]>
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 | 7 | * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org> |
f229f6ce | 8 | * (C) 2005-2012 Patrick McHardy <kaber@trash.net> |
9fb9cbb1 YK |
9 | * |
10 | * This program is free software; you can redistribute it and/or modify | |
11 | * it under the terms of the GNU General Public License version 2 as | |
12 | * published by the Free Software Foundation. | |
9fb9cbb1 YK |
13 | */ |
14 | ||
9fb9cbb1 YK |
15 | #include <linux/types.h> |
16 | #include <linux/netfilter.h> | |
17 | #include <linux/module.h> | |
d43c36dc | 18 | #include <linux/sched.h> |
9fb9cbb1 YK |
19 | #include <linux/skbuff.h> |
20 | #include <linux/proc_fs.h> | |
21 | #include <linux/vmalloc.h> | |
22 | #include <linux/stddef.h> | |
23 | #include <linux/slab.h> | |
24 | #include <linux/random.h> | |
25 | #include <linux/jhash.h> | |
26 | #include <linux/err.h> | |
27 | #include <linux/percpu.h> | |
28 | #include <linux/moduleparam.h> | |
29 | #include <linux/notifier.h> | |
30 | #include <linux/kernel.h> | |
31 | #include <linux/netdevice.h> | |
32 | #include <linux/socket.h> | |
d7fe0f24 | 33 | #include <linux/mm.h> |
d696c7bd | 34 | #include <linux/nsproxy.h> |
ea781f19 | 35 | #include <linux/rculist_nulls.h> |
9fb9cbb1 | 36 | |
9fb9cbb1 YK |
37 | #include <net/netfilter/nf_conntrack.h> |
38 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
605dcad6 | 39 | #include <net/netfilter/nf_conntrack_l4proto.h> |
77ab9cff | 40 | #include <net/netfilter/nf_conntrack_expect.h> |
9fb9cbb1 | 41 | #include <net/netfilter/nf_conntrack_helper.h> |
41d73ec0 | 42 | #include <net/netfilter/nf_conntrack_seqadj.h> |
9fb9cbb1 | 43 | #include <net/netfilter/nf_conntrack_core.h> |
ecfab2c9 | 44 | #include <net/netfilter/nf_conntrack_extend.h> |
58401572 | 45 | #include <net/netfilter/nf_conntrack_acct.h> |
a0891aa6 | 46 | #include <net/netfilter/nf_conntrack_ecache.h> |
5d0aa2cc | 47 | #include <net/netfilter/nf_conntrack_zones.h> |
a992ca2a | 48 | #include <net/netfilter/nf_conntrack_timestamp.h> |
dd705072 | 49 | #include <net/netfilter/nf_conntrack_timeout.h> |
c539f017 | 50 | #include <net/netfilter/nf_conntrack_labels.h> |
48b1de4c | 51 | #include <net/netfilter/nf_conntrack_synproxy.h> |
e6a7d3c0 | 52 | #include <net/netfilter/nf_nat.h> |
e17b666a | 53 | #include <net/netfilter/nf_nat_core.h> |
49376368 | 54 | #include <net/netfilter/nf_nat_helper.h> |
9fb9cbb1 | 55 | |
dc808fe2 | 56 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 57 | |
e17b666a PM |
58 | int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, |
59 | enum nf_nat_manip_type manip, | |
39938324 | 60 | const struct nlattr *attr) __read_mostly; |
e6a7d3c0 PNA |
61 | EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); |
62 | ||
93bb0ceb JDB |
63 | __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS]; |
64 | EXPORT_SYMBOL_GPL(nf_conntrack_locks); | |
9fb9cbb1 | 65 | |
ca7433df JDB |
66 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock); |
67 | EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock); | |
68 | ||
93bb0ceb JDB |
69 | static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2) |
70 | { | |
71 | h1 %= CONNTRACK_LOCKS; | |
72 | h2 %= CONNTRACK_LOCKS; | |
73 | spin_unlock(&nf_conntrack_locks[h1]); | |
74 | if (h1 != h2) | |
75 | spin_unlock(&nf_conntrack_locks[h2]); | |
76 | } | |
77 | ||
78 | /* return true if we need to recompute hashes (in case hash table was resized) */ | |
79 | static bool nf_conntrack_double_lock(struct net *net, unsigned int h1, | |
80 | unsigned int h2, unsigned int sequence) | |
81 | { | |
82 | h1 %= CONNTRACK_LOCKS; | |
83 | h2 %= CONNTRACK_LOCKS; | |
84 | if (h1 <= h2) { | |
85 | spin_lock(&nf_conntrack_locks[h1]); | |
86 | if (h1 != h2) | |
87 | spin_lock_nested(&nf_conntrack_locks[h2], | |
88 | SINGLE_DEPTH_NESTING); | |
89 | } else { | |
90 | spin_lock(&nf_conntrack_locks[h2]); | |
91 | spin_lock_nested(&nf_conntrack_locks[h1], | |
92 | SINGLE_DEPTH_NESTING); | |
93 | } | |
94 | if (read_seqcount_retry(&net->ct.generation, sequence)) { | |
95 | nf_conntrack_double_unlock(h1, h2); | |
96 | return true; | |
97 | } | |
98 | return false; | |
99 | } | |
100 | ||
101 | static void nf_conntrack_all_lock(void) | |
102 | { | |
103 | int i; | |
104 | ||
105 | for (i = 0; i < CONNTRACK_LOCKS; i++) | |
106 | spin_lock_nested(&nf_conntrack_locks[i], i); | |
107 | } | |
108 | ||
109 | static void nf_conntrack_all_unlock(void) | |
110 | { | |
111 | int i; | |
112 | ||
113 | for (i = 0; i < CONNTRACK_LOCKS; i++) | |
114 | spin_unlock(&nf_conntrack_locks[i]); | |
115 | } | |
116 | ||
e2b7606c | 117 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
118 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
119 | ||
e478075c | 120 | unsigned int nf_conntrack_max __read_mostly; |
a999e683 | 121 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 122 | |
b3c5163f ED |
123 | DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked); |
124 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked); | |
13b18339 | 125 | |
f682cefa | 126 | unsigned int nf_conntrack_hash_rnd __read_mostly; |
4d4e61c6 | 127 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd); |
9fb9cbb1 | 128 | |
99f07e91 | 129 | static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone) |
9fb9cbb1 | 130 | { |
0794935e | 131 | unsigned int n; |
0794935e PM |
132 | |
133 | /* The direction must be ignored, so we hash everything up to the | |
134 | * destination ports (which is a multiple of 4) and treat the last | |
135 | * three bytes manually. | |
136 | */ | |
137 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); | |
99f07e91 CG |
138 | return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^ |
139 | (((__force __u16)tuple->dst.u.all << 16) | | |
140 | tuple->dst.protonum)); | |
141 | } | |
142 | ||
143 | static u32 __hash_bucket(u32 hash, unsigned int size) | |
144 | { | |
145 | return ((u64)hash * size) >> 32; | |
146 | } | |
147 | ||
148 | static u32 hash_bucket(u32 hash, const struct net *net) | |
149 | { | |
150 | return __hash_bucket(hash, net->ct.htable_size); | |
151 | } | |
0794935e | 152 | |
99f07e91 CG |
153 | static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, |
154 | u16 zone, unsigned int size) | |
155 | { | |
156 | return __hash_bucket(hash_conntrack_raw(tuple, zone), size); | |
9fb9cbb1 YK |
157 | } |
158 | ||
5d0aa2cc | 159 | static inline u_int32_t hash_conntrack(const struct net *net, u16 zone, |
d696c7bd | 160 | const struct nf_conntrack_tuple *tuple) |
9fb9cbb1 | 161 | { |
99f07e91 | 162 | return __hash_conntrack(tuple, zone, net->ct.htable_size); |
9fb9cbb1 YK |
163 | } |
164 | ||
5f2b4c90 | 165 | bool |
9fb9cbb1 YK |
166 | nf_ct_get_tuple(const struct sk_buff *skb, |
167 | unsigned int nhoff, | |
168 | unsigned int dataoff, | |
169 | u_int16_t l3num, | |
170 | u_int8_t protonum, | |
171 | struct nf_conntrack_tuple *tuple, | |
172 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 173 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 174 | { |
443a70d5 | 175 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
176 | |
177 | tuple->src.l3num = l3num; | |
178 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 179 | return false; |
9fb9cbb1 YK |
180 | |
181 | tuple->dst.protonum = protonum; | |
182 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
183 | ||
605dcad6 | 184 | return l4proto->pkt_to_tuple(skb, dataoff, tuple); |
9fb9cbb1 | 185 | } |
13b18339 | 186 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 187 | |
5f2b4c90 JE |
188 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
189 | u_int16_t l3num, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
190 | { |
191 | struct nf_conntrack_l3proto *l3proto; | |
192 | struct nf_conntrack_l4proto *l4proto; | |
193 | unsigned int protoff; | |
194 | u_int8_t protonum; | |
195 | int ret; | |
196 | ||
197 | rcu_read_lock(); | |
198 | ||
199 | l3proto = __nf_ct_l3proto_find(l3num); | |
200 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
201 | if (ret != NF_ACCEPT) { | |
202 | rcu_read_unlock(); | |
5f2b4c90 | 203 | return false; |
e2a3123f YK |
204 | } |
205 | ||
206 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
207 | ||
208 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, | |
209 | l3proto, l4proto); | |
210 | ||
211 | rcu_read_unlock(); | |
212 | return ret; | |
213 | } | |
214 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
215 | ||
5f2b4c90 | 216 | bool |
9fb9cbb1 YK |
217 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
218 | const struct nf_conntrack_tuple *orig, | |
219 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 220 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 221 | { |
443a70d5 | 222 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
223 | |
224 | inverse->src.l3num = orig->src.l3num; | |
225 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 226 | return false; |
9fb9cbb1 YK |
227 | |
228 | inverse->dst.dir = !orig->dst.dir; | |
229 | ||
230 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 231 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 232 | } |
13b18339 | 233 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 234 | |
