]>
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 | ||
ccd63c20 WJ |
15 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
16 | ||
9fb9cbb1 YK |
17 | #include <linux/types.h> |
18 | #include <linux/netfilter.h> | |
19 | #include <linux/module.h> | |
d43c36dc | 20 | #include <linux/sched.h> |
9fb9cbb1 YK |
21 | #include <linux/skbuff.h> |
22 | #include <linux/proc_fs.h> | |
23 | #include <linux/vmalloc.h> | |
24 | #include <linux/stddef.h> | |
25 | #include <linux/slab.h> | |
26 | #include <linux/random.h> | |
27 | #include <linux/jhash.h> | |
28 | #include <linux/err.h> | |
29 | #include <linux/percpu.h> | |
30 | #include <linux/moduleparam.h> | |
31 | #include <linux/notifier.h> | |
32 | #include <linux/kernel.h> | |
33 | #include <linux/netdevice.h> | |
34 | #include <linux/socket.h> | |
d7fe0f24 | 35 | #include <linux/mm.h> |
d696c7bd | 36 | #include <linux/nsproxy.h> |
ea781f19 | 37 | #include <linux/rculist_nulls.h> |
9fb9cbb1 | 38 | |
9fb9cbb1 YK |
39 | #include <net/netfilter/nf_conntrack.h> |
40 | #include <net/netfilter/nf_conntrack_l3proto.h> | |
605dcad6 | 41 | #include <net/netfilter/nf_conntrack_l4proto.h> |
77ab9cff | 42 | #include <net/netfilter/nf_conntrack_expect.h> |
9fb9cbb1 | 43 | #include <net/netfilter/nf_conntrack_helper.h> |
41d73ec0 | 44 | #include <net/netfilter/nf_conntrack_seqadj.h> |
9fb9cbb1 | 45 | #include <net/netfilter/nf_conntrack_core.h> |
ecfab2c9 | 46 | #include <net/netfilter/nf_conntrack_extend.h> |
58401572 | 47 | #include <net/netfilter/nf_conntrack_acct.h> |
a0891aa6 | 48 | #include <net/netfilter/nf_conntrack_ecache.h> |
5d0aa2cc | 49 | #include <net/netfilter/nf_conntrack_zones.h> |
a992ca2a | 50 | #include <net/netfilter/nf_conntrack_timestamp.h> |
dd705072 | 51 | #include <net/netfilter/nf_conntrack_timeout.h> |
c539f017 | 52 | #include <net/netfilter/nf_conntrack_labels.h> |
48b1de4c | 53 | #include <net/netfilter/nf_conntrack_synproxy.h> |
e6a7d3c0 | 54 | #include <net/netfilter/nf_nat.h> |
e17b666a | 55 | #include <net/netfilter/nf_nat_core.h> |
49376368 | 56 | #include <net/netfilter/nf_nat_helper.h> |
1b8c8a9f | 57 | #include <net/netns/hash.h> |
9fb9cbb1 | 58 | |
dc808fe2 | 59 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 60 | |
e17b666a PM |
61 | int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, |
62 | enum nf_nat_manip_type manip, | |
39938324 | 63 | const struct nlattr *attr) __read_mostly; |
e6a7d3c0 PNA |
64 | EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); |
65 | ||
93bb0ceb JDB |
66 | __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS]; |
67 | EXPORT_SYMBOL_GPL(nf_conntrack_locks); | |
9fb9cbb1 | 68 | |
ca7433df JDB |
69 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock); |
70 | EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock); | |
71 | ||
56d52d48 FW |
72 | struct hlist_nulls_head *nf_conntrack_hash __read_mostly; |
73 | EXPORT_SYMBOL_GPL(nf_conntrack_hash); | |
74 | ||
0c5366b3 | 75 | static __read_mostly struct kmem_cache *nf_conntrack_cachep; |
b16c2919 | 76 | static __read_mostly spinlock_t nf_conntrack_locks_all_lock; |
a3efd812 | 77 | static __read_mostly seqcount_t nf_conntrack_generation; |
70d72b7e | 78 | static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock); |
b16c2919 SL |
79 | static __read_mostly bool nf_conntrack_locks_all; |
80 | ||
81 | void nf_conntrack_lock(spinlock_t *lock) __acquires(lock) | |
82 | { | |
83 | spin_lock(lock); | |
84 | while (unlikely(nf_conntrack_locks_all)) { | |
85 | spin_unlock(lock); | |
e39365be | 86 | spin_unlock_wait(&nf_conntrack_locks_all_lock); |
b16c2919 SL |
87 | spin_lock(lock); |
88 | } | |
89 | } | |
90 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); | |
91 | ||
93bb0ceb JDB |
92 | static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2) |
93 | { | |
94 | h1 %= CONNTRACK_LOCKS; | |
95 | h2 %= CONNTRACK_LOCKS; | |
96 | spin_unlock(&nf_conntrack_locks[h1]); | |
97 | if (h1 != h2) | |
98 | spin_unlock(&nf_conntrack_locks[h2]); | |
99 | } | |
100 | ||
101 | /* return true if we need to recompute hashes (in case hash table was resized) */ | |
102 | static bool nf_conntrack_double_lock(struct net *net, unsigned int h1, | |
103 | unsigned int h2, unsigned int sequence) | |
104 | { | |
105 | h1 %= CONNTRACK_LOCKS; | |
106 | h2 %= CONNTRACK_LOCKS; | |
107 | if (h1 <= h2) { | |
b16c2919 | 108 | nf_conntrack_lock(&nf_conntrack_locks[h1]); |
93bb0ceb JDB |
109 | if (h1 != h2) |
110 | spin_lock_nested(&nf_conntrack_locks[h2], | |
111 | SINGLE_DEPTH_NESTING); | |
112 | } else { | |
b16c2919 | 113 | nf_conntrack_lock(&nf_conntrack_locks[h2]); |
93bb0ceb JDB |
114 | spin_lock_nested(&nf_conntrack_locks[h1], |
115 | SINGLE_DEPTH_NESTING); | |
116 | } | |
a3efd812 | 117 | if (read_seqcount_retry(&nf_conntrack_generation, sequence)) { |
