the dying list */ | |
273 | spin_lock_bh(&nf_conntrack_lock); | |
274 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
275 | &net->ct.dying); | |
276 | spin_unlock_bh(&nf_conntrack_lock); | |
277 | /* set a new timer to retry event delivery */ | |
278 | setup_timer(&ct->timeout, death_by_event, (unsigned long)ct); | |
279 | ct->timeout.expires = jiffies + | |
280 | (random32() % net->ct.sysctl_events_retry_timeout); | |
281 | add_timer(&ct->timeout); | |
282 | } | |
283 | EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list); | |
284 | ||
285 | static void death_by_timeout(unsigned long ul_conntrack) | |
286 | { | |
287 | struct nf_conn *ct = (void *)ul_conntrack; | |
a992ca2a PNA |
288 | struct nf_conn_tstamp *tstamp; |
289 | ||
290 | tstamp = nf_conn_tstamp_find(ct); | |
291 | if (tstamp && tstamp->stop == 0) | |
292 | tstamp->stop = ktime_to_ns(ktime_get_real()); | |
dd7669a9 PNA |
293 | |
294 | if (!test_bit(IPS_DYING_BIT, &ct->status) && | |
295 | unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) { | |
296 | /* destroy event was not delivered */ | |
297 | nf_ct_delete_from_lists(ct); | |
298 | nf_ct_insert_dying_list(ct); | |
299 | return; | |
300 | } | |
301 | set_bit(IPS_DYING_BIT, &ct->status); | |
302 | nf_ct_delete_from_lists(ct); | |
9fb9cbb1 YK |
303 | nf_ct_put(ct); |
304 | } | |
305 | ||
ea781f19 ED |
306 | /* |
307 | * Warning : | |
308 | * - Caller must take a reference on returned object | |
309 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
310 | * OR | |
311 | * - Caller must lock nf_conntrack_lock before calling this function | |
312 | */ | |
99f07e91 CG |
313 | static struct nf_conntrack_tuple_hash * |
314 | ____nf_conntrack_find(struct net *net, u16 zone, | |
315 | const struct nf_conntrack_tuple *tuple, u32 hash) | |
9fb9cbb1 YK |
316 | { |
317 | struct nf_conntrack_tuple_hash *h; | |
ea781f19 | 318 | struct hlist_nulls_node *n; |
99f07e91 | 319 | unsigned int bucket = hash_bucket(hash, net); |
9fb9cbb1 | 320 | |
4e29e9ec PM |
321 | /* Disable BHs the entire time since we normally need to disable them |
322 | * at least once for the stats anyway. | |
323 | */ | |
324 | local_bh_disable(); | |
ea781f19 | 325 | begin: |
99f07e91 | 326 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) { |
5d0aa2cc PM |
327 | if (nf_ct_tuple_equal(tuple, &h->tuple) && |
328 | nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) { | |
0d55af87 | 329 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 330 | local_bh_enable(); |
9fb9cbb1 YK |
331 | return h; |
332 | } | |
0d55af87 | 333 | NF_CT_STAT_INC(net, searched); |
9fb9cbb1 | 334 | } |
ea781f19 ED |
335 | /* |
336 | * if the nulls value we got at the end of this lookup is | |
337 | * not the expected one, we must restart lookup. | |
338 | * We probably met an item that was moved to another chain. | |
339 | */ | |
99f07e91 | 340 | if (get_nulls_value(n) != bucket) { |
af740b2c | 341 | NF_CT_STAT_INC(net, search_restart); |
ea781f19 | 342 | goto begin; |
af740b2c | 343 | } |
4e29e9ec | 344 | local_bh_enable(); |
9fb9cbb1 YK |
345 | |
346 | return NULL; | |
347 | } | |
99f07e91 CG |
348 | |
349 | struct nf_conntrack_tuple_hash * | |
350 | __nf_conntrack_find(struct net *net, u16 zone, | |
351 | const struct nf_conntrack_tuple *tuple) | |
352 | { | |
353 | return ____nf_conntrack_find(net, zone, tuple, | |
354 | hash_conntrack_raw(tuple, zone)); | |
355 | } | |
13b18339 | 356 | EXPORT_SYMBOL_GPL(__nf_conntrack_find); |
9fb9cbb1 YK |
357 | |
358 | /* Find a connection corresponding to a tuple. */ | |
99f07e91 CG |
359 | static struct nf_conntrack_tuple_hash * |
360 | __nf_conntrack_find_get(struct net *net, u16 zone, | |
361 | const struct nf_conntrack_tuple *tuple, u32 hash) | |
9fb9cbb1 YK |
362 | { |
363 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 364 | struct nf_conn *ct; |
9fb9cbb1 | 365 | |
76507f69 | 366 | rcu_read_lock(); |
ea781f19 | 367 | begin: |
99f07e91 | 368 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
369 | if (h) { |
370 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
371 | if (unlikely(nf_ct_is_dying(ct) || |
372 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 373 | h = NULL; |
ea781f19 | 374 | else { |
5d0aa2cc PM |
375 | if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) || |
376 | nf_ct_zone(ct) != zone)) { | |
ea781f19 ED |
377 | nf_ct_put(ct); |
378 | goto begin; | |
379 | } | |
380 | } | |
76507f69 PM |
381 | } |
382 | rcu_read_unlock(); | |
9fb9cbb1 YK |
383 | |
384 | return h; | |
385 | } | |
99f07e91 CG |
386 | |
387 | struct nf_conntrack_tuple_hash * | |
388 | nf_conntrack_find_get(struct net *net, u16 zone, | |
389 | const struct nf_conntrack_tuple *tuple) | |
390 | { | |
391 | return __nf_conntrack_find_get(net, zone, tuple, | |
392 | hash_conntrack_raw(tuple, zone)); | |
393 | } | |
13b18339 | 394 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 395 | |
c1d10adb PNA |
396 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
397 | unsigned int hash, | |
601e68e1 | 398 | unsigned int repl_hash) |
c1d10adb | 399 | { |
400dad39 AD |
400 | struct net *net = nf_ct_net(ct); |
401 | ||
ea781f19 | 402 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
400dad39 | 403 | &net->ct.hash[hash]); |
ea781f19 | 404 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
400dad39 | 405 | &net->ct.hash[repl_hash]); |
c1d10adb PNA |
406 | } |
407 | ||
7d367e06 JK |
408 | int |
409 | nf_conntrack_hash_check_insert(struct nf_conn *ct) | |
c1d10adb | 410 | { |
d696c7bd | 411 | struct net *net = nf_ct_net(ct); |
c1d10adb | 412 | unsigned int hash, repl_hash; |
7d367e06 JK |
413 | struct nf_conntrack_tuple_hash *h; |
414 | struct hlist_nulls_node *n; | |
5d0aa2cc | 415 | u16 zone; |
c1d10adb | 416 | |
5d0aa2cc | 417 | zone = nf_ct_zone(ct); |
7d367e06 JK |
418 | hash = hash_conntrack(net, zone, |
419 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
420 | repl_hash = hash_conntrack(net, zone, | |
421 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
422 | ||
423 | spin_lock_bh(&nf_conntrack_lock); | |
424 | ||
425 | /* See if there's one in the list already, including reverse */ | |
426 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) | |
427 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, | |
428 | &h->tuple) && | |
429 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
430 | goto out; | |
431 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode) | |
432 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, | |
433 | &h->tuple) && | |
434 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
435 | goto out; | |
c1d10adb | 436 | |
7d367e06 JK |
437 | add_timer(&ct->timeout); |
438 | nf_conntrack_get(&ct->ct_general); | |
c1d10adb | 439 | __nf_conntrack_hash_insert(ct, hash, repl_hash); |
7d367e06 JK |
440 | NF_CT_STAT_INC(net, insert); |
441 | spin_unlock_bh(&nf_conntrack_lock); | |
442 | ||