9fb9cbb1 YK |
235 | static void |
236 | clean_from_lists(struct nf_conn *ct) | |
237 | { | |
0d53778e | 238 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
239 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
240 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
241 | |
242 | /* Destroy all pending expectations */ | |
c1d10adb | 243 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
244 | } |
245 | ||
b7779d06 JDB |
246 | /* must be called with local_bh_disable */ |
247 | static void nf_ct_add_to_dying_list(struct nf_conn *ct) | |
248 | { | |
249 | struct ct_pcpu *pcpu; | |
250 | ||
251 | /* add this conntrack to the (per cpu) dying list */ | |
252 | ct->cpu = smp_processor_id(); | |
253 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
254 | ||
255 | spin_lock(&pcpu->lock); | |
256 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
257 | &pcpu->dying); | |
258 | spin_unlock(&pcpu->lock); | |
259 | } | |
260 | ||
261 | /* must be called with local_bh_disable */ | |
262 | static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct) | |
263 | { | |
264 | struct ct_pcpu *pcpu; | |
265 | ||
266 | /* add this conntrack to the (per cpu) unconfirmed list */ | |
267 | ct->cpu = smp_processor_id(); | |
268 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
269 | ||
270 | spin_lock(&pcpu->lock); | |
271 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
272 | &pcpu->unconfirmed); | |
273 | spin_unlock(&pcpu->lock); | |
274 | } | |
275 | ||
276 | /* must be called with local_bh_disable */ | |
277 | static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct) | |
278 | { | |
279 | struct ct_pcpu *pcpu; | |
280 | ||
281 | /* We overload first tuple to link into unconfirmed or dying list.*/ | |
282 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
283 | ||
284 | spin_lock(&pcpu->lock); | |
285 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); | |
286 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
287 | spin_unlock(&pcpu->lock); | |
288 | } | |
289 | ||
9fb9cbb1 YK |
290 | static void |
291 | destroy_conntrack(struct nf_conntrack *nfct) | |
292 | { | |
293 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
0d55af87 | 294 | struct net *net = nf_ct_net(ct); |
605dcad6 | 295 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 296 | |
0d53778e | 297 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
298 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
299 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
300 | ||
923f4902 | 301 | rcu_read_lock(); |
5e8fbe2a | 302 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
605dcad6 MJ |
303 | if (l4proto && l4proto->destroy) |
304 | l4proto->destroy(ct); | |
9fb9cbb1 | 305 | |
982d9a9c | 306 | rcu_read_unlock(); |
9fb9cbb1 | 307 | |
ca7433df | 308 | local_bh_disable(); |
9fb9cbb1 YK |
309 | /* Expectations will have been removed in clean_from_lists, |
310 | * except TFTP can create an expectation on the first packet, | |
311 | * before connection is in the list, so we need to clean here, | |
ca7433df JDB |
312 | * too. |
313 | */ | |
c1d10adb | 314 | nf_ct_remove_expectations(ct); |
9fb9cbb1 | 315 | |
b7779d06 | 316 | nf_ct_del_from_dying_or_unconfirmed_list(ct); |
9fb9cbb1 | 317 | |
0d55af87 | 318 | NF_CT_STAT_INC(net, delete); |
ca7433df | 319 | local_bh_enable(); |
9fb9cbb1 YK |
320 | |
321 | if (ct->master) | |
322 | nf_ct_put(ct->master); | |
323 | ||
0d53778e | 324 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
325 | nf_conntrack_free(ct); |
326 | } | |
327 | ||
02982c27 | 328 | static void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 329 | { |
0d55af87 | 330 | struct net *net = nf_ct_net(ct); |
93bb0ceb JDB |
331 | unsigned int hash, reply_hash; |
332 | u16 zone = nf_ct_zone(ct); | |
333 | unsigned int sequence; | |
9fb9cbb1 | 334 | |
9858a3ae | 335 | nf_ct_helper_destroy(ct); |
93bb0ceb JDB |
336 | |
337 | local_bh_disable(); | |
338 | do { | |
339 | sequence = read_seqcount_begin(&net->ct.generation); | |
340 | hash = hash_conntrack(net, zone, | |
341 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
342 | reply_hash = hash_conntrack(net, zone, | |
343 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
344 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
345 | ||
9fb9cbb1 | 346 | clean_from_lists(ct); |
93bb0ceb JDB |
347 | nf_conntrack_double_unlock(hash, reply_hash); |
348 | ||
b7779d06 | 349 | nf_ct_add_to_dying_list(ct); |
93bb0ceb JDB |
350 | |
351 | NF_CT_STAT_INC(net, delete_list); | |
352 | local_bh_enable(); | |
dd7669a9 | 353 | } |
dd7669a9 | 354 | |
02982c27 | 355 | bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report) |
dd7669a9 | 356 | { |
a992ca2a PNA |
357 | struct nf_conn_tstamp *tstamp; |
358 | ||
359 | tstamp = nf_conn_tstamp_find(ct); | |
360 | if (tstamp && tstamp->stop == 0) | |
d2de875c | 361 | tstamp->stop = ktime_get_real_ns(); |
dd7669a9 | 362 | |
9500507c FW |
363 | if (nf_ct_is_dying(ct)) |
364 | goto delete; | |
365 | ||
366 | if (nf_conntrack_event_report(IPCT_DESTROY, ct, | |
367 | portid, report) < 0) { | |
dd7669a9 PNA |
368 | /* destroy event was not delivered */ |
369 | nf_ct_delete_from_lists(ct); | |
9500507c | 370 | nf_conntrack_ecache_delayed_work(nf_ct_net(ct)); |
02982c27 | 371 | return false; |
dd7669a9 | 372 | } |
9500507c FW |
373 | |
374 | nf_conntrack_ecache_work(nf_ct_net(ct)); | |
dd7669a9 | 375 | set_bit(IPS_DYING_BIT, &ct->status); |
9500507c | 376 | delete: |
dd7669a9 | 377 | nf_ct_delete_from_lists(ct); |
9fb9cbb1 | 378 | nf_ct_put(ct); |
02982c27 FW |
379 | return true; |
380 | } | |
381 | EXPORT_SYMBOL_GPL(nf_ct_delete); | |
382 | ||
383 | static void death_by_timeout(unsigned long ul_conntrack) | |
384 | { | |
385 | nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0); | |
9fb9cbb1 YK |
386 | } |
387 | ||
c6825c09 AV |
388 | static inline bool |
389 | nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, | |
390 | const struct nf_conntrack_tuple *tuple, | |
391 | u16 zone) | |
392 | { | |
393 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
394 | ||
395 | /* A conntrack can be recreated with the equal tuple, | |
396 | * so we need to check that the conntrack is confirmed | |
397 | */ | |
398 | return nf_ct_tuple_equal(tuple, &h->tuple) && | |
399 | nf_ct_zone(ct) == zone && | |
400 | nf_ct_is_confirmed(ct); | |
401 | } | |
402 | ||
ea781f19 ED |
403 | /* |
404 | * Warning : | |
405 | * - Caller must take a reference on returned object | |
406 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
ea781f19 | 407 | */ |
99f07e91 CG |
408 | static struct nf_conntrack_tuple_hash * |
409 | ____nf_conntrack_find(struct net *net, u16 zone, | |
410 | const struct nf_conntrack_tuple *tuple, u32 hash) | |
9fb9cbb1 YK |
411 | { |
412 | struct nf_conntrack_tuple_hash *h; | |
ea781f19 | 413 | struct hlist_nulls_node *n; |
99f07e91 | 414 | unsigned int bucket = hash_bucket(hash, net); |
9fb9cbb1 | 415 | |
4e29e9ec PM |
416 | /* Disable BHs the entire time since we normally need to disable them |
417 | * at least once for the stats anyway. | |
418 | */ | |
419 | local_bh_disable(); | |
ea781f19 | 420 | begin: |
99f07e91 | 421 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) { |
c6825c09 | 422 | if (nf_ct_key_equal(h, tuple, zone)) { |
0d55af87 | 423 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 424 | local_bh_enable(); |
9fb9cbb1 YK |
425 | return h; |
426 | } | |
0d55af87 | 427 | NF_CT_STAT_INC(net, searched); |
9fb9cbb1 | 428 | } |
ea781f19 ED |
429 | /* |
430 | * if the nulls value we got at the end of this lookup is | |
431 | * not the expected one, we must restart lookup. | |
432 | * We probably met an item that was moved to another chain. | |
433 | */ | |
99f07e91 | 434 | if (get_nulls_value(n) != bucket) { |
af740b2c | 435 | NF_CT_STAT_INC(net, search_restart); |
ea781f19 | 436 | goto begin; |
af740b2c | 437 | } |
4e29e9ec | 438 | local_bh_enable(); |
9fb9cbb1 YK |
439 | |
440 | return NULL; | |
441 | } | |
99f07e91 | 442 | |
9fb9cbb1 | 443 | /* Find a connection corresponding to a tuple. */ |
99f07e91 CG |
444 | static struct nf_conntrack_tuple_hash * |
445 | __nf_conntrack_find_get(struct net *net, u16 zone, | |
446 | const struct nf_conntrack_tuple *tuple, u32 hash) | |
9fb9cbb1 YK |
447 | { |
448 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 449 | struct nf_conn *ct; |
9fb9cbb1 | 450 | |
76507f69 | 451 | rcu_read_lock(); |
ea781f19 | 452 | begin: |
99f07e91 | 453 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
454 | if (h) { |
455 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
456 | if (unlikely(nf_ct_is_dying(ct) || |
457 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 458 | h = NULL; |
ea781f19 | 459 | else { |
c6825c09 | 460 | if (unlikely(!nf_ct_key_equal(h, tuple, zone))) { |
ea781f19 ED |
461 | nf_ct_put(ct); |
462 | goto begin; | |
463 | } | |
464 | } | |
76507f69 PM |
465 | } |
466 | rcu_read_unlock(); | |
9fb9cbb1 YK |
467 | |
468 | return h; | |
469 | } | |
99f07e91 CG |
470 | |
471 | struct nf_conntrack_tuple_hash * | |
472 | nf_conntrack_find_get(struct net *net, u16 zone, | |
473 | const struct nf_conntrack_tuple *tuple) | |
474 | { | |