93bb0ceb JDB |
118 | nf_conntrack_double_unlock(h1, h2); |
119 | return true; | |
120 | } | |
121 | return false; | |
122 | } | |
123 | ||
124 | static void nf_conntrack_all_lock(void) | |
125 | { | |
126 | int i; | |
127 | ||
b16c2919 SL |
128 | spin_lock(&nf_conntrack_locks_all_lock); |
129 | nf_conntrack_locks_all = true; | |
130 | ||
131 | for (i = 0; i < CONNTRACK_LOCKS; i++) { | |
e39365be | 132 | spin_unlock_wait(&nf_conntrack_locks[i]); |
b16c2919 | 133 | } |
93bb0ceb JDB |
134 | } |
135 | ||
136 | static void nf_conntrack_all_unlock(void) | |
137 | { | |
b16c2919 SL |
138 | nf_conntrack_locks_all = false; |
139 | spin_unlock(&nf_conntrack_locks_all_lock); | |
93bb0ceb JDB |
140 | } |
141 | ||
e2b7606c | 142 | unsigned int nf_conntrack_htable_size __read_mostly; |
13b18339 PM |
143 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
144 | ||
e478075c | 145 | unsigned int nf_conntrack_max __read_mostly; |
a999e683 | 146 | EXPORT_SYMBOL_GPL(nf_conntrack_max); |
13b18339 | 147 | |
b3c5163f ED |
148 | DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked); |
149 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked); | |
13b18339 | 150 | |
141658fb | 151 | static unsigned int nf_conntrack_hash_rnd __read_mostly; |
9fb9cbb1 | 152 | |
1b8c8a9f FW |
153 | static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, |
154 | const struct net *net) | |
9fb9cbb1 | 155 | { |
0794935e | 156 | unsigned int n; |
1b8c8a9f | 157 | u32 seed; |
0794935e | 158 | |
141658fb FW |
159 | get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd)); |
160 | ||
0794935e PM |
161 | /* The direction must be ignored, so we hash everything up to the |
162 | * destination ports (which is a multiple of 4) and treat the last | |
163 | * three bytes manually. | |
164 | */ | |
1b8c8a9f | 165 | seed = nf_conntrack_hash_rnd ^ net_hash_mix(net); |
0794935e | 166 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); |
1b8c8a9f | 167 | return jhash2((u32 *)tuple, n, seed ^ |
99f07e91 CG |
168 | (((__force __u16)tuple->dst.u.all << 16) | |
169 | tuple->dst.protonum)); | |
170 | } | |
171 | ||
56d52d48 | 172 | static u32 scale_hash(u32 hash) |
99f07e91 | 173 | { |
56d52d48 | 174 | return reciprocal_scale(hash, nf_conntrack_htable_size); |
99f07e91 | 175 | } |
0794935e | 176 | |
1b8c8a9f FW |
177 | static u32 __hash_conntrack(const struct net *net, |
178 | const struct nf_conntrack_tuple *tuple, | |
179 | unsigned int size) | |
99f07e91 | 180 | { |
1b8c8a9f | 181 | return reciprocal_scale(hash_conntrack_raw(tuple, net), size); |
9fb9cbb1 YK |
182 | } |
183 | ||
1b8c8a9f FW |
184 | static u32 hash_conntrack(const struct net *net, |
185 | const struct nf_conntrack_tuple *tuple) | |
9fb9cbb1 | 186 | { |
56d52d48 | 187 | return scale_hash(hash_conntrack_raw(tuple, net)); |
9fb9cbb1 YK |
188 | } |
189 | ||
5f2b4c90 | 190 | bool |
9fb9cbb1 YK |
191 | nf_ct_get_tuple(const struct sk_buff *skb, |
192 | unsigned int nhoff, | |
193 | unsigned int dataoff, | |
194 | u_int16_t l3num, | |
195 | u_int8_t protonum, | |
a31f1adc | 196 | struct net *net, |
9fb9cbb1 YK |
197 | struct nf_conntrack_tuple *tuple, |
198 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 199 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 200 | { |
443a70d5 | 201 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
202 | |
203 | tuple->src.l3num = l3num; | |
204 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 205 | return false; |
9fb9cbb1 YK |
206 | |
207 | tuple->dst.protonum = protonum; | |
208 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
209 | ||
a31f1adc | 210 | return l4proto->pkt_to_tuple(skb, dataoff, net, tuple); |
9fb9cbb1 | 211 | } |
13b18339 | 212 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 213 | |
5f2b4c90 | 214 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
a31f1adc EB |
215 | u_int16_t l3num, |
216 | struct net *net, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
217 | { |
218 | struct nf_conntrack_l3proto *l3proto; | |
219 | struct nf_conntrack_l4proto *l4proto; | |
220 | unsigned int protoff; | |
221 | u_int8_t protonum; | |
222 | int ret; | |
223 | ||
224 | rcu_read_lock(); | |
225 | ||
226 | l3proto = __nf_ct_l3proto_find(l3num); | |
227 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
228 | if (ret != NF_ACCEPT) { | |
229 | rcu_read_unlock(); | |
5f2b4c90 | 230 | return false; |
e2a3123f YK |
231 | } |
232 | ||
233 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
234 | ||
a31f1adc | 235 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple, |
e2a3123f YK |
236 | l3proto, l4proto); |
237 | ||
238 | rcu_read_unlock(); | |
239 | return ret; | |
240 | } | |
241 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
242 | ||
5f2b4c90 | 243 | bool |
9fb9cbb1 YK |