443 | return 0; | |
444 | ||
445 | out: | |
446 | NF_CT_STAT_INC(net, insert_failed); | |
447 | spin_unlock_bh(&nf_conntrack_lock); | |
448 | return -EEXIST; | |
c1d10adb | 449 | } |
7d367e06 | 450 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); |
c1d10adb | 451 | |
9fb9cbb1 YK |
452 | /* Confirm a connection given skb; places it in hash table */ |
453 | int | |
3db05fea | 454 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 YK |
455 | { |
456 | unsigned int hash, repl_hash; | |
df0933dc | 457 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 458 | struct nf_conn *ct; |
df0933dc | 459 | struct nf_conn_help *help; |
a992ca2a | 460 | struct nf_conn_tstamp *tstamp; |
ea781f19 | 461 | struct hlist_nulls_node *n; |
9fb9cbb1 | 462 | enum ip_conntrack_info ctinfo; |
400dad39 | 463 | struct net *net; |
5d0aa2cc | 464 | u16 zone; |
9fb9cbb1 | 465 | |
3db05fea | 466 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 467 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
468 | |
469 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
470 | ICMP/TCP RST packets in other direction. Actual packet | |
471 | which created connection will be IP_CT_NEW or for an | |
472 | expected connection, IP_CT_RELATED. */ | |
473 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
474 | return NF_ACCEPT; | |
475 | ||
5d0aa2cc | 476 | zone = nf_ct_zone(ct); |
99f07e91 CG |
477 | /* reuse the hash saved before */ |
478 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
479 | hash = hash_bucket(hash, net); | |
480 | repl_hash = hash_conntrack(net, zone, | |
481 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
9fb9cbb1 YK |
482 | |
483 | /* We're not in hash table, and we refuse to set up related | |
484 | connections for unconfirmed conns. But packet copies and | |
485 | REJECT will give spurious warnings here. */ | |
486 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ | |
487 | ||
25985edc | 488 | /* No external references means no one else could have |
9fb9cbb1 YK |
489 | confirmed us. */ |
490 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
0d53778e | 491 | pr_debug("Confirming conntrack %p\n", ct); |
9fb9cbb1 | 492 | |
f8ba1aff | 493 | spin_lock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 494 | |
fc350777 JM |
495 | /* We have to check the DYING flag inside the lock to prevent |
496 | a race against nf_ct_get_next_corpse() possibly called from | |
497 | user context, else we insert an already 'dead' hash, blocking | |
498 | further use of that particular connection -JM */ | |
499 | ||
500 | if (unlikely(nf_ct_is_dying(ct))) { | |
501 | spin_unlock_bh(&nf_conntrack_lock); | |
502 | return NF_ACCEPT; | |
503 | } | |
504 | ||
9fb9cbb1 YK |
505 | /* See if there's one in the list already, including reverse: |
506 | NAT could have grabbed it without realizing, since we're | |
507 | not in the hash. If there is, we lost race. */ | |
ea781f19 | 508 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) |
df0933dc | 509 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
5d0aa2cc PM |
510 | &h->tuple) && |
511 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
df0933dc | 512 | goto out; |
ea781f19 | 513 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode) |
df0933dc | 514 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
5d0aa2cc PM |
515 | &h->tuple) && |
516 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
df0933dc | 517 | goto out; |
9fb9cbb1 | 518 | |
df0933dc | 519 | /* Remove from unconfirmed list */ |
ea781f19 | 520 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
df0933dc | 521 | |
df0933dc PM |
522 | /* Timer relative to confirmation time, not original |
523 | setting time, otherwise we'd get timer wrap in | |
524 | weird delay cases. */ | |
525 | ct->timeout.expires += jiffies; | |
526 | add_timer(&ct->timeout); | |
527 | atomic_inc(&ct->ct_general.use); | |
45eec341 | 528 | ct->status |= IPS_CONFIRMED; |
5c8ec910 | 529 | |
a992ca2a PNA |
530 | /* set conntrack timestamp, if enabled. */ |
531 | tstamp = nf_conn_tstamp_find(ct); | |
532 | if (tstamp) { | |
533 | if (skb->tstamp.tv64 == 0) | |
534 | __net_timestamp((struct sk_buff *)skb); | |
535 | ||
536 | tstamp->start = ktime_to_ns(skb->tstamp); | |
537 | } | |
5c8ec910 PM |
538 | /* Since the lookup is lockless, hash insertion must be done after |
539 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
540 | * guarantee that no other CPU can find the conntrack before the above | |
541 | * stores are visible. | |
542 | */ | |
543 | __nf_conntrack_hash_insert(ct, hash, repl_hash); | |
0d55af87 | 544 | NF_CT_STAT_INC(net, insert); |
f8ba1aff | 545 | spin_unlock_bh(&nf_conntrack_lock); |
5c8ec910 | 546 | |
df0933dc PM |
547 | help = nfct_help(ct); |
548 | if (help && help->helper) | |
a71996fc | 549 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 550 | |
df0933dc | 551 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 552 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 553 | return NF_ACCEPT; |
9fb9cbb1 | 554 | |
df0933dc | 555 | out: |
0d55af87 | 556 | NF_CT_STAT_INC(net, insert_failed); |
f8ba1aff | 557 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
558 | return NF_DROP; |
559 | } | |
13b18339 | 560 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
561 | |
562 | /* Returns true if a connection correspondings to the tuple (required | |
563 | for NAT). */ | |
564 | int | |
565 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
566 | const struct nf_conn *ignored_conntrack) | |
567 | { | |
400dad39 | 568 | struct net *net = nf_ct_net(ignored_conntrack); |
9fb9cbb1 | 569 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 570 | struct hlist_nulls_node *n; |
5d0aa2cc PM |
571 | struct nf_conn *ct; |
572 | u16 zone = nf_ct_zone(ignored_conntrack); | |
573 | unsigned int hash = hash_conntrack(net, zone, tuple); | |
9fb9cbb1 | 574 | |
4e29e9ec PM |
575 | /* Disable BHs the entire time since we need to disable them at |
576 | * least once for the stats anyway. | |
577 | */ | |
578 | rcu_read_lock_bh(); | |
ea781f19 | 579 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { |
5d0aa2cc PM |
580 | ct = nf_ct_tuplehash_to_ctrack(h); |
581 | if (ct != ignored_conntrack && | |
582 | nf_ct_tuple_equal(tuple, &h->tuple) && | |
583 | nf_ct_zone(ct) == zone) { | |
0d55af87 | 584 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 585 | rcu_read_unlock_bh(); |
ba419aff PM |
586 | return 1; |
587 | } | |
0d55af87 | 588 | NF_CT_STAT_INC(net, searched); |
ba419aff | 589 | } |
4e29e9ec | 590 | rcu_read_unlock_bh(); |
9fb9cbb1 | 591 | |
ba419aff | 592 | return 0; |
9fb9cbb1 | 593 | } |
13b18339 | 594 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 595 | |
7ae7730f PM |
596 | #define NF_CT_EVICTION_RANGE 8 |
597 | ||
9fb9cbb1 YK |
598 | /* There's a small race here where we may free a just-assured |
599 | connection. Too bad: we're in trouble anyway. */ | |