475 | return __nf_conntrack_find_get(net, zone, tuple, | |
476 | hash_conntrack_raw(tuple, zone)); | |
477 | } | |
13b18339 | 478 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 479 | |
c1d10adb PNA |
480 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
481 | unsigned int hash, | |
b476b72a | 482 | unsigned int reply_hash) |
c1d10adb | 483 | { |
400dad39 AD |
484 | struct net *net = nf_ct_net(ct); |
485 | ||
ea781f19 | 486 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
400dad39 | 487 | &net->ct.hash[hash]); |
ea781f19 | 488 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
b476b72a | 489 | &net->ct.hash[reply_hash]); |
c1d10adb PNA |
490 | } |
491 | ||
7d367e06 JK |
492 | int |
493 | nf_conntrack_hash_check_insert(struct nf_conn *ct) | |
c1d10adb | 494 | { |
d696c7bd | 495 | struct net *net = nf_ct_net(ct); |
b476b72a | 496 | unsigned int hash, reply_hash; |
7d367e06 JK |
497 | struct nf_conntrack_tuple_hash *h; |
498 | struct hlist_nulls_node *n; | |
5d0aa2cc | 499 | u16 zone; |
93bb0ceb | 500 | unsigned int sequence; |
c1d10adb | 501 | |
5d0aa2cc | 502 | zone = nf_ct_zone(ct); |
7d367e06 | 503 | |
93bb0ceb JDB |
504 | local_bh_disable(); |
505 | do { | |
506 | sequence = read_seqcount_begin(&net->ct.generation); | |
507 | hash = hash_conntrack(net, zone, | |
508 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
509 | reply_hash = hash_conntrack(net, zone, | |
510 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
511 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
7d367e06 JK |
512 | |
513 | /* See if there's one in the list already, including reverse */ | |
514 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) | |
515 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, | |
516 | &h->tuple) && | |
517 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
518 | goto out; | |
b476b72a | 519 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode) |
7d367e06 JK |
520 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
521 | &h->tuple) && | |
522 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
523 | goto out; | |
c1d10adb | 524 | |
7d367e06 | 525 | add_timer(&ct->timeout); |
e53376be PNA |
526 | smp_wmb(); |
527 | /* The caller holds a reference to this object */ | |
528 | atomic_set(&ct->ct_general.use, 2); | |
b476b72a | 529 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 530 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 531 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 532 | local_bh_enable(); |
7d367e06 JK |
533 | return 0; |
534 | ||
535 | out: | |
93bb0ceb | 536 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 537 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 538 | local_bh_enable(); |
7d367e06 | 539 | return -EEXIST; |
c1d10adb | 540 | } |
7d367e06 | 541 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); |
c1d10adb | 542 | |
e53376be PNA |
543 | /* deletion from this larval template list happens via nf_ct_put() */ |
544 | void nf_conntrack_tmpl_insert(struct net *net, struct nf_conn *tmpl) | |
545 | { | |
b7779d06 JDB |
546 | struct ct_pcpu *pcpu; |
547 | ||
e53376be PNA |
548 | __set_bit(IPS_TEMPLATE_BIT, &tmpl->status); |
549 | __set_bit(IPS_CONFIRMED_BIT, &tmpl->status); | |
550 | nf_conntrack_get(&tmpl->ct_general); | |
551 | ||
b7779d06 JDB |
552 | /* add this conntrack to the (per cpu) tmpl list */ |
553 | local_bh_disable(); | |
554 | tmpl->cpu = smp_processor_id(); | |
555 | pcpu = per_cpu_ptr(nf_ct_net(tmpl)->ct.pcpu_lists, tmpl->cpu); | |
556 | ||
557 | spin_lock(&pcpu->lock); | |
e53376be PNA |
558 | /* Overload tuple linked list to put us in template list. */ |
559 | hlist_nulls_add_head_rcu(&tmpl->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
b7779d06 JDB |
560 | &pcpu->tmpl); |
561 | spin_unlock_bh(&pcpu->lock); | |
e53376be PNA |
562 | } |
563 | EXPORT_SYMBOL_GPL(nf_conntrack_tmpl_insert); | |
564 | ||
9fb9cbb1 YK |
565 | /* Confirm a connection given skb; places it in hash table */ |
566 | int | |
3db05fea | 567 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 | 568 | { |
b476b72a | 569 | unsigned int hash, reply_hash; |
df0933dc | 570 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 571 | struct nf_conn *ct; |
df0933dc | 572 | struct nf_conn_help *help; |
a992ca2a | 573 | struct nf_conn_tstamp *tstamp; |
ea781f19 | 574 | struct hlist_nulls_node *n; |
9fb9cbb1 | 575 | enum ip_conntrack_info ctinfo; |
400dad39 | 576 | struct net *net; |
5d0aa2cc | 577 | u16 zone; |
93bb0ceb | 578 | unsigned int sequence; |
9fb9cbb1 | 579 | |
3db05fea | 580 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 581 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
582 | |
583 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
584 | ICMP/TCP RST packets in other direction. Actual packet | |
585 | which created connection will be IP_CT_NEW or for an | |
586 | expected connection, IP_CT_RELATED. */ | |
587 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
588 | return NF_ACCEPT; | |
589 | ||
5d0aa2cc | 590 | zone = nf_ct_zone(ct); |
93bb0ceb JDB |
591 | local_bh_disable(); |
592 | ||
593 | do { | |
594 | sequence = read_seqcount_begin(&net->ct.generation); | |
595 | /* reuse the hash saved before */ | |
596 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
597 | hash = hash_bucket(hash, net); | |
598 | reply_hash = hash_conntrack(net, zone, | |
599 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); | |
600 | ||
601 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
9fb9cbb1 YK |
602 | |
603 | /* We're not in hash table, and we refuse to set up related | |
93bb0ceb JDB |
604 | * connections for unconfirmed conns. But packet copies and |
605 | * REJECT will give spurious warnings here. | |
606 | */ | |
9fb9cbb1 YK |
607 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ |
608 | ||
25985edc | 609 | /* No external references means no one else could have |
93bb0ceb JDB |
610 | * confirmed us. |
611 | */ | |
9fb9cbb1 | 612 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); |
0d53778e | 613 | pr_debug("Confirming conntrack %p\n", ct); |
fc350777 JM |
614 | /* We have to check the DYING flag inside the lock to prevent |
615 | a race against nf_ct_get_next_corpse() possibly called from | |
616 | user context, else we insert an already 'dead' hash, blocking | |
617 | further use of that particular connection -JM */ | |
618 | ||
619 | if (unlikely(nf_ct_is_dying(ct))) { | |
93bb0ceb JDB |
620 | nf_conntrack_double_unlock(hash, reply_hash); |
621 | local_bh_enable(); | |
fc350777 JM |
622 | return NF_ACCEPT; |
623 | } | |
624 | ||
9fb9cbb1 YK |
625 | /* See if there's one in the list already, including reverse: |
626 | NAT could have grabbed it without realizing, since we're | |
627 | not in the hash. If there is, we lost race. */ | |
ea781f19 | 628 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) |
df0933dc | 629 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
5d0aa2cc PM |
630 | &h->tuple) && |
631 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
df0933dc | 632 | goto out; |
b476b72a | 633 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode) |
df0933dc | 634 | if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
5d0aa2cc PM |
635 | &h->tuple) && |
636 | zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) | |
df0933dc | 637 | goto out; |
9fb9cbb1 | 638 | |
b7779d06 | 639 | nf_ct_del_from_dying_or_unconfirmed_list(ct); |
df0933dc | 640 | |
df0933dc PM |
641 | /* Timer relative to confirmation time, not original |
642 | setting time, otherwise we'd get timer wrap in | |
643 | weird delay cases. */ | |
644 | ct->timeout.expires += jiffies; | |
645 | add_timer(&ct->timeout); | |
646 | atomic_inc(&ct->ct_general.use); | |
45eec341 | 647 | ct->status |= IPS_CONFIRMED; |
5c8ec910 | 648 | |
a992ca2a PNA |
649 | /* set conntrack timestamp, if enabled. */ |
650 | tstamp = nf_conn_tstamp_find(ct); | |
651 | if (tstamp) { | |
652 | if (skb->tstamp.tv64 == 0) | |
e3192690 | 653 | __net_timestamp(skb); |
a992ca2a PNA |
654 | |
655 | tstamp->start = ktime_to_ns(skb->tstamp); | |
656 | } | |
5c8ec910 PM |
657 | /* Since the lookup is lockless, hash insertion must be done after |
658 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
659 | * guarantee that no other CPU can find the conntrack before the above | |
660 | * stores are visible. | |
661 | */ | |
b476b72a | 662 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 663 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 664 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 665 | local_bh_enable(); |
5c8ec910 | 666 | |
df0933dc PM |
667 | help = nfct_help(ct); |
668 | if (help && help->helper) | |
a71996fc | 669 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 670 | |
df0933dc | 671 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 672 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 673 | return NF_ACCEPT; |
9fb9cbb1 | 674 | |
df0933dc | 675 | out: |
93bb0ceb | 676 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 677 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 678 | local_bh_enable(); |
9fb9cbb1 YK |
679 | return NF_DROP; |
680 | } | |
13b18339 | 681 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