244 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
245 | const struct nf_conntrack_tuple *orig, | |
246 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 247 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 248 | { |
443a70d5 | 249 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
250 | |
251 | inverse->src.l3num = orig->src.l3num; | |
252 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 253 | return false; |
9fb9cbb1 YK |
254 | |
255 | inverse->dst.dir = !orig->dst.dir; | |
256 | ||
257 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 258 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 259 | } |
13b18339 | 260 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 261 | |
9fb9cbb1 YK |
262 | static void |
263 | clean_from_lists(struct nf_conn *ct) | |
264 | { | |
0d53778e | 265 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
266 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
267 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
268 | |
269 | /* Destroy all pending expectations */ | |
c1d10adb | 270 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
271 | } |
272 | ||
b7779d06 JDB |
273 | /* must be called with local_bh_disable */ |
274 | static void nf_ct_add_to_dying_list(struct nf_conn *ct) | |
275 | { | |
276 | struct ct_pcpu *pcpu; | |
277 | ||
278 | /* add this conntrack to the (per cpu) dying list */ | |
279 | ct->cpu = smp_processor_id(); | |
280 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
281 | ||
282 | spin_lock(&pcpu->lock); | |
283 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
284 | &pcpu->dying); | |
285 | spin_unlock(&pcpu->lock); | |
286 | } | |
287 | ||
288 | /* must be called with local_bh_disable */ | |
289 | static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct) | |
290 | { | |
291 | struct ct_pcpu *pcpu; | |
292 | ||
293 | /* add this conntrack to the (per cpu) unconfirmed list */ | |
294 | ct->cpu = smp_processor_id(); | |
295 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
296 | ||
297 | spin_lock(&pcpu->lock); | |
298 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
299 | &pcpu->unconfirmed); | |
300 | spin_unlock(&pcpu->lock); | |
301 | } | |
302 | ||
303 | /* must be called with local_bh_disable */ | |
304 | static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct) | |
305 | { | |
306 | struct ct_pcpu *pcpu; | |
307 | ||
308 | /* We overload first tuple to link into unconfirmed or dying list.*/ | |
309 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
310 | ||
311 | spin_lock(&pcpu->lock); | |
312 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); | |
313 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
314 | spin_unlock(&pcpu->lock); | |
315 | } | |
316 | ||
0838aa7f | 317 | /* Released via destroy_conntrack() */ |
308ac914 DB |
318 | struct nf_conn *nf_ct_tmpl_alloc(struct net *net, |
319 | const struct nf_conntrack_zone *zone, | |
320 | gfp_t flags) | |
0838aa7f PNA |
321 | { |
322 | struct nf_conn *tmpl; | |
323 | ||
f58e5aa7 | 324 | tmpl = kzalloc(sizeof(*tmpl), flags); |
0838aa7f PNA |
325 | if (tmpl == NULL) |
326 | return NULL; | |
327 | ||
328 | tmpl->status = IPS_TEMPLATE; | |
329 | write_pnet(&tmpl->ct_net, net); | |
6c8dee98 | 330 | nf_ct_zone_add(tmpl, zone); |
0838aa7f PNA |
331 | atomic_set(&tmpl->ct_general.use, 0); |
332 | ||
333 | return tmpl; | |
0838aa7f PNA |
334 | } |
335 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc); | |
336 | ||
9cf94eab | 337 | void nf_ct_tmpl_free(struct nf_conn *tmpl) |
0838aa7f PNA |
338 | { |
339 | nf_ct_ext_destroy(tmpl); | |
340 | nf_ct_ext_free(tmpl); | |
341 | kfree(tmpl); | |
342 | } | |
9cf94eab | 343 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_free); |
0838aa7f | 344 | |
9fb9cbb1 YK |
345 | static void |
346 | destroy_conntrack(struct nf_conntrack *nfct) | |
347 | { | |
348 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
0d55af87 | 349 | struct net *net = nf_ct_net(ct); |
605dcad6 | 350 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 351 | |
0d53778e | 352 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 YK |
353 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
354 | NF_CT_ASSERT(!timer_pending(&ct->timeout)); | |
355 | ||
0838aa7f PNA |
356 | if (unlikely(nf_ct_is_template(ct))) { |
357 | nf_ct_tmpl_free(ct); | |
358 | return; | |
359 | } | |
923f4902 | 360 | rcu_read_lock(); |
5e8fbe2a | 361 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
4b4ceb9d | 362 | if (l4proto->destroy) |
605dcad6 | 363 | l4proto->destroy(ct); |
9fb9cbb1 | 364 | |
982d9a9c | 365 | rcu_read_unlock(); |
9fb9cbb1 | 366 | |
ca7433df | 367 | local_bh_disable(); |
9fb9cbb1 YK |
368 | /* Expectations will have been removed in clean_from_lists, |
369 | * except TFTP can create an expectation on the first packet, | |
