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