682 | |
683 | /* Returns true if a connection correspondings to the tuple (required | |
684 | for NAT). */ | |
685 | int | |
686 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
687 | const struct nf_conn *ignored_conntrack) | |
688 | { | |
400dad39 | 689 | struct net *net = nf_ct_net(ignored_conntrack); |
9fb9cbb1 | 690 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 691 | struct hlist_nulls_node *n; |
5d0aa2cc PM |
692 | struct nf_conn *ct; |
693 | u16 zone = nf_ct_zone(ignored_conntrack); | |
694 | unsigned int hash = hash_conntrack(net, zone, tuple); | |
9fb9cbb1 | 695 | |
4e29e9ec PM |
696 | /* Disable BHs the entire time since we need to disable them at |
697 | * least once for the stats anyway. | |
698 | */ | |
699 | rcu_read_lock_bh(); | |
ea781f19 | 700 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { |
5d0aa2cc PM |
701 | ct = nf_ct_tuplehash_to_ctrack(h); |
702 | if (ct != ignored_conntrack && | |
703 | nf_ct_tuple_equal(tuple, &h->tuple) && | |
704 | nf_ct_zone(ct) == zone) { | |
0d55af87 | 705 | NF_CT_STAT_INC(net, found); |
4e29e9ec | 706 | rcu_read_unlock_bh(); |
ba419aff PM |
707 | return 1; |
708 | } | |
0d55af87 | 709 | NF_CT_STAT_INC(net, searched); |
ba419aff | 710 | } |
4e29e9ec | 711 | rcu_read_unlock_bh(); |
9fb9cbb1 | 712 | |
ba419aff | 713 | return 0; |
9fb9cbb1 | 714 | } |
13b18339 | 715 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 716 | |
7ae7730f PM |
717 | #define NF_CT_EVICTION_RANGE 8 |
718 | ||
9fb9cbb1 YK |
719 | /* There's a small race here where we may free a just-assured |
720 | connection. Too bad: we're in trouble anyway. */ | |
93bb0ceb | 721 | static noinline int early_drop(struct net *net, unsigned int _hash) |
9fb9cbb1 | 722 | { |
f205c5e0 | 723 | /* Use oldest entry, which is roughly LRU */ |
9fb9cbb1 | 724 | struct nf_conntrack_tuple_hash *h; |
df0933dc | 725 | struct nf_conn *ct = NULL, *tmp; |
ea781f19 | 726 | struct hlist_nulls_node *n; |
93bb0ceb | 727 | unsigned int i = 0, cnt = 0; |
9fb9cbb1 | 728 | int dropped = 0; |
93bb0ceb JDB |
729 | unsigned int hash, sequence; |
730 | spinlock_t *lockp; | |
9fb9cbb1 | 731 | |
93bb0ceb JDB |
732 | local_bh_disable(); |
733 | restart: | |
734 | sequence = read_seqcount_begin(&net->ct.generation); | |
735 | hash = hash_bucket(_hash, net); | |
736 | for (; i < net->ct.htable_size; i++) { | |
737 | lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS]; | |
738 | spin_lock(lockp); | |
739 | if (read_seqcount_retry(&net->ct.generation, sequence)) { | |
740 | spin_unlock(lockp); | |
741 | goto restart; | |
742 | } | |
ea781f19 ED |
743 | hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], |
744 | hnnode) { | |
7ae7730f | 745 | tmp = nf_ct_tuplehash_to_ctrack(h); |
93bb0ceb JDB |
746 | if (!test_bit(IPS_ASSURED_BIT, &tmp->status) && |
747 | !nf_ct_is_dying(tmp) && | |
748 | atomic_inc_not_zero(&tmp->ct_general.use)) { | |
7ae7730f | 749 | ct = tmp; |
93bb0ceb JDB |
750 | break; |
751 | } | |
7ae7730f PM |
752 | cnt++; |
753 | } | |
76507f69 | 754 | |
93bb0ceb JDB |
755 | hash = (hash + 1) % net->ct.htable_size; |
756 | spin_unlock(lockp); | |
5ae27aa2 | 757 | |
93bb0ceb | 758 | if (ct || cnt >= NF_CT_EVICTION_RANGE) |
7ae7730f | 759 | break; |
5ae27aa2 | 760 | |
9fb9cbb1 | 761 | } |
93bb0ceb | 762 | local_bh_enable(); |
9fb9cbb1 YK |
763 | |
764 | if (!ct) | |
765 | return dropped; | |
766 | ||
767 | if (del_timer(&ct->timeout)) { | |
02982c27 | 768 | if (nf_ct_delete(ct, 0, 0)) { |
74138511 PNA |
769 | dropped = 1; |
770 | NF_CT_STAT_INC_ATOMIC(net, early_drop); | |
771 | } | |
9fb9cbb1 YK |
772 | } |
773 | nf_ct_put(ct); | |
774 | return dropped; | |
775 | } | |
776 | ||
f682cefa CG |
777 | void init_nf_conntrack_hash_rnd(void) |
778 | { | |
779 | unsigned int rand; | |
780 | ||
781 | /* | |
782 | * Why not initialize nf_conntrack_rnd in a "init()" function ? | |
783 | * Because there isn't enough entropy when system initializing, | |
784 | * and we initialize it as late as possible. | |
785 | */ | |
786 | do { | |
787 | get_random_bytes(&rand, sizeof(rand)); | |
788 | } while (!rand); | |
789 | cmpxchg(&nf_conntrack_hash_rnd, 0, rand); | |
790 | } | |
791 | ||
99f07e91 CG |
792 | static struct nf_conn * |
793 | __nf_conntrack_alloc(struct net *net, u16 zone, | |
794 | const struct nf_conntrack_tuple *orig, | |
795 | const struct nf_conntrack_tuple *repl, | |
796 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 797 | { |
cd7fcbf1 | 798 | struct nf_conn *ct; |
9fb9cbb1 | 799 | |
b2390969 | 800 | if (unlikely(!nf_conntrack_hash_rnd)) { |
f682cefa | 801 | init_nf_conntrack_hash_rnd(); |
99f07e91 CG |
802 | /* recompute the hash as nf_conntrack_hash_rnd is initialized */ |
803 | hash = hash_conntrack_raw(orig, zone); | |
9fb9cbb1 YK |
804 | } |
805 | ||
5251e2d2 | 806 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 807 | atomic_inc(&net->ct.count); |
5251e2d2 | 808 | |
76eb9460 | 809 | if (nf_conntrack_max && |
49ac8713 | 810 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
93bb0ceb | 811 | if (!early_drop(net, hash)) { |
49ac8713 | 812 | atomic_dec(&net->ct.count); |
e87cc472 | 813 | net_warn_ratelimited("nf_conntrack: table full, dropping packet\n"); |
9fb9cbb1 YK |
814 | return ERR_PTR(-ENOMEM); |
815 | } | |
816 | } | |
817 | ||
941297f4 ED |
818 | /* |
819 | * Do not use kmem_cache_zalloc(), as this cache uses | |
820 | * SLAB_DESTROY_BY_RCU. | |
821 | */ | |
5b3501fa | 822 | ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp); |
c88130bc | 823 | if (ct == NULL) { |
49ac8713 | 824 | atomic_dec(&net->ct.count); |
dacd2a1a | 825 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 826 | } |
941297f4 ED |
827 | /* |
828 | * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next | |
829 | * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged. | |
830 | */ | |
831 | memset(&ct->tuplehash[IP_CT_DIR_MAX], 0, | |
e5fc9e7a CG |
832 | offsetof(struct nf_conn, proto) - |
833 | offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX])); | |
440f0d58 | 834 | spin_lock_init(&ct->lock); |
c88130bc | 835 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 836 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 837 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
838 | /* save hash for reusing when confirming */ |
839 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
9fb9cbb1 | 840 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 841 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
c2d9ba9b | 842 | write_pnet(&ct->ct_net, net); |
5d0aa2cc PM |
843 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
844 | if (zone) { | |
845 | struct nf_conntrack_zone *nf_ct_zone; | |
846 | ||
847 | nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC); | |
848 | if (!nf_ct_zone) | |
849 | goto out_free; | |
850 | nf_ct_zone->id = zone; | |
851 | } | |
852 | #endif | |
e53376be PNA |
853 | /* Because we use RCU lookups, we set ct_general.use to zero before |
854 | * this is inserted in any list. | |
941297f4 | 855 | */ |
e53376be | 856 | atomic_set(&ct->ct_general.use, 0); |
c88130bc | 857 | return ct; |
5d0aa2cc PM |
858 | |
859 | #ifdef CONFIG_NF_CONNTRACK_ZONES | |
860 | out_free: | |
d96fc659 | 861 | atomic_dec(&net->ct.count); |
5d0aa2cc PM |
862 | kmem_cache_free(net->ct.nf_conntrack_cachep, ct); |
863 | return ERR_PTR(-ENOMEM); | |
864 | #endif | |
9fb9cbb1 | 865 | } |
99f07e91 CG |
866 | |
867 | struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone, | |
868 | const struct nf_conntrack_tuple *orig, | |
869 | const struct nf_conntrack_tuple *repl, | |
870 | gfp_t gfp) | |
871 | { | |
872 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
873 | } | |
13b18339 | 874 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 875 | |
c88130bc | 876 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 877 | { |
1d45209d ED |
878 | struct net *net = nf_ct_net(ct); |
879 | ||
e53376be PNA |
880 | /* A freed object has refcnt == 0, that's |
881 | * the golden rule for SLAB_DESTROY_BY_RCU | |
882 | */ | |
883 | NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0); | |
884 | ||
ceeff754 | 885 | nf_ct_ext_destroy(ct); |
ea781f19 | 886 | nf_ct_ext_free(ct); |
5b3501fa | 887 | kmem_cache_free(net->ct.nf_conntrack_cachep, ct); |
4e857c58 | 888 | smp_mb__before_atomic(); |
0c3c6c00 | 889 | atomic_dec(&net->ct.count); |
76507f69 | 890 | } |
13b18339 | 891 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 | 892 | |
c539f017 | 893 | |
9fb9cbb1 YK |
894 | /* Allocate a new conntrack: we return -ENOMEM if classification |
895 | failed due to stress. Otherwise it really is unclassifiable. */ | |
896 | static struct nf_conntrack_tuple_hash * | |
b2a15a60 | 897 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 898 | const struct nf_conntrack_tuple *tuple, |
9fb9cbb1 | 899 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 900 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 901 | struct sk_buff *skb, |
60b5f8f7 | 902 | unsigned int dataoff, u32 hash) |
9fb9cbb1 | 903 | { |
c88130bc | 904 | struct nf_conn *ct; |
3c158f7f | 905 | struct nf_conn_help *help; |
9fb9cbb1 | 906 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 907 | struct nf_conntrack_ecache *ecache; |