370 | * before connection is in the list, so we need to clean here, | |
ca7433df JDB |
371 | * too. |
372 | */ | |
c1d10adb | 373 | nf_ct_remove_expectations(ct); |
9fb9cbb1 | 374 | |
b7779d06 | 375 | nf_ct_del_from_dying_or_unconfirmed_list(ct); |
9fb9cbb1 | 376 | |
0d55af87 | 377 | NF_CT_STAT_INC(net, delete); |
ca7433df | 378 | local_bh_enable(); |
9fb9cbb1 YK |
379 | |
380 | if (ct->master) | |
381 | nf_ct_put(ct->master); | |
382 | ||
0d53778e | 383 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
384 | nf_conntrack_free(ct); |
385 | } | |
386 | ||
02982c27 | 387 | static void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 388 | { |
0d55af87 | 389 | struct net *net = nf_ct_net(ct); |
93bb0ceb | 390 | unsigned int hash, reply_hash; |
93bb0ceb | 391 | unsigned int sequence; |
9fb9cbb1 | 392 | |
9858a3ae | 393 | nf_ct_helper_destroy(ct); |
93bb0ceb JDB |
394 | |
395 | local_bh_disable(); | |
396 | do { | |
a3efd812 | 397 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 398 | hash = hash_conntrack(net, |
93bb0ceb | 399 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 400 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
401 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
402 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
403 | ||
9fb9cbb1 | 404 | clean_from_lists(ct); |
93bb0ceb JDB |
405 | nf_conntrack_double_unlock(hash, reply_hash); |
406 | ||
b7779d06 | 407 | nf_ct_add_to_dying_list(ct); |
93bb0ceb JDB |
408 | |
409 | NF_CT_STAT_INC(net, delete_list); | |
410 | local_bh_enable(); | |
dd7669a9 | 411 | } |
dd7669a9 | 412 | |
02982c27 | 413 | bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report) |
dd7669a9 | 414 | { |
a992ca2a PNA |
415 | struct nf_conn_tstamp *tstamp; |
416 | ||
417 | tstamp = nf_conn_tstamp_find(ct); | |
418 | if (tstamp && tstamp->stop == 0) | |
d2de875c | 419 | tstamp->stop = ktime_get_real_ns(); |
dd7669a9 | 420 | |
9500507c FW |
421 | if (nf_ct_is_dying(ct)) |
422 | goto delete; | |
423 | ||
424 | if (nf_conntrack_event_report(IPCT_DESTROY, ct, | |
425 | portid, report) < 0) { | |
dd7669a9 PNA |
426 | /* destroy event was not delivered */ |
427 | nf_ct_delete_from_lists(ct); | |
9500507c | 428 | nf_conntrack_ecache_delayed_work(nf_ct_net(ct)); |
02982c27 | 429 | return false; |
dd7669a9 | 430 | } |
9500507c FW |
431 | |
432 | nf_conntrack_ecache_work(nf_ct_net(ct)); | |
dd7669a9 | 433 | set_bit(IPS_DYING_BIT, &ct->status); |
9500507c | 434 | delete: |
dd7669a9 | 435 | nf_ct_delete_from_lists(ct); |
9fb9cbb1 | 436 | nf_ct_put(ct); |
02982c27 FW |
437 | return true; |
438 | } | |
439 | EXPORT_SYMBOL_GPL(nf_ct_delete); | |
440 | ||
441 | static void death_by_timeout(unsigned long ul_conntrack) | |
442 | { | |
443 | nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0); | |
9fb9cbb1 YK |
444 | } |
445 | ||
c6825c09 AV |
446 | static inline bool |
447 | nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, | |
308ac914 | 448 | const struct nf_conntrack_tuple *tuple, |
e0c7d472 FW |
449 | const struct nf_conntrack_zone *zone, |
450 | const struct net *net) | |
c6825c09 AV |
451 | { |
452 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
453 | ||
454 | /* A conntrack can be recreated with the equal tuple, | |
455 | * so we need to check that the conntrack is confirmed | |
456 | */ | |
457 | return nf_ct_tuple_equal(tuple, &h->tuple) && | |
deedb590 | 458 | nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) && |
e0c7d472 FW |
459 | nf_ct_is_confirmed(ct) && |
460 | net_eq(net, nf_ct_net(ct)); | |
c6825c09 AV |
461 | } |
462 | ||
64b87639 LZ |
463 | /* must be called with rcu read lock held */ |
464 | void nf_conntrack_get_ht(struct hlist_nulls_head **hash, unsigned int *hsize) | |
465 | { | |
466 | struct hlist_nulls_head *hptr; | |
467 | unsigned int sequence, hsz; | |
468 | ||
469 | do { | |
470 | sequence = read_seqcount_begin(&nf_conntrack_generation); | |
471 | hsz = nf_conntrack_htable_size; | |
472 | hptr = nf_conntrack_hash; | |
473 | } while (read_seqcount_retry(&nf_conntrack_generation, sequence)); | |
474 | ||
475 | *hash = hptr; | |
476 | *hsize = hsz; | |
477 | } | |
478 | EXPORT_SYMBOL_GPL(nf_conntrack_get_ht); | |
479 | ||
ea781f19 ED |
480 | /* |
481 | * Warning : | |
482 | * - Caller must take a reference on returned object | |
483 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
ea781f19 | 484 | */ |
99f07e91 | 485 | static struct nf_conntrack_tuple_hash * |
308ac914 | 486 | ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 487 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
488 | { |
489 | struct nf_conntrack_tuple_hash *h; | |
5e3c61f9 | 490 | struct hlist_nulls_head *ct_hash; |
ea781f19 | 491 | struct hlist_nulls_node *n; |
5e3c61f9 | 492 | unsigned int bucket, sequence; |
9fb9cbb1 | 493 | |