400dad39 | 600 | static noinline int early_drop(struct net *net, unsigned int hash) |
9fb9cbb1 | 601 | { |
f205c5e0 | 602 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 603 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 604 | struct nf_conn *ct = NULL, *tmp; |
ea781f19 | 605 | struct hlist_nulls_node *n; |
7ae7730f | 606 | unsigned int i, cnt = 0; |
9fb9cbb1 YK |
607 | int dropped = 0; |
608 | ||
76507f69 | 609 | rcu_read_lock(); |
d696c7bd | 610 | for (i = 0; i < net->ct.htable_size; i++) { |
ea781f19 ED |
611 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], |
612 | hnnode) { | |
7ae7730f PM |
613 | tmp = nf_ct_tuplehash_to_ctrack(h); |
614 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) | |
615 | ct = tmp; | |
616 | cnt++; | |
617 | } | |
76507f69 | 618 | |
5ae27aa2 CG |
619 | if (ct != NULL) { |
620 | if (likely(!nf_ct_is_dying(ct) && | |
621 | atomic_inc_not_zero(&ct->ct_general.use))) | |
622 | break; | |
623 | else | |
624 | ct = NULL; | |
625 | } | |
626 | ||
627 | if (cnt >= NF_CT_EVICTION_RANGE) | |
7ae7730f | 628 | break; |
5ae27aa2 | 629 | |
d696c7bd | 630 | hash = (hash + 1) % net->ct.htable_size; |
9fb9cbb1 | 631 | } |
76507f69 | 632 | rcu_read_unlock(); |
9fb9cbb1 YK |
633 | |
634 | if (!ct) | |
635 | return dropped; | |
636 | ||
637 | if (del_timer(&ct->timeout)) { | |
638 | death_by_timeout((unsigned long)ct); | |
74138511 PNA |
639 | /* Check if we indeed killed this entry. Reliable event |
640 | delivery may have inserted it into the dying list. */ | |
641 | if (test_bit(IPS_DYING_BIT, &ct->status)) { | |
642 | dropped = 1; | |
643 | NF_CT_STAT_INC_ATOMIC(net, early_drop); | |
644 | } | |
9fb9cbb1 YK |
645 | } |
646 | nf_ct_put(ct); | |
647 | return dropped; | |
648 | } | |
649 | ||
f682cefa CG |
650 | void init_nf_conntrack_hash_rnd(void) |
651 | { | |
652 | unsigned int rand; | |
653 | ||
654 | /* | |
655 | * Why not initialize nf_conntrack_rnd in a "init()" function ? | |
656 | * Because there isn't enough entropy when system initializing, | |
657 | * and we initialize it as late as possible. | |
658 | */ | |
659 | do { | |
660 | get_random_bytes(&rand, sizeof(rand)); | |
661 | } while (!rand); | |
662 | cmpxchg(&nf_conntrack_hash_rnd, 0, rand); | |
663 | } | |
664 | ||
99f07e91 CG |
665 | static struct nf_conn * |
666 | __nf_conntrack_alloc(struct net *net, u16 zone, | |
667 | const struct nf_conntrack_tuple *orig, | |
668 | const struct nf_conntrack_tuple *repl, | |
669 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 670 | { |
cd7fcbf1 | 671 | struct nf_conn *ct; |
9fb9cbb1 | 672 | |
b2390969 | 673 | if (unlikely(!nf_conntrack_hash_rnd)) { |
f682cefa | 674 | init_nf_conntrack_hash_rnd(); |
99f07e91 CG |
675 | /* recompute the hash as nf_conntrack_hash_rnd is initialized */ |
676 | hash = hash_conntrack_raw(orig, zone); | |
9fb9cbb1 YK |
677 | } |
678 | ||
5251e2d2 | 679 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 680 | atomic_inc(&net->ct.count); |
5251e2d2 | 681 | |
76eb9460 | 682 | if (nf_conntrack_max && |
49ac8713 | 683 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
99f07e91 | 684 | if (!early_drop(net, hash_bucket(hash, net))) { |
49ac8713 | 685 | atomic_dec(&net->ct.count); |
9fb9cbb1 YK |
686 | if (net_ratelimit()) |
687 | printk(KERN_WARNING | |
688 | "nf_conntrack: table full, dropping" | |
689 | " packet.\n"); | |
690 | return ERR_PTR(-ENOMEM); | |
691 | } | |
692 | } | |
693 | ||
941297f4 ED |
694 | /* |
695 | * Do not use kmem_cache_zalloc(), as this cache uses | |
696 | * SLAB_DESTROY_BY_RCU. | |
697 | */ | |
5b3501fa | 698 | ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp); |
c88130bc | 699 | if (ct == NULL) { |
49ac8713 | 700 | atomic_dec(&net->ct.count); |
dacd2a1a | 701 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 702 | } |
941297f4 ED |
703 | /* |
704 | * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next | |
705 | * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged. | |
706 | */ | |
707 | memset(&ct->tuplehash[IP_CT_DIR_MAX], 0, | |
e5fc9e7a CG |
708 | offsetof(struct nf_conn, proto) - |
709 | offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX])); | |
440f0d58 | 710 | spin_lock_init(&ct->lock); |
c88130bc | 711 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 712 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 713 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
714 | /* save hash for reusing when confirming */ |
715 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
9fb9cbb1 | 716 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 717 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
c2d9ba9b | 718 | write_pnet(&ct->ct_net, net); |
5d0aa2cc PM |
719 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
720 | if (zone) { | |
721 | struct nf_conntrack_zone *nf_ct_zone; | |
722 | ||
723 | nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC); | |
724 | if (!nf_ct_zone) | |
725 | goto out_free; | |
726 | nf_ct_zone->id = zone; | |
727 | } | |
728 | #endif | |
941297f4 ED |
729 | /* |
730 | * changes to lookup keys must be done before setting refcnt to 1 | |
731 | */ | |
732 | smp_wmb(); | |
733 | atomic_set(&ct->ct_general.use, 1); | |
c88130bc | 734 | return ct; |
5d0aa2cc PM |
735 | |
736 | #ifdef CONFIG_NF_CONNTRACK_ZONES | |
737 | out_free: | |
738 | kmem_cache_free(net->ct.nf_conntrack_cachep, ct); | |
739 | return ERR_PTR(-ENOMEM); | |
740 | #endif | |
9fb9cbb1 | 741 | } |
99f07e91 CG |
742 | |
743 | struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone, | |
744 | const struct nf_conntrack_tuple *orig, | |
745 | const struct nf_conntrack_tuple *repl, | |
746 | gfp_t gfp) | |
747 | { | |
748 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
749 | } | |
13b18339 | 750 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 751 | |
c88130bc | 752 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 753 | { |
1d45209d ED |
754 | struct net *net = nf_ct_net(ct); |
755 | ||
ceeff754 | 756 | nf_ct_ext_destroy(ct); |
1d45209d | 757 | atomic_dec(&net->ct.count); |
ea781f19 | 758 | nf_ct_ext_free(ct); |
5b3501fa | 759 | kmem_cache_free(net->ct.nf_conntrack_cachep, ct); |
76507f69 | 760 | } |
13b18339 | 761 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 YK |
762 | |
763 | /* Allocate a new conntrack: we return -ENOMEM if classification | |
764 | failed due to stress. Otherwise it really is unclassifiable. */ | |
765 | static struct nf_conntrack_tuple_hash * | |
b2a15a60 | 766 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 767 | const struct nf_conntrack_tuple *tuple, |
9fb9cbb1 | 768 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 769 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 770 | struct sk_buff *skb, |
2c8503f5 PNA |
771 | unsigned int dataoff, u32 hash, |
772 | unsigned int *timeouts) | |
9fb9cbb1 | 773 | { |
c88130bc | 774 | struct nf_conn *ct; |