ca7433df | 908 | struct nf_conntrack_expect *exp = NULL; |
5d0aa2cc | 909 | u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; |
60b5f8f7 PNA |
910 | struct nf_conn_timeout *timeout_ext; |
911 | unsigned int *timeouts; | |
9fb9cbb1 | 912 | |
605dcad6 | 913 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 914 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
915 | return NULL; |
916 | } | |
917 | ||
99f07e91 CG |
918 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
919 | hash); | |
0a9ee813 | 920 | if (IS_ERR(ct)) |
c88130bc | 921 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 | 922 | |
48b1de4c PM |
923 | if (tmpl && nfct_synproxy(tmpl)) { |
924 | nfct_seqadj_ext_add(ct); | |
925 | nfct_synproxy_ext_add(ct); | |
926 | } | |
927 | ||
60b5f8f7 PNA |
928 | timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; |
929 | if (timeout_ext) | |
930 | timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext); | |
931 | else | |
932 | timeouts = l4proto->get_timeouts(net); | |
933 | ||
2c8503f5 | 934 | if (!l4proto->new(ct, skb, dataoff, timeouts)) { |
c88130bc | 935 | nf_conntrack_free(ct); |
0d53778e | 936 | pr_debug("init conntrack: can't track with proto module\n"); |
9fb9cbb1 YK |
937 | return NULL; |
938 | } | |
939 | ||
60b5f8f7 PNA |
940 | if (timeout_ext) |
941 | nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC); | |
942 | ||
58401572 | 943 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a992ca2a | 944 | nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); |
c539f017 | 945 | nf_ct_labels_ext_add(ct); |
b2a15a60 PM |
946 | |
947 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
948 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
949 | ecache ? ecache->expmask : 0, | |
950 | GFP_ATOMIC); | |
58401572 | 951 | |
ca7433df JDB |
952 | local_bh_disable(); |
953 | if (net->ct.expect_count) { | |
954 | spin_lock(&nf_conntrack_expect_lock); | |
955 | exp = nf_ct_find_expectation(net, zone, tuple); | |
956 | if (exp) { | |
957 | pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", | |
958 | ct, exp); | |
959 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
960 | __set_bit(IPS_EXPECTED_BIT, &ct->status); | |
961 | /* exp->master safe, refcnt bumped in nf_ct_find_expectation */ | |
962 | ct->master = exp->master; | |
963 | if (exp->helper) { | |
964 | help = nf_ct_helper_ext_add(ct, exp->helper, | |
965 | GFP_ATOMIC); | |
966 | if (help) | |
967 | rcu_assign_pointer(help->helper, exp->helper); | |
968 | } | |
ceceae1b | 969 | |
9fb9cbb1 | 970 | #ifdef CONFIG_NF_CONNTRACK_MARK |
ca7433df | 971 | ct->mark = exp->master->mark; |
7c9728c3 JM |
972 | #endif |
973 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
ca7433df | 974 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 975 | #endif |
ca7433df JDB |
976 | NF_CT_STAT_INC(net, expect_new); |
977 | } | |
978 | spin_unlock(&nf_conntrack_expect_lock); | |
979 | } | |
980 | if (!exp) { | |
b2a15a60 | 981 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
0d55af87 | 982 | NF_CT_STAT_INC(net, new); |
22e7410b | 983 | } |
9fb9cbb1 | 984 | |
e53376be PNA |
985 | /* Now it is inserted into the unconfirmed list, bump refcount */ |
986 | nf_conntrack_get(&ct->ct_general); | |
b7779d06 | 987 | nf_ct_add_to_unconfirmed_list(ct); |
9fb9cbb1 | 988 | |
ca7433df | 989 | local_bh_enable(); |
9fb9cbb1 YK |
990 | |
991 | if (exp) { | |
992 | if (exp->expectfn) | |
c88130bc | 993 | exp->expectfn(ct, exp); |
6823645d | 994 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
995 | } |
996 | ||
c88130bc | 997 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
998 | } |
999 | ||
1000 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
1001 | static inline struct nf_conn * | |
b2a15a60 | 1002 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 1003 | struct sk_buff *skb, |
9fb9cbb1 YK |
1004 | unsigned int dataoff, |
1005 | u_int16_t l3num, | |
1006 | u_int8_t protonum, | |
1007 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 1008 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 1009 | int *set_reply, |
60b5f8f7 | 1010 | enum ip_conntrack_info *ctinfo) |
9fb9cbb1 YK |
1011 | { |
1012 | struct nf_conntrack_tuple tuple; | |
1013 | struct nf_conntrack_tuple_hash *h; | |
1014 | struct nf_conn *ct; | |
5d0aa2cc | 1015 | u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; |
99f07e91 | 1016 | u32 hash; |
9fb9cbb1 | 1017 | |
bbe735e4 | 1018 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
9fb9cbb1 | 1019 | dataoff, l3num, protonum, &tuple, l3proto, |
605dcad6 | 1020 | l4proto)) { |
0d53778e | 1021 | pr_debug("resolve_normal_ct: Can't get tuple\n"); |
9fb9cbb1 YK |
1022 | return NULL; |
1023 | } | |
1024 | ||
1025 | /* look for tuple match */ | |
99f07e91 CG |
1026 | hash = hash_conntrack_raw(&tuple, zone); |
1027 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); | |
9fb9cbb1 | 1028 | if (!h) { |
b2a15a60 | 1029 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
60b5f8f7 | 1030 | skb, dataoff, hash); |
9fb9cbb1 YK |
1031 | if (!h) |
1032 | return NULL; | |
1033 | if (IS_ERR(h)) | |
1034 | return (void *)h; | |
1035 | } | |
1036 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1037 | ||
1038 | /* It exists; we have (non-exclusive) reference. */ | |
1039 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
fb048833 | 1040 | *ctinfo = IP_CT_ESTABLISHED_REPLY; |
9fb9cbb1 YK |
1041 | /* Please set reply bit if this packet OK */ |
1042 | *set_reply = 1; | |
1043 | } else { | |
1044 | /* Once we've had two way comms, always ESTABLISHED. */ | |
1045 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
0d53778e | 1046 | pr_debug("nf_conntrack_in: normal packet for %p\n", ct); |
9fb9cbb1 YK |
1047 | *ctinfo = IP_CT_ESTABLISHED; |
1048 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
0d53778e PM |
1049 | pr_debug("nf_conntrack_in: related packet for %p\n", |
1050 | ct); | |
9fb9cbb1 YK |
1051 | *ctinfo = IP_CT_RELATED; |
1052 | } else { | |
0d53778e | 1053 | pr_debug("nf_conntrack_in: new packet for %p\n", ct); |
9fb9cbb1 YK |
1054 | *ctinfo = IP_CT_NEW; |
1055 | } | |
1056 | *set_reply = 0; | |
1057 | } | |
1058 | skb->nfct = &ct->ct_general; | |
1059 | skb->nfctinfo = *ctinfo; | |
1060 | return ct; | |
1061 | } | |
1062 | ||
1063 | unsigned int | |
a702a65f AD |
1064 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
1065 | struct sk_buff *skb) | |
9fb9cbb1 | 1066 | { |
b2a15a60 | 1067 | struct nf_conn *ct, *tmpl = NULL; |
9fb9cbb1 YK |
1068 | enum ip_conntrack_info ctinfo; |
1069 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 1070 | struct nf_conntrack_l4proto *l4proto; |
2c8503f5 | 1071 | unsigned int *timeouts; |
9fb9cbb1 YK |
1072 | unsigned int dataoff; |
1073 | u_int8_t protonum; | |
1074 | int set_reply = 0; | |
1075 | int ret; | |
1076 | ||
3db05fea | 1077 | if (skb->nfct) { |
b2a15a60 PM |
1078 | /* Previously seen (loopback or untracked)? Ignore. */ |
1079 | tmpl = (struct nf_conn *)skb->nfct; | |
1080 | if (!nf_ct_is_template(tmpl)) { | |
1081 | NF_CT_STAT_INC_ATOMIC(net, ignore); | |
1082 | return NF_ACCEPT; | |
1083 | } | |
1084 | skb->nfct = NULL; | |
9fb9cbb1 YK |
1085 | } |
1086 | ||
923f4902 | 1087 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 1088 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 1089 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
1090 | &dataoff, &protonum); |
1091 | if (ret <= 0) { | |
25985edc | 1092 | pr_debug("not prepared to track yet or error occurred\n"); |
0d55af87 AD |
1093 | NF_CT_STAT_INC_ATOMIC(net, error); |
1094 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1095 | ret = -ret; |
1096 | goto out; | |
9fb9cbb1 YK |
1097 | } |
1098 | ||
76108cea | 1099 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
1100 | |
1101 | /* It may be an special packet, error, unclean... | |
1102 | * inverse of the return code tells to the netfilter | |
1103 | * core what to do with the packet. */ | |
74c51a14 | 1104 | if (l4proto->error != NULL) { |
8fea97ec PM |
1105 | ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo, |
1106 | pf, hooknum); | |
74c51a14 | 1107 | if (ret <= 0) { |
0d55af87 AD |
1108 | NF_CT_STAT_INC_ATOMIC(net, error); |
1109 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1110 | ret = -ret; |
1111 | goto out; | |
74c51a14 | 1112 | } |
88ed01d1 PNA |
1113 | /* ICMP[v6] protocol trackers may assign one conntrack. */ |
1114 | if (skb->nfct) | |
1115 | goto out; | |
9fb9cbb1 YK |
1116 | } |
1117 | ||
b2a15a60 | 1118 | ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
60b5f8f7 | 1119 | l3proto, l4proto, &set_reply, &ctinfo); |
9fb9cbb1 YK |
1120 | if (!ct) { |
1121 | /* Not valid part of a connection */ | |
0d55af87 | 1122 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
b2a15a60 PM |
1123 | ret = NF_ACCEPT; |
1124 | goto out; | |
9fb9cbb1 YK |
1125 | } |
1126 | ||
1127 | if (IS_ERR(ct)) { | |
1128 | /* Too stressed to deal. */ | |
0d55af87 | 1129 | NF_CT_STAT_INC_ATOMIC(net, drop); |
b2a15a60 PM |
1130 | ret = NF_DROP; |
1131 | goto out; | |
9fb9cbb1 YK |
1132 | } |
1133 | ||
3db05fea | 1134 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 1135 | |