ea781f19 | 494 | begin: |
5e3c61f9 FW |
495 | do { |
496 | sequence = read_seqcount_begin(&nf_conntrack_generation); | |
56d52d48 FW |
497 | bucket = scale_hash(hash); |
498 | ct_hash = nf_conntrack_hash; | |
5e3c61f9 FW |
499 | } while (read_seqcount_retry(&nf_conntrack_generation, sequence)); |
500 | ||
501 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) { | |
e0c7d472 | 502 | if (nf_ct_key_equal(h, tuple, zone, net)) { |
2cf12348 | 503 | NF_CT_STAT_INC_ATOMIC(net, found); |
9fb9cbb1 YK |
504 | return h; |
505 | } | |
2cf12348 | 506 | NF_CT_STAT_INC_ATOMIC(net, searched); |
9fb9cbb1 | 507 | } |
ea781f19 ED |
508 | /* |
509 | * if the nulls value we got at the end of this lookup is | |
510 | * not the expected one, we must restart lookup. | |
511 | * We probably met an item that was moved to another chain. | |
512 | */ | |
99f07e91 | 513 | if (get_nulls_value(n) != bucket) { |
2cf12348 | 514 | NF_CT_STAT_INC_ATOMIC(net, search_restart); |
ea781f19 | 515 | goto begin; |
af740b2c | 516 | } |
9fb9cbb1 YK |
517 | |
518 | return NULL; | |
519 | } | |
99f07e91 | 520 | |
9fb9cbb1 | 521 | /* Find a connection corresponding to a tuple. */ |
99f07e91 | 522 | static struct nf_conntrack_tuple_hash * |
308ac914 | 523 | __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 524 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
525 | { |
526 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 527 | struct nf_conn *ct; |
9fb9cbb1 | 528 | |
76507f69 | 529 | rcu_read_lock(); |
ea781f19 | 530 | begin: |
99f07e91 | 531 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
532 | if (h) { |
533 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
534 | if (unlikely(nf_ct_is_dying(ct) || |
535 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 536 | h = NULL; |
ea781f19 | 537 | else { |
e0c7d472 | 538 | if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) { |
ea781f19 ED |
539 | nf_ct_put(ct); |
540 | goto begin; | |
541 | } | |
542 | } | |
76507f69 PM |
543 | } |
544 | rcu_read_unlock(); | |
9fb9cbb1 YK |
545 | |
546 | return h; | |
547 | } | |
99f07e91 CG |
548 | |
549 | struct nf_conntrack_tuple_hash * | |
308ac914 | 550 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 CG |
551 | const struct nf_conntrack_tuple *tuple) |
552 | { | |
553 | return __nf_conntrack_find_get(net, zone, tuple, | |
1b8c8a9f | 554 | hash_conntrack_raw(tuple, net)); |
99f07e91 | 555 | } |
13b18339 | 556 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 557 | |
c1d10adb PNA |
558 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
559 | unsigned int hash, | |
b476b72a | 560 | unsigned int reply_hash) |
c1d10adb | 561 | { |
ea781f19 | 562 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
56d52d48 | 563 | &nf_conntrack_hash[hash]); |
ea781f19 | 564 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
56d52d48 | 565 | &nf_conntrack_hash[reply_hash]); |
c1d10adb PNA |
566 | } |
567 | ||
7d367e06 JK |
568 | int |
569 | nf_conntrack_hash_check_insert(struct nf_conn *ct) | |
c1d10adb | 570 | { |
308ac914 | 571 | const struct nf_conntrack_zone *zone; |
d696c7bd | 572 | struct net *net = nf_ct_net(ct); |
b476b72a | 573 | unsigned int hash, reply_hash; |
7d367e06 JK |
574 | struct nf_conntrack_tuple_hash *h; |
575 | struct hlist_nulls_node *n; | |
93bb0ceb | 576 | unsigned int sequence; |
c1d10adb | 577 | |
5d0aa2cc | 578 | zone = nf_ct_zone(ct); |
7d367e06 | 579 | |
93bb0ceb JDB |
580 | local_bh_disable(); |
581 | do { | |
a3efd812 | 582 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 583 | hash = hash_conntrack(net, |
93bb0ceb | 584 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 585 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
586 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
587 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
7d367e06 JK |
588 | |
589 | /* See if there's one in the list already, including reverse */ | |
56d52d48 | 590 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 591 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 592 | zone, net)) |
7d367e06 | 593 | goto out; |
86804348 | 594 | |
56d52d48 | 595 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 596 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 597 | zone, net)) |
7d367e06 | 598 | goto out; |
c1d10adb | 599 | |
7d367e06 | 600 | add_timer(&ct->timeout); |
e53376be PNA |
601 | smp_wmb(); |
602 | /* The caller holds a reference to this object */ | |
603 | atomic_set(&ct->ct_general.use, 2); | |
b476b72a | 604 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 605 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 606 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 607 | local_bh_enable(); |
7d367e06 JK |
608 | return 0; |
609 | ||
610 | out: | |