3c158f7f | 775 | struct nf_conn_help *help; |
9fb9cbb1 | 776 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 777 | struct nf_conntrack_ecache *ecache; |
9fb9cbb1 | 778 | struct nf_conntrack_expect *exp; |
5d0aa2cc | 779 | u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; |
9fb9cbb1 | 780 | |
605dcad6 | 781 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 782 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
783 | return NULL; |
784 | } | |
785 | ||
99f07e91 CG |
786 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
787 | hash); | |
0a9ee813 | 788 | if (IS_ERR(ct)) |
c88130bc | 789 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 | 790 | |
2c8503f5 | 791 | if (!l4proto->new(ct, skb, dataoff, timeouts)) { |
c88130bc | 792 | nf_conntrack_free(ct); |
0d53778e | 793 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
794 | return NULL; |
795 | } | |
796 | ||
58401572 | 797 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a992ca2a | 798 | nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); |
b2a15a60 PM |
799 | |
800 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
801 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
802 | ecache ? ecache->expmask : 0, | |
803 | GFP_ATOMIC); | |
58401572 | 804 | |
f8ba1aff | 805 | spin_lock_bh(&nf_conntrack_lock); |
5d0aa2cc | 806 | exp = nf_ct_find_expectation(net, zone, tuple); |
9fb9cbb1 | 807 | if (exp) { |
0d53778e | 808 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", |
c88130bc | 809 | ct, exp); |
9fb9cbb1 | 810 | /* Welcome, Mr. Bond. We've been expecting you... */ |
c88130bc PM |
811 | __set_bit(IPS_EXPECTED_BIT, &ct->status); |
812 | ct->master = exp->master; | |
ceceae1b | 813 | if (exp->helper) { |
c88130bc | 814 | help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); |
ceceae1b | 815 | if (help) |
cf778b00 | 816 | rcu_assign_pointer(help->helper, exp->helper); |
ceceae1b YK |
817 | } |
818 | ||
9fb9cbb1 | 819 | #ifdef CONFIG_NF_CONNTRACK_MARK |
c88130bc | 820 | ct->mark = exp->master->mark; |
7c9728c3 JM |
821 | #endif |
822 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
c88130bc | 823 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 824 | #endif |
c88130bc | 825 | nf_conntrack_get(&ct->master->ct_general); |
0d55af87 | 826 | NF_CT_STAT_INC(net, expect_new); |
22e7410b | 827 | } else { |
b2a15a60 | 828 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
0d55af87 | 829 | NF_CT_STAT_INC(net, new); |
22e7410b | 830 | } |
9fb9cbb1 YK |
831 | |
832 | /* Overload tuple linked list to put us in unconfirmed list. */ | |
ea781f19 | 833 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
63c9a262 | 834 | &net->ct.unconfirmed); |
9fb9cbb1 | 835 | |
f8ba1aff | 836 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 YK |
837 | |
838 | if (exp) { | |
839 | if (exp->expectfn) | |
c88130bc | 840 | exp->expectfn(ct, exp); |
6823645d | 841 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
842 | } |
843 | ||
c88130bc | 844 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
845 | } |
846 | ||
847 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
848 | static inline struct nf_conn * | |
b2a15a60 | 849 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 850 | struct sk_buff *skb, |
9fb9cbb1 YK |
851 | unsigned int dataoff, |
852 | u_int16_t l3num, | |
853 | u_int8_t protonum, | |
854 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 855 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 856 | int *set_reply, |
2c8503f5 PNA |
857 | enum ip_conntrack_info *ctinfo, |
858 | unsigned int *timeouts) | |
9fb9cbb1 YK |
859 | { |
860 | struct nf_conntrack_tuple tuple; | |
861 | struct nf_conntrack_tuple_hash *h; | |
862 | struct nf_conn *ct; | |
5d0aa2cc | 863 | u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; |
99f07e91 | 864 | u32 hash; |
9fb9cbb1 | 865 | |
bbe735e4 | 866 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 867 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 868 | l4proto)) { |
0d53778e | 869 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
870 | return NULL; |
871 | } | |
872 | ||
873 | /* look for tuple match */ | |
99f07e91 CG |
874 | hash = hash_conntrack_raw(&tuple, zone); |
875 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); | |
9fb9cbb1 | 876 | if (!h) { |
b2a15a60 | 877 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
2c8503f5 | 878 | skb, dataoff, hash, timeouts); |
9fb9cbb1 YK |
879 | if (!h) |
880 | return NULL; | |
881 | if (IS_ERR(h)) | |
882 | return (void *)h; | |
883 | } | |
884 | ct = nf_ct_tuplehash_to_ctrack(h); | |
885 | ||
886 | /* It exists; we have (non-exclusive) reference. */ | |
887 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
fb048833 | 888 | *ctinfo = IP_CT_ESTABLISHED_REPLY; |
9fb9cbb1 YK |
889 | /* Please set reply bit if this packet OK */ |
890 | *set_reply = 1; | |
891 | } else { | |
892 | /* Once we've had two way comms, always ESTABLISHED. */ | |
893 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 894 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
895 | *ctinfo = IP_CT_ESTABLISHED; |
896 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
897 | pr_debug("nf_conntrack_in: related packet for %p\n", |
898 | ct); | |
9fb9cbb1 YK |
899 | *ctinfo = IP_CT_RELATED; |
900 | } else { | |
0d53778e | 901 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
902 | *ctinfo = IP_CT_NEW; |
903 | } | |
904 | *set_reply = 0; | |
905 | } | |
906 | skb->nfct = &ct->ct_general; | |
907 | skb->nfctinfo = *ctinfo; | |
908 | return ct; | |
909 | } | |
910 | ||
911 | unsigned int | |
a702a65f AD |
912 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
913 | struct sk_buff *skb) | |
9fb9cbb1 | 914 | { |
b2a15a60 | 915 | struct nf_conn *ct, *tmpl = NULL; |
9fb9cbb1 YK |
916 | enum ip_conntrack_info ctinfo; |
917 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 918 | struct nf_conntrack_l4proto *l4proto; |
24de58f4 | 919 | struct nf_conn_timeout *timeout_ext; |
2c8503f5 | 920 | unsigned int *timeouts; |
9fb9cbb1 YK |
921 | unsigned int dataoff; |
922 | u_int8_t protonum; | |
923 | int set_reply = 0; | |
924 | int ret; | |
925 | ||
3db05fea | 926 | if (skb->nfct) { |
b2a15a60 PM |
927 | /* Previously seen (loopback or untracked)? Ignore. */ |
928 | tmpl = (struct nf_conn *)skb->nfct; | |
929 | if (!nf_ct_is_template(tmpl)) { | |
930 | NF_CT_STAT_INC_ATOMIC(net, ignore); | |
931 | return NF_ACCEPT; | |
932 | } | |
933 | skb->nfct = NULL; | |
9fb9cbb1 YK |
934 | } |
935 | ||
923f4902 | 936 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 937 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 938 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
939 | &dataoff, &protonum); |
940 | if (ret <= 0) { | |
25985edc | 941 | pr_debug("not prepared to track yet or error occurred\n"); |
0d55af87 AD |
942 | NF_CT_STAT_INC_ATOMIC(net, error); |
943 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