60b5f8f7 | 1136 | /* Decide what timeout policy we want to apply to this flow. */ |
84b5ee93 | 1137 | timeouts = nf_ct_timeout_lookup(net, ct, l4proto); |
60b5f8f7 | 1138 | |
2c8503f5 | 1139 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); |
ec8d5409 | 1140 | if (ret <= 0) { |
9fb9cbb1 YK |
1141 | /* Invalid: inverse of the return code tells |
1142 | * the netfilter core what to do */ | |
0d53778e | 1143 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
1144 | nf_conntrack_put(skb->nfct); |
1145 | skb->nfct = NULL; | |
0d55af87 | 1146 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
1147 | if (ret == -NF_DROP) |
1148 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
b2a15a60 PM |
1149 | ret = -ret; |
1150 | goto out; | |
9fb9cbb1 YK |
1151 | } |
1152 | ||
1153 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 1154 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 | 1155 | out: |
c3174286 PNA |
1156 | if (tmpl) { |
1157 | /* Special case: we have to repeat this hook, assign the | |
1158 | * template again to this packet. We assume that this packet | |
1159 | * has no conntrack assigned. This is used by nf_ct_tcp. */ | |
1160 | if (ret == NF_REPEAT) | |
1161 | skb->nfct = (struct nf_conntrack *)tmpl; | |
1162 | else | |
1163 | nf_ct_put(tmpl); | |
1164 | } | |
9fb9cbb1 YK |
1165 | |
1166 | return ret; | |
1167 | } | |
13b18339 | 1168 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 1169 | |
5f2b4c90 JE |
1170 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
1171 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 1172 | { |
5f2b4c90 | 1173 | bool ret; |
923f4902 PM |
1174 | |
1175 | rcu_read_lock(); | |
1176 | ret = nf_ct_invert_tuple(inverse, orig, | |
1177 | __nf_ct_l3proto_find(orig->src.l3num), | |
1178 | __nf_ct_l4proto_find(orig->src.l3num, | |
1179 | orig->dst.protonum)); | |
1180 | rcu_read_unlock(); | |
1181 | return ret; | |
9fb9cbb1 | 1182 | } |
13b18339 | 1183 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 1184 | |
5b1158e9 JK |
1185 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
1186 | implicitly racy: see __nf_conntrack_confirm */ | |
1187 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
1188 | const struct nf_conntrack_tuple *newreply) | |
1189 | { | |
1190 | struct nf_conn_help *help = nfct_help(ct); | |
1191 | ||
5b1158e9 JK |
1192 | /* Should be unconfirmed, so not in hash table yet */ |
1193 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
1194 | ||
0d53778e | 1195 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 1196 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
1197 | |
1198 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 1199 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 1200 | return; |
ceceae1b | 1201 | |
c52fbb41 | 1202 | rcu_read_lock(); |
b2a15a60 | 1203 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 1204 | rcu_read_unlock(); |
5b1158e9 | 1205 | } |
13b18339 | 1206 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 1207 | |
9fb9cbb1 YK |
1208 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
1209 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
1210 | enum ip_conntrack_info ctinfo, | |
1211 | const struct sk_buff *skb, | |
1212 | unsigned long extra_jiffies, | |
1213 | int do_acct) | |
1214 | { | |
9fb9cbb1 YK |
1215 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); |
1216 | NF_CT_ASSERT(skb); | |
1217 | ||
997ae831 | 1218 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1219 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1220 | goto acct; | |
997ae831 | 1221 | |
9fb9cbb1 YK |
1222 | /* If not in hash table, timer will not be active yet */ |
1223 | if (!nf_ct_is_confirmed(ct)) { | |
1224 | ct->timeout.expires = extra_jiffies; | |
9fb9cbb1 | 1225 | } else { |
be00c8e4 MJ |
1226 | unsigned long newtime = jiffies + extra_jiffies; |
1227 | ||
1228 | /* Only update the timeout if the new timeout is at least | |
1229 | HZ jiffies from the old timeout. Need del_timer for race | |
1230 | avoidance (may already be dying). */ | |
65cb9fda PM |
1231 | if (newtime - ct->timeout.expires >= HZ) |
1232 | mod_timer_pending(&ct->timeout, newtime); | |
9fb9cbb1 YK |
1233 | } |
1234 | ||
47d95045 | 1235 | acct: |
9fb9cbb1 | 1236 | if (do_acct) { |
f7b13e43 | 1237 | struct nf_conn_acct *acct; |
3ffd5eeb | 1238 | |
58401572 KPO |
1239 | acct = nf_conn_acct_find(ct); |
1240 | if (acct) { | |
f7b13e43 HE |
1241 | struct nf_conn_counter *counter = acct->counter; |
1242 | ||
1243 | atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); | |
1244 | atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes); | |
58401572 | 1245 | } |
9fb9cbb1 | 1246 | } |
9fb9cbb1 | 1247 | } |
13b18339 | 1248 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1249 | |
4c889498 DM |
1250 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
1251 | enum ip_conntrack_info ctinfo, | |
1252 | const struct sk_buff *skb, | |
1253 | int do_acct) | |
51091764 | 1254 | { |
718d4ad9 | 1255 | if (do_acct) { |
f7b13e43 | 1256 | struct nf_conn_acct *acct; |
58401572 | 1257 | |
58401572 KPO |
1258 | acct = nf_conn_acct_find(ct); |
1259 | if (acct) { | |
f7b13e43 HE |
1260 | struct nf_conn_counter *counter = acct->counter; |
1261 | ||
1262 | atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); | |
b3e0bfa7 | 1263 | atomic64_add(skb->len - skb_network_offset(skb), |
f7b13e43 | 1264 | &counter[CTINFO2DIR(ctinfo)].bytes); |
58401572 | 1265 | } |
718d4ad9 | 1266 | } |
58401572 | 1267 | |
4c889498 | 1268 | if (del_timer(&ct->timeout)) { |
51091764 | 1269 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
1270 | return true; |
1271 | } | |
1272 | return false; | |
51091764 | 1273 | } |
718d4ad9 | 1274 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 1275 | |
5d0aa2cc PM |
1276 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1277 | static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = { | |
1278 | .len = sizeof(struct nf_conntrack_zone), | |
1279 | .align = __alignof__(struct nf_conntrack_zone), | |
1280 | .id = NF_CT_EXT_ZONE, | |
1281 | }; | |
1282 | #endif | |
1283 | ||
c0cd1156 | 1284 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
c1d10adb PNA |
1285 | |
1286 | #include <linux/netfilter/nfnetlink.h> | |
1287 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1288 | #include <linux/mutex.h> |
1289 | ||
c1d10adb PNA |
1290 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1291 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1292 | * or depend on ctnetlink */ | |
fdf70832 | 1293 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1294 | const struct nf_conntrack_tuple *tuple) |
1295 | { | |
bae65be8 DM |
1296 | if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) || |
1297 | nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port)) | |
1298 | goto nla_put_failure; | |
c1d10adb PNA |
1299 | return 0; |
1300 | ||
df6fb868 | 1301 | nla_put_failure: |
c1d10adb PNA |
1302 | return -1; |
1303 | } | |
fdf70832 | 1304 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1305 | |
f73e924c PM |
1306 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1307 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1308 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1309 | }; |
f73e924c | 1310 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1311 | |
fdf70832 | 1312 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1313 | struct nf_conntrack_tuple *t) |
1314 | { | |
df6fb868 | 1315 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1316 | return -EINVAL; |
1317 | ||
77236b6e PM |
1318 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1319 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1320 | |
1321 | return 0; | |
1322 | } | |
fdf70832 | 1323 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1324 | |
1325 | int nf_ct_port_nlattr_tuple_size(void) | |
1326 | { | |
1327 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1328 | } | |
1329 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1330 | #endif |
1331 | ||
9fb9cbb1 | 1332 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
312a0c16 | 1333 | static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb) |
9fb9cbb1 YK |
1334 | { |
1335 | struct nf_conn *ct; | |
1336 | enum ip_conntrack_info ctinfo; | |
1337 | ||
1338 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1339 | ct = nf_ct_get(skb, &ctinfo); | |
1340 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
fb048833 | 1341 | ctinfo = IP_CT_RELATED_REPLY; |
9fb9cbb1 YK |
1342 | else |
1343 | ctinfo = IP_CT_RELATED; | |
1344 | ||
1345 | /* Attach to new skbuff, and increment count */ | |
1346 | nskb->nfct = &ct->ct_general; | |
1347 | nskb->nfctinfo = ctinfo; | |
1348 | nf_conntrack_get(nskb->nfct); | |
1349 | } | |
1350 | ||
9fb9cbb1 | 1351 | /* Bring out ya dead! */ |
df0933dc | 1352 | static struct nf_conn * |
400dad39 | 1353 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1354 | void *data, unsigned int *bucket) |
1355 | { | |
df0933dc PM |
1356 | struct nf_conntrack_tuple_hash *h; |
1357 | struct nf_conn *ct; | |
ea781f19 | 1358 | struct hlist_nulls_node *n; |
b7779d06 | 1359 | int cpu; |
93bb0ceb | 1360 | spinlock_t *lockp; |
9fb9cbb1 | 1361 | |