93bb0ceb | 611 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 612 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 613 | local_bh_enable(); |
7d367e06 | 614 | return -EEXIST; |
c1d10adb | 615 | } |
7d367e06 | 616 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); |
c1d10adb | 617 | |
ba76738c PNA |
618 | static inline void nf_ct_acct_update(struct nf_conn *ct, |
619 | enum ip_conntrack_info ctinfo, | |
620 | unsigned int len) | |
621 | { | |
622 | struct nf_conn_acct *acct; | |
623 | ||
624 | acct = nf_conn_acct_find(ct); | |
625 | if (acct) { | |
626 | struct nf_conn_counter *counter = acct->counter; | |
627 | ||
628 | atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); | |
629 | atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes); | |
630 | } | |
631 | } | |
632 | ||
71d8c47f PNA |
633 | static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo, |
634 | const struct nf_conn *loser_ct) | |
635 | { | |
636 | struct nf_conn_acct *acct; | |
637 | ||
638 | acct = nf_conn_acct_find(loser_ct); | |
639 | if (acct) { | |
640 | struct nf_conn_counter *counter = acct->counter; | |
71d8c47f PNA |
641 | unsigned int bytes; |
642 | ||
643 | /* u32 should be fine since we must have seen one packet. */ | |
644 | bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes); | |
645 | nf_ct_acct_update(ct, ctinfo, bytes); | |
646 | } | |
647 | } | |
648 | ||
649 | /* Resolve race on insertion if this protocol allows this. */ | |
650 | static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb, | |
651 | enum ip_conntrack_info ctinfo, | |
652 | struct nf_conntrack_tuple_hash *h) | |
653 | { | |
654 | /* This is the conntrack entry already in hashes that won race. */ | |
655 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
656 | struct nf_conntrack_l4proto *l4proto; | |
657 | ||
658 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); | |
659 | if (l4proto->allow_clash && | |
660 | !nf_ct_is_dying(ct) && | |
661 | atomic_inc_not_zero(&ct->ct_general.use)) { | |
662 | nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct); | |
663 | nf_conntrack_put(skb->nfct); | |
664 | /* Assign conntrack already in hashes to this skbuff. Don't | |
665 | * modify skb->nfctinfo to ensure consistent stateful filtering. | |
666 | */ | |
667 | skb->nfct = &ct->ct_general; | |
668 | return NF_ACCEPT; | |
669 | } | |
670 | NF_CT_STAT_INC(net, drop); | |
671 | return NF_DROP; | |
672 | } | |
673 | ||
9fb9cbb1 YK |
674 | /* Confirm a connection given skb; places it in hash table */ |
675 | int | |
3db05fea | 676 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 | 677 | { |
308ac914 | 678 | const struct nf_conntrack_zone *zone; |
b476b72a | 679 | unsigned int hash, reply_hash; |
df0933dc | 680 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 681 | struct nf_conn *ct; |
df0933dc | 682 | struct nf_conn_help *help; |
a992ca2a | 683 | struct nf_conn_tstamp *tstamp; |
ea781f19 | 684 | struct hlist_nulls_node *n; |
9fb9cbb1 | 685 | enum ip_conntrack_info ctinfo; |
400dad39 | 686 | struct net *net; |
93bb0ceb | 687 | unsigned int sequence; |
71d8c47f | 688 | int ret = NF_DROP; |
9fb9cbb1 | 689 | |
3db05fea | 690 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 691 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
692 | |
693 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
694 | ICMP/TCP RST packets in other direction. Actual packet | |
695 | which created connection will be IP_CT_NEW or for an | |
696 | expected connection, IP_CT_RELATED. */ | |
697 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
698 | return NF_ACCEPT; | |
699 | ||
5d0aa2cc | 700 | zone = nf_ct_zone(ct); |
93bb0ceb JDB |
701 | local_bh_disable(); |
702 | ||
703 | do { | |
a3efd812 | 704 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb JDB |
705 | /* reuse the hash saved before */ |
706 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
56d52d48 | 707 | hash = scale_hash(hash); |
deedb590 | 708 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
709 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
710 | ||
711 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
9fb9cbb1 YK |
712 | |
713 | /* We're not in hash table, and we refuse to set up related | |
93bb0ceb JDB |
714 | * connections for unconfirmed conns. But packet copies and |
715 | * REJECT will give spurious warnings here. | |
716 | */ | |
9fb9cbb1 YK |
717 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ |
718 | ||
25985edc | 719 | /* No external references means no one else could have |
93bb0ceb JDB |
720 | * confirmed us. |
721 | */ | |
9fb9cbb1 | 722 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); |
0d53778e | 723 | pr_debug("Confirming conntrack %p\n", ct); |
8ca3f5e9 PNA |
724 | /* We have to check the DYING flag after unlink to prevent |