944 | ret = -ret; |
945 | goto out; | |
9fb9cbb1 YK |
946 | } |
947 | ||
76108cea | 948 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
949 | |
950 | /* It may be an special packet, error, unclean... | |
951 | * inverse of the return code tells to the netfilter | |
952 | * core what to do with the packet. */ | |
74c51a14 | 953 | if (l4proto->error != NULL) { |
8fea97ec PM |
954 | ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo, |
955 | pf, hooknum); | |
74c51a14 | 956 | if (ret <= 0) { |
0d55af87 AD |
957 | NF_CT_STAT_INC_ATOMIC(net, error); |
958 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
959 | ret = -ret; |
960 | goto out; | |
74c51a14 | 961 | } |
88ed01d1 PNA |
962 | /* ICMP[v6] protocol trackers may assign one conntrack. */ |
963 | if (skb->nfct) | |
964 | goto out; | |
9fb9cbb1 YK |
965 | } |
966 | ||
24de58f4 PNA |
967 | /* Decide what timeout policy we want to apply to this flow. */ |
968 | if (tmpl) { | |
969 | timeout_ext = nf_ct_timeout_find(tmpl); | |
970 | if (timeout_ext) | |
971 | timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext); | |
972 | else | |
973 | timeouts = l4proto->get_timeouts(net); | |
974 | } else | |
975 | timeouts = l4proto->get_timeouts(net); | |
2c8503f5 | 976 | |
b2a15a60 | 977 | ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
2c8503f5 PNA |
978 | l3proto, l4proto, &set_reply, &ctinfo, |
979 | timeouts); | |
9fb9cbb1 YK |
980 | if (!ct) { |
981 | /* Not valid part of a connection */ | |
0d55af87 | 982 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
b2a15a60 PM |
983 | ret = NF_ACCEPT; |
984 | goto out; | |
9fb9cbb1 YK |
985 | } |
986 | ||
987 | if (IS_ERR(ct)) { | |
988 | /* Too stressed to deal. */ | |
0d55af87 | 989 | NF_CT_STAT_INC_ATOMIC(net, drop); |
b2a15a60 PM |
990 | ret = NF_DROP; |
991 | goto out; | |
9fb9cbb1 YK |
992 | } |
993 | ||
3db05fea | 994 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 995 | |
2c8503f5 | 996 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); |
ec8d5409 | 997 | if (ret <= 0) { |
9fb9cbb1 YK |
998 | /* Invalid: inverse of the return code tells |
999 | * the netfilter core what to do */ | |
0d53778e | 1000 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
1001 | nf_conntrack_put(skb->nfct); |
1002 | skb->nfct = NULL; | |
0d55af87 | 1003 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
1004 | if (ret == -NF_DROP) |
1005 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
b2a15a60 PM |
1006 | ret = -ret; |
1007 | goto out; | |
9fb9cbb1 YK |
1008 | } |
1009 | ||
1010 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 1011 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 | 1012 | out: |
c3174286 PNA |
1013 | if (tmpl) { |
1014 | /* Special case: we have to repeat this hook, assign the | |
1015 | * template again to this packet. We assume that this packet | |
1016 | * has no conntrack assigned. This is used by nf_ct_tcp. */ | |
1017 | if (ret == NF_REPEAT) | |
1018 | skb->nfct = (struct nf_conntrack *)tmpl; | |
1019 | else | |
1020 | nf_ct_put(tmpl); | |
1021 | } | |
9fb9cbb1 YK |
1022 | |
1023 | return ret; | |
1024 | } | |
13b18339 | 1025 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 1026 | |
5f2b4c90 JE |
1027 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
1028 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 1029 | { |
5f2b4c90 | 1030 | bool ret; |
923f4902 PM |
1031 | |
1032 | rcu_read_lock(); | |
1033 | ret = nf_ct_invert_tuple(inverse, orig, | |
1034 | __nf_ct_l3proto_find(orig->src.l3num), | |
1035 | __nf_ct_l4proto_find(orig->src.l3num, | |
1036 | orig->dst.protonum)); | |
1037 | rcu_read_unlock(); | |
1038 | return ret; | |
9fb9cbb1 | 1039 | } |
13b18339 | 1040 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 1041 | |
5b1158e9 JK |
1042 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
1043 | implicitly racy: see __nf_conntrack_confirm */ | |
1044 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
1045 | const struct nf_conntrack_tuple *newreply) | |
1046 | { | |
1047 | struct nf_conn_help *help = nfct_help(ct); | |
1048 | ||
5b1158e9 JK |
1049 | /* Should be unconfirmed, so not in hash table yet */ |
1050 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
1051 | ||
0d53778e | 1052 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 1053 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
1054 | |
1055 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 1056 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 1057 | return; |
ceceae1b | 1058 | |
c52fbb41 | 1059 | rcu_read_lock(); |
b2a15a60 | 1060 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 1061 | rcu_read_unlock(); |
5b1158e9 | 1062 | } |
13b18339 | 1063 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 1064 | |
9fb9cbb1 YK |
1065 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
1066 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
1067 | enum ip_conntrack_info ctinfo, | |
1068 | const struct sk_buff *skb, | |
1069 | unsigned long extra_jiffies, | |
1070 | int do_acct) | |
1071 | { | |
9fb9cbb1 YK |
1072 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); |
1073 | NF_CT_ASSERT(skb); | |
1074 | ||
997ae831 | 1075 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1076 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1077 | goto acct; | |
997ae831 | 1078 | |
9fb9cbb1 YK |
1079 | /* If not in hash table, timer will not be active yet */ |
1080 | if (!nf_ct_is_confirmed(ct)) { | |
1081 | ct->timeout.expires = extra_jiffies; | |
9fb9cbb1 | 1082 | } else { |
be00c8e4 MJ |
1083 | unsigned long newtime = jiffies + extra_jiffies; |
1084 | ||
1085 | /* Only update the timeout if the new timeout is at least | |
1086 | HZ jiffies from the old timeout. Need del_timer for race | |
1087 | avoidance (may already be dying). */ | |
65cb9fda PM |
1088 | if (newtime - ct->timeout.expires >= HZ) |
1089 | mod_timer_pending(&ct->timeout, newtime); | |
9fb9cbb1 YK |
1090 | } |
1091 | ||
47d95045 | 1092 | acct: |
9fb9cbb1 | 1093 | if (do_acct) { |
58401572 | 1094 | struct nf_conn_counter *acct; |
3ffd5eeb | 1095 | |
58401572 KPO |
1096 | acct = nf_conn_acct_find(ct); |
1097 | if (acct) { | |
b3e0bfa7 ED |
1098 | atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets); |
1099 | atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes); | |
58401572 | 1100 | } |
9fb9cbb1 | 1101 | } |
9fb9cbb1 | 1102 | } |
13b18339 | 1103 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1104 | |
4c889498 DM |
1105 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
1106 | enum ip_conntrack_info ctinfo, | |
1107 | const struct sk_buff *skb, | |
1108 | int do_acct) | |
51091764 | 1109 | { |
718d4ad9 | 1110 | if (do_acct) { |
58401572 KPO |
1111 | struct nf_conn_counter *acct; |
1112 | ||
58401572 KPO |
1113 | acct = nf_conn_acct_find(ct); |
1114 | if (acct) { | |
b3e0bfa7 ED |