d696c7bd | 1362 | for (; *bucket < net->ct.htable_size; (*bucket)++) { |
93bb0ceb JDB |
1363 | lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS]; |
1364 | local_bh_disable(); | |
1365 | spin_lock(lockp); | |
1366 | if (*bucket < net->ct.htable_size) { | |
1367 | hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) { | |
1368 | if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL) | |
1369 | continue; | |
1370 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1371 | if (iter(ct, data)) | |
1372 | goto found; | |
1373 | } | |
df0933dc | 1374 | } |
93bb0ceb JDB |
1375 | spin_unlock(lockp); |
1376 | local_bh_enable(); | |
601e68e1 | 1377 | } |
b7779d06 JDB |
1378 | |
1379 | for_each_possible_cpu(cpu) { | |
1380 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1381 | ||
1382 | spin_lock_bh(&pcpu->lock); | |
1383 | hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) { | |
1384 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1385 | if (iter(ct, data)) | |
1386 | set_bit(IPS_DYING_BIT, &ct->status); | |
1387 | } | |
1388 | spin_unlock_bh(&pcpu->lock); | |
1389 | } | |
df0933dc PM |
1390 | return NULL; |
1391 | found: | |
c073e3fa | 1392 | atomic_inc(&ct->ct_general.use); |
93bb0ceb JDB |
1393 | spin_unlock(lockp); |
1394 | local_bh_enable(); | |
df0933dc | 1395 | return ct; |
9fb9cbb1 YK |
1396 | } |
1397 | ||
400dad39 AD |
1398 | void nf_ct_iterate_cleanup(struct net *net, |
1399 | int (*iter)(struct nf_conn *i, void *data), | |
c655bc68 | 1400 | void *data, u32 portid, int report) |
9fb9cbb1 | 1401 | { |
df0933dc | 1402 | struct nf_conn *ct; |
9fb9cbb1 YK |
1403 | unsigned int bucket = 0; |
1404 | ||
400dad39 | 1405 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
1406 | /* Time to push up daises... */ |
1407 | if (del_timer(&ct->timeout)) | |
c655bc68 | 1408 | nf_ct_delete(ct, portid, report); |
02982c27 | 1409 | |
9fb9cbb1 YK |
1410 | /* ... else the timer will get him soon. */ |
1411 | ||
1412 | nf_ct_put(ct); | |
1413 | } | |
1414 | } | |
13b18339 | 1415 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1416 | |
274d383b PNA |
1417 | static int kill_all(struct nf_conn *i, void *data) |
1418 | { | |
1419 | return 1; | |
1420 | } | |
1421 | ||
d862a662 | 1422 | void nf_ct_free_hashtable(void *hash, unsigned int size) |
9fb9cbb1 | 1423 | { |
d862a662 | 1424 | if (is_vmalloc_addr(hash)) |
9fb9cbb1 YK |
1425 | vfree(hash); |
1426 | else | |
601e68e1 | 1427 | free_pages((unsigned long)hash, |
f205c5e0 | 1428 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1429 | } |
ac565e5f | 1430 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1431 | |
ec464e5d | 1432 | void nf_conntrack_flush_report(struct net *net, u32 portid, int report) |
c1d10adb | 1433 | { |
c655bc68 | 1434 | nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report); |
c1d10adb | 1435 | } |
274d383b | 1436 | EXPORT_SYMBOL_GPL(nf_conntrack_flush_report); |
c1d10adb | 1437 | |
b3c5163f ED |
1438 | static int untrack_refs(void) |
1439 | { | |
1440 | int cnt = 0, cpu; | |
1441 | ||
1442 | for_each_possible_cpu(cpu) { | |
1443 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
1444 | ||
1445 | cnt += atomic_read(&ct->ct_general.use) - 1; | |
1446 | } | |
1447 | return cnt; | |
1448 | } | |
1449 | ||
f94161c1 | 1450 | void nf_conntrack_cleanup_start(void) |
9fb9cbb1 | 1451 | { |
f94161c1 G |
1452 | RCU_INIT_POINTER(ip_ct_attach, NULL); |
1453 | } | |
1454 | ||
1455 | void nf_conntrack_cleanup_end(void) | |
1456 | { | |
1457 | RCU_INIT_POINTER(nf_ct_destroy, NULL); | |
b3c5163f | 1458 | while (untrack_refs() > 0) |
9edd7ca0 PM |
1459 | schedule(); |
1460 | ||
5d0aa2cc PM |
1461 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1462 | nf_ct_extend_unregister(&nf_ct_zone_extend); | |
1463 | #endif | |
04d87001 | 1464 | nf_conntrack_proto_fini(); |
41d73ec0 | 1465 | nf_conntrack_seqadj_fini(); |
5f69b8f5 | 1466 | nf_conntrack_labels_fini(); |
5e615b22 | 1467 | nf_conntrack_helper_fini(); |
8684094c | 1468 | nf_conntrack_timeout_fini(); |
3fe0f943 | 1469 | nf_conntrack_ecache_fini(); |
73f4001a | 1470 | nf_conntrack_tstamp_fini(); |
b7ff3a1f | 1471 | nf_conntrack_acct_fini(); |
83b4dbe1 | 1472 | nf_conntrack_expect_fini(); |
08f6547d | 1473 | } |
9fb9cbb1 | 1474 | |
f94161c1 G |
1475 | /* |
1476 | * Mishearing the voices in his head, our hero wonders how he's | |
1477 | * supposed to kill the mall. | |
1478 | */ | |
1479 | void nf_conntrack_cleanup_net(struct net *net) | |
08f6547d | 1480 | { |
dece40e8 VD |
1481 | LIST_HEAD(single); |
1482 | ||
1483 | list_add(&net->exit_list, &single); | |
1484 | nf_conntrack_cleanup_net_list(&single); | |
1485 | } | |
1486 | ||
1487 | void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list) | |
1488 | { | |
1489 | int busy; | |
1490 | struct net *net; | |
1491 | ||
f94161c1 G |
1492 | /* |
1493 | * This makes sure all current packets have passed through | |
1494 | * netfilter framework. Roll on, two-stage module | |
1495 | * delete... | |
1496 | */ | |
1497 | synchronize_net(); | |
dece40e8 VD |
1498 | i_see_dead_people: |
1499 | busy = 0; | |
1500 | list_for_each_entry(net, net_exit_list, exit_list) { | |
c655bc68 | 1501 | nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0); |
dece40e8 VD |
1502 | if (atomic_read(&net->ct.count) != 0) |
1503 | busy = 1; | |
1504 | } | |
1505 | if (busy) { | |
9fb9cbb1 YK |
1506 | schedule(); |
1507 | goto i_see_dead_people; | |
1508 | } | |
1509 | ||
dece40e8 VD |
1510 | list_for_each_entry(net, net_exit_list, exit_list) { |
1511 | nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size); | |
1512 | nf_conntrack_proto_pernet_fini(net); | |
1513 | nf_conntrack_helper_pernet_fini(net); | |
1514 | nf_conntrack_ecache_pernet_fini(net); | |
1515 | nf_conntrack_tstamp_pernet_fini(net); | |
1516 | nf_conntrack_acct_pernet_fini(net); | |
1517 | nf_conntrack_expect_pernet_fini(net); | |
1518 | kmem_cache_destroy(net->ct.nf_conntrack_cachep); | |
1519 | kfree(net->ct.slabname); | |
1520 | free_percpu(net->ct.stat); | |
b7779d06 | 1521 | free_percpu(net->ct.pcpu_lists); |
dece40e8 | 1522 | } |
08f6547d AD |
1523 | } |
1524 | ||
d862a662 | 1525 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) |
9fb9cbb1 | 1526 | { |
ea781f19 ED |
1527 | struct hlist_nulls_head *hash; |
1528 | unsigned int nr_slots, i; | |
1529 | size_t sz; | |
9fb9cbb1 | 1530 | |
ea781f19 ED |
1531 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1532 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
1533 | sz = nr_slots * sizeof(struct hlist_nulls_head); | |
1534 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1535 | get_order(sz)); | |
601e68e1 | 1536 | if (!hash) { |
9fb9cbb1 | 1537 | printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); |
966567b7 | 1538 | hash = vzalloc(sz); |
9fb9cbb1 YK |
1539 | } |
1540 | ||
ea781f19 ED |
1541 | if (hash && nulls) |
1542 | for (i = 0; i < nr_slots; i++) | |
1543 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1544 | |
1545 | return hash; | |
1546 | } | |
ac565e5f | 1547 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1548 | |
fae718dd | 1549 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) |
9fb9cbb1 | 1550 | { |
4b5511eb | 1551 | int i, bucket, rc; |
96eb24d7 | 1552 | unsigned int hashsize, old_size; |
ea781f19 | 1553 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1554 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1555 | struct nf_conn *ct; |
9fb9cbb1 | 1556 | |
d696c7bd PM |
1557 | if (current->nsproxy->net_ns != &init_net) |
1558 | return -EOPNOTSUPP; | |
1559 | ||
9fb9cbb1 YK |
1560 | /* On boot, we can set this without any fancy locking. */ |
1561 | if (!nf_conntrack_htable_size) | |
1562 | return param_set_uint(val, kp); | |
1563 | ||
4b5511eb AP |
1564 | rc = kstrtouint(val, 0, &hashsize); |
1565 | if (rc) | |
1566 | return rc; | |
9fb9cbb1 YK |
1567 | if (!hashsize) |
1568 | return -EINVAL; | |
1569 | ||
d862a662 | 1570 | hash = nf_ct_alloc_hashtable(&hashsize, 1); |
9fb9cbb1 YK |
1571 | if (!hash) |
1572 | return -ENOMEM; | |
1573 | ||
93bb0ceb JDB |
1574 | local_bh_disable(); |
1575 | nf_conntrack_all_lock(); | |
1576 | write_seqcount_begin(&init_net.ct.generation); | |
1577 | ||
76507f69 PM |
1578 | /* Lookups in the old hash might happen in parallel, which means we |
1579 | * might get false negatives during connection lookup. New connections | |
1580 | * created because of a false negative won't make it into the hash | |
93bb0ceb | 1581 | * though since that required taking the locks. |
76507f69 | 1582 | */ |
93bb0ceb | 1583 | |
d696c7bd | 1584 | for (i = 0; i < init_net.ct.htable_size; i++) { |
ea781f19 ED |
1585 | while (!hlist_nulls_empty(&init_net.ct.hash[i])) { |
1586 | h = hlist_nulls_entry(init_net.ct.hash[i].first, | |
1587 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1588 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1589 | hlist_nulls_del_rcu(&h->hnnode); |
5d0aa2cc | 1590 | bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct), |
99f07e91 | 1591 | hashsize); |
ea781f19 | 1592 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1593 | } |
1594 | } | |
d696c7bd | 1595 | old_size = init_net.ct.htable_size; |
400dad39 | 1596 | old_hash = init_net.ct.hash; |