725 | * a race against nf_ct_get_next_corpse() possibly called from | |
726 | * user context, else we insert an already 'dead' hash, blocking | |
727 | * further use of that particular connection -JM. | |
728 | */ | |
729 | nf_ct_del_from_dying_or_unconfirmed_list(ct); | |
730 | ||
71d8c47f PNA |
731 | if (unlikely(nf_ct_is_dying(ct))) { |
732 | nf_ct_add_to_dying_list(ct); | |
733 | goto dying; | |
734 | } | |
fc350777 | 735 | |
9fb9cbb1 YK |
736 | /* See if there's one in the list already, including reverse: |
737 | NAT could have grabbed it without realizing, since we're | |
738 | not in the hash. If there is, we lost race. */ | |
56d52d48 | 739 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 740 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 741 | zone, net)) |
df0933dc | 742 | goto out; |
86804348 | 743 | |
56d52d48 | 744 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 745 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 746 | zone, net)) |
df0933dc | 747 | goto out; |
9fb9cbb1 | 748 | |
df0933dc PM |
749 | /* Timer relative to confirmation time, not original |
750 | setting time, otherwise we'd get timer wrap in | |
751 | weird delay cases. */ | |
752 | ct->timeout.expires += jiffies; | |
753 | add_timer(&ct->timeout); | |
754 | atomic_inc(&ct->ct_general.use); | |
45eec341 | 755 | ct->status |= IPS_CONFIRMED; |
5c8ec910 | 756 | |
a992ca2a PNA |
757 | /* set conntrack timestamp, if enabled. */ |
758 | tstamp = nf_conn_tstamp_find(ct); | |
759 | if (tstamp) { | |
760 | if (skb->tstamp.tv64 == 0) | |
e3192690 | 761 | __net_timestamp(skb); |
a992ca2a PNA |
762 | |
763 | tstamp->start = ktime_to_ns(skb->tstamp); | |
764 | } | |
5c8ec910 PM |
765 | /* Since the lookup is lockless, hash insertion must be done after |
766 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
767 | * guarantee that no other CPU can find the conntrack before the above | |
768 | * stores are visible. | |
769 | */ | |
b476b72a | 770 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 771 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 772 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 773 | local_bh_enable(); |
5c8ec910 | 774 | |
df0933dc PM |
775 | help = nfct_help(ct); |
776 | if (help && help->helper) | |
a71996fc | 777 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 778 | |
df0933dc | 779 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 780 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 781 | return NF_ACCEPT; |
9fb9cbb1 | 782 | |
df0933dc | 783 | out: |
8ca3f5e9 | 784 | nf_ct_add_to_dying_list(ct); |
71d8c47f PNA |
785 | ret = nf_ct_resolve_clash(net, skb, ctinfo, h); |
786 | dying: | |
93bb0ceb | 787 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 788 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 789 | local_bh_enable(); |
71d8c47f | 790 | return ret; |
9fb9cbb1 | 791 | } |
13b18339 | 792 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
793 | |
794 | /* Returns true if a connection correspondings to the tuple (required | |
795 | for NAT). */ | |
796 | int | |
797 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
798 | const struct nf_conn *ignored_conntrack) | |
799 | { | |
400dad39 | 800 | struct net *net = nf_ct_net(ignored_conntrack); |
308ac914 | 801 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 | 802 | struct nf_conntrack_tuple_hash *h; |
5e3c61f9 FW |
803 | struct hlist_nulls_head *ct_hash; |
804 | unsigned int hash, sequence; | |
ea781f19 | 805 | struct hlist_nulls_node *n; |
5d0aa2cc | 806 | struct nf_conn *ct; |
308ac914 DB |
807 | |
808 | zone = nf_ct_zone(ignored_conntrack); | |
9fb9cbb1 | 809 | |
2cf12348 | 810 | rcu_read_lock(); |
5e3c61f9 FW |
811 | do { |
812 | sequence = read_seqcount_begin(&nf_conntrack_generation); | |
813 | hash = hash_conntrack(net, tuple); | |
56d52d48 | 814 | ct_hash = nf_conntrack_hash; |
5e3c61f9 FW |
815 | } while (read_seqcount_retry(&nf_conntrack_generation, sequence)); |
816 | ||
817 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) { | |
5d0aa2cc PM |
818 | ct = nf_ct_tuplehash_to_ctrack(h); |
819 | if (ct != ignored_conntrack && | |
e0c7d472 | 820 | nf_ct_key_equal(h, tuple, zone, net)) { |
2cf12348 FW |
821 | NF_CT_STAT_INC_ATOMIC(net, found); |
822 | rcu_read_unlock(); | |
ba419aff PM |
823 | return 1; |
824 | } | |
2cf12348 | 825 | NF_CT_STAT_INC_ATOMIC(net, searched); |
ba419aff | 826 | } |
2cf12348 | 827 | rcu_read_unlock(); |
9fb9cbb1 | 828 | |
ba419aff | 829 | return 0; |
9fb9cbb1 | 830 | } |
13b18339 | 831 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 832 | |
7ae7730f PM |
833 | #define NF_CT_EVICTION_RANGE 8 |
834 | ||
9fb9cbb1 YK |
835 | /* There's a small race here where we may free a just-assured |
836 | connection. Too bad: we're in trouble anyway. */ | |
242922a0 FW |