1115 | atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets); |
1116 | atomic64_add(skb->len - skb_network_offset(skb), | |
1117 | &acct[CTINFO2DIR(ctinfo)].bytes); | |
58401572 | 1118 | } |
718d4ad9 | 1119 | } |
58401572 | 1120 | |
4c889498 | 1121 | if (del_timer(&ct->timeout)) { |
51091764 | 1122 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
1123 | return true; |
1124 | } | |
1125 | return false; | |
51091764 | 1126 | } |
718d4ad9 | 1127 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 1128 | |
5d0aa2cc PM |
1129 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1130 | static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = { | |
1131 | .len = sizeof(struct nf_conntrack_zone), | |
1132 | .align = __alignof__(struct nf_conntrack_zone), | |
1133 | .id = NF_CT_EXT_ZONE, | |
1134 | }; | |
1135 | #endif | |
1136 | ||
c0cd1156 | 1137 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
c1d10adb PNA |
1138 | |
1139 | #include <linux/netfilter/nfnetlink.h> | |
1140 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1141 | #include <linux/mutex.h> |
1142 | ||
c1d10adb PNA |
1143 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1144 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1145 | * or depend on ctnetlink */ | |
fdf70832 | 1146 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1147 | const struct nf_conntrack_tuple *tuple) |
1148 | { | |
77236b6e PM |
1149 | NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); |
1150 | NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); | |
c1d10adb PNA |
1151 | return 0; |
1152 | ||
df6fb868 | 1153 | nla_put_failure: |
c1d10adb PNA |
1154 | return -1; |
1155 | } | |
fdf70832 | 1156 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1157 | |
f73e924c PM |
1158 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1159 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1160 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1161 | }; |
f73e924c | 1162 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1163 | |
fdf70832 | 1164 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1165 | struct nf_conntrack_tuple *t) |
1166 | { | |
df6fb868 | 1167 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1168 | return -EINVAL; |
1169 | ||
77236b6e PM |
1170 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1171 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1172 | |
1173 | return 0; | |
1174 | } | |
fdf70832 | 1175 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1176 | |
1177 | int nf_ct_port_nlattr_tuple_size(void) | |
1178 | { | |
1179 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1180 | } | |
1181 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1182 | #endif |
1183 | ||
9fb9cbb1 | 1184 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
b334aadc | 1185 | static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
9fb9cbb1 YK |
1186 | { |
1187 | struct nf_conn *ct; | |
1188 | enum ip_conntrack_info ctinfo; | |
1189 | ||
1190 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1191 | ct = nf_ct_get(skb, &ctinfo); | |
1192 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
fb048833 | 1193 | ctinfo = IP_CT_RELATED_REPLY; |
9fb9cbb1 YK |
1194 | else |
1195 | ctinfo = IP_CT_RELATED; | |
1196 | ||
1197 | /* Attach to new skbuff, and increment count */ | |
1198 | nskb->nfct = &ct->ct_general; | |
1199 | nskb->nfctinfo = ctinfo; | |
1200 | nf_conntrack_get(nskb->nfct); | |
1201 | } | |
1202 | ||
9fb9cbb1 | 1203 | /* Bring out ya dead! */ |
df0933dc | 1204 | static struct nf_conn * |
400dad39 | 1205 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1206 | void *data, unsigned int *bucket) |
1207 | { | |
df0933dc PM |
1208 | struct nf_conntrack_tuple_hash *h; |
1209 | struct nf_conn *ct; | |
ea781f19 | 1210 | struct hlist_nulls_node *n; |
9fb9cbb1 | 1211 | |
f8ba1aff | 1212 | spin_lock_bh(&nf_conntrack_lock); |
d696c7bd | 1213 | for (; *bucket < net->ct.htable_size; (*bucket)++) { |
ea781f19 | 1214 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) { |
df0933dc PM |
1215 | ct = nf_ct_tuplehash_to_ctrack(h); |
1216 | if (iter(ct, data)) | |
1217 | goto found; | |
1218 | } | |
601e68e1 | 1219 | } |
ea781f19 | 1220 | hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) { |
df0933dc PM |
1221 | ct = nf_ct_tuplehash_to_ctrack(h); |
1222 | if (iter(ct, data)) | |
ec68e97d | 1223 | set_bit(IPS_DYING_BIT, &ct->status); |
df0933dc | 1224 | } |
f8ba1aff | 1225 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc PM |
1226 | return NULL; |
1227 | found: | |
c073e3fa | 1228 | atomic_inc(&ct->ct_general.use); |
f8ba1aff | 1229 | spin_unlock_bh(&nf_conntrack_lock); |
df0933dc | 1230 | return ct; |
9fb9cbb1 YK |
1231 | } |
1232 | ||
400dad39 AD |
1233 | void nf_ct_iterate_cleanup(struct net *net, |
1234 | int (*iter)(struct nf_conn *i, void *data), | |
1235 | void *data) | |
9fb9cbb1 | 1236 | { |
df0933dc | 1237 | struct nf_conn *ct; |
9fb9cbb1 YK |
1238 | unsigned int bucket = 0; |
1239 | ||
400dad39 | 1240 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
1241 | /* Time to push up daises... */ |
1242 | if (del_timer(&ct->timeout)) | |
1243 | death_by_timeout((unsigned long)ct); | |
1244 | /* ... else the timer will get him soon. */ | |
1245 | ||
1246 | nf_ct_put(ct); | |
1247 | } | |
1248 | } | |
13b18339 | 1249 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1250 | |
19abb7b0 PNA |
1251 | struct __nf_ct_flush_report { |
1252 | u32 pid; | |
1253 | int report; | |
1254 | }; | |
1255 | ||
274d383b | 1256 | static int kill_report(struct nf_conn *i, void *data) |
9fb9cbb1 | 1257 | { |
19abb7b0 | 1258 | struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data; |
a992ca2a PNA |
1259 | struct nf_conn_tstamp *tstamp; |
1260 | ||
1261 | tstamp = nf_conn_tstamp_find(i); | |
1262 | if (tstamp && tstamp->stop == 0) | |
1263 | tstamp->stop = ktime_to_ns(ktime_get_real()); | |
19abb7b0 | 1264 | |
dd7669a9 PNA |
1265 | /* If we fail to deliver the event, death_by_timeout() will retry */ |
1266 | if (nf_conntrack_event_report(IPCT_DESTROY, i, | |
1267 | fr->pid, fr->report) < 0) | |
1268 | return 1; | |
1269 | ||
1270 | /* Avoid the delivery of the destroy event in death_by_timeout(). */ | |
1271 | set_bit(IPS_DYING_BIT, &i->status); | |
9fb9cbb1 YK |
1272 | return 1; |
1273 | } | |
1274 | ||
274d383b PNA |
1275 | static int kill_all(struct nf_conn *i, void *data) |
1276 | { | |
1277 | return 1; | |
1278 | } | |
1279 | ||
d862a662 | 1280 | void nf_ct_free_hashtable(void *hash, unsigned int size) |
9fb9cbb1 | 1281 | { |
d862a662 | 1282 | if (is_vmalloc_addr(hash)) |
9fb9cbb1 YK |
1283 | vfree(hash); |
1284 | else | |
601e68e1 | 1285 | free_pages((unsigned long)hash, |
f205c5e0 | 1286 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1287 | } |
ac565e5f | 1288 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1289 | |
274d383b | 1290 | void nf_conntrack_flush_report(struct net *net, u32 pid, int report) |