9fb9cbb1 | 1597 | |
d696c7bd | 1598 | init_net.ct.htable_size = nf_conntrack_htable_size = hashsize; |
400dad39 | 1599 | init_net.ct.hash = hash; |
93bb0ceb JDB |
1600 | |
1601 | write_seqcount_end(&init_net.ct.generation); | |
1602 | nf_conntrack_all_unlock(); | |
1603 | local_bh_enable(); | |
9fb9cbb1 | 1604 | |
d862a662 | 1605 | nf_ct_free_hashtable(old_hash, old_size); |
9fb9cbb1 YK |
1606 | return 0; |
1607 | } | |
fae718dd | 1608 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1609 | |
fae718dd | 1610 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1611 | &nf_conntrack_htable_size, 0600); |
1612 | ||
5bfddbd4 ED |
1613 | void nf_ct_untracked_status_or(unsigned long bits) |
1614 | { | |
b3c5163f ED |
1615 | int cpu; |
1616 | ||
1617 | for_each_possible_cpu(cpu) | |
1618 | per_cpu(nf_conntrack_untracked, cpu).status |= bits; | |
5bfddbd4 ED |
1619 | } |
1620 | EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); | |
1621 | ||
f94161c1 | 1622 | int nf_conntrack_init_start(void) |
9fb9cbb1 | 1623 | { |
f205c5e0 | 1624 | int max_factor = 8; |
93bb0ceb JDB |
1625 | int i, ret, cpu; |
1626 | ||
d5d20912 | 1627 | for (i = 0; i < CONNTRACK_LOCKS; i++) |
93bb0ceb | 1628 | spin_lock_init(&nf_conntrack_locks[i]); |
9fb9cbb1 YK |
1629 | |
1630 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB | |
f205c5e0 | 1631 | * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ |
9fb9cbb1 YK |
1632 | if (!nf_conntrack_htable_size) { |
1633 | nf_conntrack_htable_size | |
4481374c | 1634 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1635 | / sizeof(struct hlist_head)); |
4481374c | 1636 | if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
f205c5e0 PM |
1637 | nf_conntrack_htable_size = 16384; |
1638 | if (nf_conntrack_htable_size < 32) | |
1639 | nf_conntrack_htable_size = 32; | |
1640 | ||
1641 | /* Use a max. factor of four by default to get the same max as | |
1642 | * with the old struct list_heads. When a table size is given | |
1643 | * we use the old value of 8 to avoid reducing the max. | |
1644 | * entries. */ | |
1645 | max_factor = 4; | |
9fb9cbb1 | 1646 | } |
f205c5e0 | 1647 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 1648 | |
654d0fbd | 1649 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
1650 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
1651 | nf_conntrack_max); | |
83b4dbe1 G |
1652 | |
1653 | ret = nf_conntrack_expect_init(); | |
1654 | if (ret < 0) | |
1655 | goto err_expect; | |
1656 | ||
b7ff3a1f G |
1657 | ret = nf_conntrack_acct_init(); |
1658 | if (ret < 0) | |
1659 | goto err_acct; | |
1660 | ||
73f4001a G |
1661 | ret = nf_conntrack_tstamp_init(); |
1662 | if (ret < 0) | |
1663 | goto err_tstamp; | |
1664 | ||
3fe0f943 G |
1665 | ret = nf_conntrack_ecache_init(); |
1666 | if (ret < 0) | |
1667 | goto err_ecache; | |
1668 | ||
8684094c G |
1669 | ret = nf_conntrack_timeout_init(); |
1670 | if (ret < 0) | |
1671 | goto err_timeout; | |
1672 | ||
5e615b22 G |
1673 | ret = nf_conntrack_helper_init(); |
1674 | if (ret < 0) | |
1675 | goto err_helper; | |
1676 | ||
5f69b8f5 G |
1677 | ret = nf_conntrack_labels_init(); |
1678 | if (ret < 0) | |
1679 | goto err_labels; | |
1680 | ||
41d73ec0 PM |
1681 | ret = nf_conntrack_seqadj_init(); |
1682 | if (ret < 0) | |
1683 | goto err_seqadj; | |
1684 | ||
5d0aa2cc PM |
1685 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
1686 | ret = nf_ct_extend_register(&nf_ct_zone_extend); | |
1687 | if (ret < 0) | |
1688 | goto err_extend; | |
1689 | #endif | |
04d87001 G |
1690 | ret = nf_conntrack_proto_init(); |
1691 | if (ret < 0) | |
1692 | goto err_proto; | |
1693 | ||
9edd7ca0 | 1694 | /* Set up fake conntrack: to never be deleted, not in any hashes */ |
b3c5163f ED |
1695 | for_each_possible_cpu(cpu) { |
1696 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
b3c5163f ED |
1697 | write_pnet(&ct->ct_net, &init_net); |
1698 | atomic_set(&ct->ct_general.use, 1); | |
1699 | } | |
9edd7ca0 | 1700 | /* - and look it like as a confirmed connection */ |
5bfddbd4 | 1701 | nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); |
08f6547d AD |
1702 | return 0; |
1703 | ||
04d87001 | 1704 | err_proto: |
5d0aa2cc | 1705 | #ifdef CONFIG_NF_CONNTRACK_ZONES |
04d87001 | 1706 | nf_ct_extend_unregister(&nf_ct_zone_extend); |
5d0aa2cc | 1707 | err_extend: |
a9006892 | 1708 | #endif |
41d73ec0 PM |
1709 | nf_conntrack_seqadj_fini(); |
1710 | err_seqadj: | |
04d87001 | 1711 | nf_conntrack_labels_fini(); |
5f69b8f5 G |
1712 | err_labels: |
1713 | nf_conntrack_helper_fini(); | |
5e615b22 G |
1714 | err_helper: |
1715 | nf_conntrack_timeout_fini(); | |
8684094c G |
1716 | err_timeout: |
1717 | nf_conntrack_ecache_fini(); | |
3fe0f943 G |
1718 | err_ecache: |
1719 | nf_conntrack_tstamp_fini(); | |
73f4001a G |
1720 | err_tstamp: |
1721 | nf_conntrack_acct_fini(); | |
b7ff3a1f G |
1722 | err_acct: |
1723 | nf_conntrack_expect_fini(); | |
83b4dbe1 | 1724 | err_expect: |
08f6547d AD |
1725 | return ret; |
1726 | } | |
1727 | ||
f94161c1 G |
1728 | void nf_conntrack_init_end(void) |
1729 | { | |
1730 | /* For use by REJECT target */ | |
1731 | RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); | |
1732 | RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); | |
f94161c1 G |
1733 | } |
1734 | ||
8cc20198 ED |
1735 | /* |
1736 | * We need to use special "null" values, not used in hash table | |
1737 | */ | |
1738 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1739 | #define DYING_NULLS_VAL ((1<<30)+1) | |
252b3e8c | 1740 | #define TEMPLATE_NULLS_VAL ((1<<30)+2) |
8cc20198 | 1741 | |
f94161c1 | 1742 | int nf_conntrack_init_net(struct net *net) |
08f6547d | 1743 | { |
b7779d06 JDB |
1744 | int ret = -ENOMEM; |
1745 | int cpu; | |
ceceae1b | 1746 | |
08f6547d | 1747 | atomic_set(&net->ct.count, 0); |
ee214d54 | 1748 | seqcount_init(&net->ct.generation); |
b7779d06 JDB |
1749 | |
1750 | net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu); | |
1751 | if (!net->ct.pcpu_lists) | |
08f6547d | 1752 | goto err_stat; |
b7779d06 JDB |
1753 | |
1754 | for_each_possible_cpu(cpu) { | |
1755 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1756 | ||
1757 | spin_lock_init(&pcpu->lock); | |
1758 | INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL); | |
1759 | INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL); | |
1760 | INIT_HLIST_NULLS_HEAD(&pcpu->tmpl, TEMPLATE_NULLS_VAL); | |
08f6547d | 1761 | } |
5b3501fa | 1762 | |
b7779d06 JDB |
1763 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1764 | if (!net->ct.stat) | |
1765 | goto err_pcpu_lists; | |
1766 | ||
5b3501fa | 1767 | net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net); |
b7779d06 | 1768 | if (!net->ct.slabname) |
5b3501fa | 1769 | goto err_slabname; |
5b3501fa ED |
1770 | |
1771 | net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname, | |
1772 | sizeof(struct nf_conn), 0, | |
1773 | SLAB_DESTROY_BY_RCU, NULL); | |
1774 | if (!net->ct.nf_conntrack_cachep) { | |
1775 | printk(KERN_ERR "Unable to create nf_conn slab cache\n"); | |
5b3501fa ED |
1776 | goto err_cache; |
1777 | } | |
d696c7bd PM |
1778 | |
1779 | net->ct.htable_size = nf_conntrack_htable_size; | |
d862a662 | 1780 | net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1); |
08f6547d | 1781 | if (!net->ct.hash) { |
08f6547d AD |
1782 | printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); |
1783 | goto err_hash; | |
1784 | } | |
83b4dbe1 | 1785 | ret = nf_conntrack_expect_pernet_init(net); |
08f6547d AD |
1786 | if (ret < 0) |
1787 | goto err_expect; | |
b7ff3a1f | 1788 | ret = nf_conntrack_acct_pernet_init(net); |
58401572 | 1789 | if (ret < 0) |
08f6547d | 1790 | goto err_acct; |
73f4001a | 1791 | ret = nf_conntrack_tstamp_pernet_init(net); |
a992ca2a PNA |
1792 | if (ret < 0) |
1793 | goto err_tstamp; | |
3fe0f943 | 1794 | ret = nf_conntrack_ecache_pernet_init(net); |
a0891aa6 PNA |
1795 | if (ret < 0) |
1796 | goto err_ecache; | |
5e615b22 | 1797 | ret = nf_conntrack_helper_pernet_init(net); |
a9006892 EL |
1798 | if (ret < 0) |
1799 | goto err_helper; | |
04d87001 | 1800 | ret = nf_conntrack_proto_pernet_init(net); |
f94161c1 G |
1801 | if (ret < 0) |
1802 | goto err_proto; | |
08f6547d | 1803 | return 0; |
c539f017 | 1804 | |
f94161c1 | 1805 | err_proto: |
5e615b22 | 1806 | nf_conntrack_helper_pernet_fini(net); |
a9006892 | 1807 | err_helper: |
3fe0f943 | 1808 | nf_conntrack_ecache_pernet_fini(net); |
a0891aa6 | 1809 | err_ecache: |
73f4001a | 1810 | nf_conntrack_tstamp_pernet_fini(net); |
a992ca2a | 1811 | err_tstamp: |
b7ff3a1f | 1812 | nf_conntrack_acct_pernet_fini(net); |
08f6547d | 1813 | err_acct: |
83b4dbe1 | 1814 | nf_conntrack_expect_pernet_fini(net); |
08f6547d | 1815 | err_expect: |
d862a662 | 1816 | nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size); |
6058fa6b | 1817 | err_hash: |
5b3501fa ED |
1818 | kmem_cache_destroy(net->ct.nf_conntrack_cachep); |
1819 | err_cache: | |
1820 | kfree(net->ct.slabname); | |
1821 | err_slabname: | |
0d55af87 | 1822 | free_percpu(net->ct.stat); |
b7779d06 JDB |
1823 | err_pcpu_lists: |
1824 | free_percpu(net->ct.pcpu_lists); | |
0d55af87 | 1825 | err_stat: |
08f6547d AD |
1826 | return ret; |
1827 | } |