c1d10adb | 1291 | { |
19abb7b0 PNA |
1292 | struct __nf_ct_flush_report fr = { |
1293 | .pid = pid, | |
1294 | .report = report, | |
1295 | }; | |
274d383b | 1296 | nf_ct_iterate_cleanup(net, kill_report, &fr); |
c1d10adb | 1297 | } |
274d383b | 1298 | EXPORT_SYMBOL_GPL(nf_conntrack_flush_report); |
c1d10adb | 1299 | |
ee254fa4 | 1300 | static void nf_ct_release_dying_list(struct net *net) |
dd7669a9 PNA |
1301 | { |
1302 | struct nf_conntrack_tuple_hash *h; | |
1303 | struct nf_conn *ct; | |
1304 | struct hlist_nulls_node *n; | |
1305 | ||
1306 | spin_lock_bh(&nf_conntrack_lock); | |
ee254fa4 | 1307 | hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) { |
dd7669a9 PNA |
1308 | ct = nf_ct_tuplehash_to_ctrack(h); |
1309 | /* never fails to remove them, no listeners at this point */ | |
1310 | nf_ct_kill(ct); | |
1311 | } | |
1312 | spin_unlock_bh(&nf_conntrack_lock); | |
1313 | } | |
1314 | ||
b3c5163f ED |
1315 | static int untrack_refs(void) |
1316 | { | |
1317 | int cnt = 0, cpu; | |
1318 | ||
1319 | for_each_possible_cpu(cpu) { | |
1320 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
1321 | ||
1322 | cnt += atomic_read(&ct->ct_general.use) - 1; | |
1323 | } | |
1324 | return cnt; | |
1325 | } | |
1326 | ||
08f6547d | 1327 | static void nf_conntrack_cleanup_init_net(void) |
9fb9cbb1 | 1328 | { |
b3c5163f | 1329 | while (untrack_refs() > 0) |
9edd7ca0 PM |
1330 | schedule(); |
1331 | ||
08f6547d AD |
1332 | nf_conntrack_helper_fini(); |
1333 | nf_conntrack_proto_fini(); | |
5d0aa2cc PM |
1334 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1335 | nf_ct_extend_unregister(&nf_ct_zone_extend); | |
1336 | #endif | |
08f6547d | 1337 | } |
9fb9cbb1 | 1338 | |
08f6547d AD |
1339 | static void nf_conntrack_cleanup_net(struct net *net) |
1340 | { | |
9fb9cbb1 | 1341 | i_see_dead_people: |
274d383b | 1342 | nf_ct_iterate_cleanup(net, kill_all, NULL); |
ee254fa4 | 1343 | nf_ct_release_dying_list(net); |
49ac8713 | 1344 | if (atomic_read(&net->ct.count) != 0) { |
9fb9cbb1 YK |
1345 | schedule(); |
1346 | goto i_see_dead_people; | |
1347 | } | |
1348 | ||
d862a662 | 1349 | nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size); |
dd705072 | 1350 | nf_conntrack_timeout_fini(net); |
a0891aa6 | 1351 | nf_conntrack_ecache_fini(net); |
fe8f661f | 1352 | nf_conntrack_tstamp_fini(net); |
d716a4df | 1353 | nf_conntrack_acct_fini(net); |
9b03f38d | 1354 | nf_conntrack_expect_fini(net); |
5b3501fa ED |
1355 | kmem_cache_destroy(net->ct.nf_conntrack_cachep); |
1356 | kfree(net->ct.slabname); | |
0d55af87 | 1357 | free_percpu(net->ct.stat); |
08f6547d AD |
1358 | } |
1359 | ||
1360 | /* Mishearing the voices in his head, our hero wonders how he's | |
1361 | supposed to kill the mall. */ | |
1362 | void nf_conntrack_cleanup(struct net *net) | |
1363 | { | |
1364 | if (net_eq(net, &init_net)) | |
a9b3cd7f | 1365 | RCU_INIT_POINTER(ip_ct_attach, NULL); |
08f6547d AD |
1366 | |
1367 | /* This makes sure all current packets have passed through | |
1368 | netfilter framework. Roll on, two-stage module | |
1369 | delete... */ | |
1370 | synchronize_net(); | |
1371 | ||
1372 | nf_conntrack_cleanup_net(net); | |
1373 | ||
1374 | if (net_eq(net, &init_net)) { | |
a9b3cd7f | 1375 | RCU_INIT_POINTER(nf_ct_destroy, NULL); |
08f6547d AD |
1376 | nf_conntrack_cleanup_init_net(); |
1377 | } | |
9fb9cbb1 YK |
1378 | } |
1379 | ||
d862a662 | 1380 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) |
9fb9cbb1 | 1381 | { |
ea781f19 ED |
1382 | struct hlist_nulls_head *hash; |
1383 | unsigned int nr_slots, i; | |
1384 | size_t sz; | |
9fb9cbb1 | 1385 | |
ea781f19 ED |
1386 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1387 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
1388 | sz = nr_slots * sizeof(struct hlist_nulls_head); | |
1389 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1390 | get_order(sz)); | |
601e68e1 | 1391 | if (!hash) { |
9fb9cbb1 | 1392 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); |
966567b7 | 1393 | hash = vzalloc(sz); |
9fb9cbb1 YK |
1394 | } |
1395 | ||
ea781f19 ED |
1396 | if (hash && nulls) |
1397 | for (i = 0; i < nr_slots; i++) | |
1398 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1399 | |
1400 | return hash; | |
1401 | } | |
ac565e5f | 1402 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1403 | |
fae718dd | 1404 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1405 | { |
d862a662 | 1406 | int i, bucket; |
96eb24d7 | 1407 | unsigned int hashsize, old_size; |
ea781f19 | 1408 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1409 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1410 | struct nf_conn *ct; |
9fb9cbb1 | 1411 | |
d696c7bd PM |
1412 | if (current->nsproxy->net_ns != &init_net) |
1413 | return -EOPNOTSUPP; | |
1414 | ||
9fb9cbb1 YK |
1415 | /* On boot, we can set this without any fancy locking. */ |
1416 | if (!nf_conntrack_htable_size) | |
1417 | return param_set_uint(val, kp); | |
1418 | ||
96eb24d7 | 1419 | hashsize = simple_strtoul(val, NULL, 0); |
9fb9cbb1 YK |
1420 | if (!hashsize) |
1421 | return -EINVAL; | |
1422 | ||
d862a662 | 1423 | hash = nf_ct_alloc_hashtable(&hashsize, 1); |
9fb9cbb1 YK |
1424 | if (!hash) |
1425 | return -ENOMEM; | |
1426 | ||
76507f69 PM |
1427 | /* Lookups in the old hash might happen in parallel, which means we |
1428 | * might get false negatives during connection lookup. New connections | |
1429 | * created because of a false negative won't make it into the hash | |
1430 | * though since that required taking the lock. | |
1431 | */ | |
f8ba1aff | 1432 | spin_lock_bh(&nf_conntrack_lock); |
d696c7bd | 1433 | for (i = 0; i < init_net.ct.htable_size; i++) { |
ea781f19 ED |
1434 | while (!hlist_nulls_empty(&init_net.ct.hash[i])) { |
1435 | h = hlist_nulls_entry(init_net.ct.hash[i].first, | |
1436 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1437 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1438 | hlist_nulls_del_rcu(&h->hnnode); |
5d0aa2cc | 1439 | bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct), |
99f07e91 | 1440 | hashsize); |
ea781f19 | 1441 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1442 | } |
1443 | } | |
d696c7bd | 1444 | old_size = init_net.ct.htable_size; |
400dad39 | 1445 | old_hash = init_net.ct.hash; |
9fb9cbb1 | 1446 | |
d696c7bd | 1447 | init_net.ct.htable_size = nf_conntrack_htable_size = hashsize; |
400dad39 | 1448 | init_net.ct.hash = hash; |
f8ba1aff | 1449 | spin_unlock_bh(&nf_conntrack_lock); |
9fb9cbb1 | 1450 | |
d862a662 | 1451 | nf_ct_free_hashtable(old_hash, old_size); |
9fb9cbb1 YK |
1452 | return 0; |
1453 | } | |
fae718dd | 1454 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1455 | |
fae718dd | 1456 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1457 | &nf_conntrack_htable_size, 0600); |
1458 | ||
5bfddbd4 ED |
1459 | void nf_ct_untracked_status_or(unsigned long bits) |
1460 | { | |
b3c5163f ED |
1461 | int cpu; |
1462 | ||
1463 | for_each_possible_cpu(cpu) | |
1464 | per_cpu(nf_conntrack_untracked, cpu).status |= bits; | |
5bfddbd4 ED |
1465 | } |
1466 | EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); | |
1467 | ||
08f6547d | 1468 | static int nf_conntrack_init_init_net(void) |
9fb9cbb1 | 1469 | { |
f205c5e0 | 1470 | int max_factor = 8; |
b3c5163f | 1471 | int ret, cpu; |
9fb9cbb1 YK |
1472 | |
1473 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1474 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1475 | if (!nf_conntrack_htable_size) { |
1476 | nf_conntrack_htable_size | |
4481374c | 1477 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1478 | / sizeof(struct hlist_head)); |
4481374c | 1479 | if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1480 | nf_conntrack_htable_size = 16384; |
1481 | if (nf_conntrack_htable_size < 32) | |
1482 | nf_conntrack_htable_size = 32; | |
1483 | ||
1484 | /* Use a max. factor of four by default to get the same max as | |
1485 | * with the old struct list_heads. When a table size is given | |
1486 | * we use the old value of 8 to avoid reducing the max. | |
1487 | * entries. */ | |
1488 | max_factor = 4; | |
9fb9cbb1 | 1489 | } |
f205c5e0 | 1490 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 1491 | |
654d0fbd | 1492 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
1493 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
1494 | nf_conntrack_max); | |
1495 | ||
e9c1b084 PM |
1496 | ret = nf_conntrack_proto_init(); |
1497 | if (ret < 0) | |
08f6547d | 1498 | goto err_proto; |
933a41e7 | 1499 | |
ceceae1b YK |
1500 | ret = nf_conntrack_helper_init(); |
1501 | if (ret < 0) | |
08f6547d AD |
1502 | goto err_helper; |
1503 | ||
5d0aa2cc PM |
1504 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1505 | ret = nf_ct_extend_register(&nf_ct_zone_extend); | |
1506 | if (ret < 0) | |
1507 | goto err_extend; | |
1508 | #endif | |
9edd7ca0 | 1509 | /* Set up fake conntrack: to never be deleted, not in any hashes */ |
b3c5163f ED |
1510 | for_each_possible_cpu(cpu) { |
1511 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
b3c5163f ED |
1512 | write_pnet(&ct->ct_net, &init_net); |
1513 | atomic_set(&ct->ct_general.use, 1); | |
1514 | } | |
9edd7ca0 | 1515 | /* - and look it like as a confirmed connection */ |
5bfddbd4 | 1516 | nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); |
08f6547d AD |
1517 | return 0; |
1518 | ||
5d0aa2cc PM |
1519 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1520 | err_extend: | |
1521 | nf_conntrack_helper_fini(); | |
1522 | #endif | |
08f6547d AD |
1523 | err_helper: |
1524 | nf_conntrack_proto_fini(); | |
1525 | err_proto: | |
08f6547d AD |
1526 | return ret; |
1527 | } | |
1528 | ||
8cc20198 ED |
1529 | /* |
1530 | * We need to use special "null" values, not used in hash table | |
1531 | */ | |
1532 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1533 | #define DYING_NULLS_VAL ((1<<30)+1) | |
1534 | ||
08f6547d AD |
1535 | static int nf_conntrack_init_net(struct net *net) |
1536 | { | |
1537 | int ret; | |
ceceae1b | 1538 | |
08f6547d | 1539 | atomic_set(&net->ct.count, 0); |
8cc20198 ED |
1540 | INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL); |
1541 | INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL); | |
08f6547d AD |
1542 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1543 | if (!net->ct.stat) { | |
1544 | ret = -ENOMEM; | |
1545 | goto err_stat; | |
1546 | } | |
5b3501fa ED |
1547 | |
1548 | net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net); | |
1549 | if (!net->ct.slabname) { | |
1550 | ret = -ENOMEM; | |
1551 | goto err_slabname; | |
1552 | } | |
1553 | ||
1554 | net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname, | |
1555 | sizeof(struct nf_conn), 0, | |
1556 | SLAB_DESTROY_BY_RCU, NULL); | |
1557 | if (!net->ct.nf_conntrack_cachep) { | |
1558 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); | |
1559 | ret = -ENOMEM; | |
1560 | goto err_cache; | |
1561 | } | |
d696c7bd PM |
1562 | |
1563 | net->ct.htable_size = nf_conntrack_htable_size; | |
d862a662 | 1564 | net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1); |
08f6547d AD |
1565 | if (!net->ct.hash) { |
1566 | ret = -ENOMEM; | |
1567 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); | |
1568 | goto err_hash; | |
1569 | } | |
1570 | ret = nf_conntrack_expect_init(net); | |
1571 | if (ret < 0) | |
1572 | goto err_expect; | |
d716a4df | 1573 | ret = nf_conntrack_acct_init(net); |
58401572 | 1574 | if (ret < 0) |
08f6547d | 1575 | goto err_acct; |
a992ca2a PNA |
1576 | ret = nf_conntrack_tstamp_init(net); |
1577 | if (ret < 0) | |
1578 | goto err_tstamp; | |
a0891aa6 PNA |
1579 | ret = nf_conntrack_ecache_init(net); |
1580 | if (ret < 0) | |
1581 | goto err_ecache; | |
dd705072 PNA |
1582 | ret = nf_conntrack_timeout_init(net); |
1583 | if (ret < 0) | |
1584 | goto err_timeout; | |
7d3cdc6b | 1585 | |
08f6547d | 1586 | return 0; |
9fb9cbb1 | 1587 | |
dd705072 PNA |
1588 | err_timeout: |
1589 | nf_conntrack_timeout_fini(net); | |
a0891aa6 | 1590 | err_ecache: |
a992ca2a PNA |
1591 | nf_conntrack_tstamp_fini(net); |
1592 | err_tstamp: | |
a0891aa6 | 1593 | nf_conntrack_acct_fini(net); |
08f6547d | 1594 | err_acct: |
9b03f38d | 1595 | nf_conntrack_expect_fini(net); |
08f6547d | 1596 | err_expect: |
d862a662 | 1597 | nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size); |
6058fa6b | 1598 | err_hash: |
5b3501fa ED |
1599 | kmem_cache_destroy(net->ct.nf_conntrack_cachep); |
1600 | err_cache: | |
1601 | kfree(net->ct.slabname); | |
1602 | err_slabname: | |
0d55af87 AD |
1603 | free_percpu(net->ct.stat); |
1604 | err_stat: | |
08f6547d AD |
1605 | return ret; |
1606 | } | |
1607 | ||
f9dd09c7 JK |
1608 | s16 (*nf_ct_nat_offset)(const struct nf_conn *ct, |
1609 | enum ip_conntrack_dir dir, | |
1610 | u32 seq); | |
1611 | EXPORT_SYMBOL_GPL(nf_ct_nat_offset); | |
1612 | ||
08f6547d AD |
1613 | int nf_conntrack_init(struct net *net) |
1614 | { | |
1615 | int ret; | |
1616 | ||
1617 | if (net_eq(net, &init_net)) { | |
1618 | ret = nf_conntrack_init_init_net(); | |
1619 | if (ret < 0) | |
1620 | goto out_init_net; | |
1621 | } | |
1622 | ret = nf_conntrack_init_net(net); | |
1623 | if (ret < 0) | |
1624 | goto out_net; | |
1625 | ||
1626 | if (net_eq(net, &init_net)) { | |
1627 | /* For use by REJECT target */ | |
a9b3cd7f SH |
1628 | RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); |
1629 | RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); | |
f9dd09c7 JK |
1630 | |
1631 | /* Howto get NAT offsets */ | |
a9b3cd7f | 1632 | RCU_INIT_POINTER(nf_ct_nat_offset, NULL); |
08f6547d AD |
1633 | } |
1634 | return 0; | |
1635 | ||
1636 | out_net: | |
1637 | if (net_eq(net, &init_net)) | |
1638 | nf_conntrack_cleanup_init_net(); | |
1639 | out_init_net: | |
1640 | return ret; | |
9fb9cbb1 | 1641 | } |