837 | static unsigned int early_drop_list(struct net *net, |
838 | struct hlist_nulls_head *head) | |
9fb9cbb1 | 839 | { |
9fb9cbb1 | 840 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 841 | struct hlist_nulls_node *n; |
242922a0 FW |
842 | unsigned int drops = 0; |
843 | struct nf_conn *tmp; | |
3e86638e | 844 | |
242922a0 FW |
845 | hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) { |
846 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
9fb9cbb1 | 847 | |
242922a0 FW |
848 | if (test_bit(IPS_ASSURED_BIT, &tmp->status) || |
849 | !net_eq(nf_ct_net(tmp), net) || | |
850 | nf_ct_is_dying(tmp)) | |
851 | continue; | |
852 | ||
853 | if (!atomic_inc_not_zero(&tmp->ct_general.use)) | |
854 | continue; | |
76507f69 | 855 | |
242922a0 FW |
856 | /* kill only if still in same netns -- might have moved due to |
857 | * SLAB_DESTROY_BY_RCU rules. | |
858 | * | |
859 | * We steal the timer reference. If that fails timer has | |
860 | * already fired or someone else deleted it. Just drop ref | |
861 | * and move to next entry. | |
862 | */ | |
863 | if (net_eq(nf_ct_net(tmp), net) && | |
864 | nf_ct_is_confirmed(tmp) && | |
865 | del_timer(&tmp->timeout) && | |
866 | nf_ct_delete(tmp, 0, 0)) | |
867 | drops++; | |
868 | ||
869 | nf_ct_put(tmp); | |
9fb9cbb1 | 870 | } |
3e86638e | 871 | |
242922a0 FW |
872 | return drops; |
873 | } | |
9fb9cbb1 | 874 | |
242922a0 FW |
875 | static noinline int early_drop(struct net *net, unsigned int _hash) |
876 | { | |
877 | unsigned int i; | |
9fb9cbb1 | 878 | |
242922a0 FW |
879 | for (i = 0; i < NF_CT_EVICTION_RANGE; i++) { |
880 | struct hlist_nulls_head *ct_hash; | |
881 | unsigned hash, sequence, drops; | |
882 | ||
883 | do { | |
884 | sequence = read_seqcount_begin(&nf_conntrack_generation); | |
885 | hash = scale_hash(_hash++); | |
886 | ct_hash = nf_conntrack_hash; | |
887 | } while (read_seqcount_retry(&nf_conntrack_generation, sequence)); | |
888 | ||
889 | drops = early_drop_list(net, &ct_hash[hash]); | |
890 | if (drops) { | |
891 | NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops); | |
892 | return true; | |
74138511 | 893 | } |
9fb9cbb1 | 894 | } |
3e86638e | 895 | |
242922a0 | 896 | return false; |
9fb9cbb1 YK |
897 | } |
898 | ||
99f07e91 | 899 | static struct nf_conn * |
308ac914 DB |
900 | __nf_conntrack_alloc(struct net *net, |
901 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
902 | const struct nf_conntrack_tuple *orig, |
903 | const struct nf_conntrack_tuple *repl, | |
904 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 905 | { |
cd7fcbf1 | 906 | struct nf_conn *ct; |
9fb9cbb1 | 907 | |
5251e2d2 | 908 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 909 | atomic_inc(&net->ct.count); |
5251e2d2 | 910 | |
76eb9460 | 911 | if (nf_conntrack_max && |
49ac8713 | 912 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
93bb0ceb | 913 | if (!early_drop(net, hash)) { |
49ac8713 | 914 | atomic_dec(&net->ct.count); |
e87cc472 | 915 | net_warn_ratelimited("nf_conntrack: table full, dropping packet\n"); |
9fb9cbb1 YK |
916 | return ERR_PTR(-ENOMEM); |
917 | } | |
918 | } | |
919 | ||
941297f4 ED |
920 | /* |
921 | * Do not use kmem_cache_zalloc(), as this cache uses | |
922 | * SLAB_DESTROY_BY_RCU. | |
923 | */ | |
0c5366b3 | 924 | ct = kmem_cache_alloc(nf_conntrack_cachep, gfp); |
5e8018fc DB |
925 | if (ct == NULL) |
926 | goto out; | |
927 | ||
440f0d58 | 928 | spin_lock_init(&ct->lock); |
c88130bc | 929 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 930 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 931 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
932 | /* save hash for reusing when confirming */ |
933 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
c41884ce | 934 | ct->status = 0; |
9fb9cbb1 | 935 | /* Don't set timer yet: wait for confirmation */ |
c88130bc | 936 | setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); |
c2d9ba9b | 937 | write_pnet(&ct->ct_net, net); |
c41884ce FW |
938 | memset(&ct->__nfct_init_offset[0], 0, |
939 | offsetof(struct nf_conn, proto) - | |
940 | offsetof(struct nf_conn, __nfct_init_offset[0])); | |
5e8018fc | 941 | |
6c8dee98 | 942 | nf_ct_zone_add(ct, zone); |
5e8018fc | 943 | |
e53376be PNA |
944 | /* Because we use RCU lookups, we set ct_general.use to zero before |
945 | * this is inserted in any list. | |
941297f4 | 946 | */ |
e53376be | 947 | atomic_set(&ct->ct_general.use, 0); |
c88130bc | 948 | return ct; |
5e8018fc DB |
949 | out: |
950 | atomic_dec(&net->ct.count); | |
5d0aa2cc | 951 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 952 | } |
99f07e91 | 953 | |
308ac914 DB |
954 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
955 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
956 | const struct nf_conntrack_tuple *orig, |
957 | const struct nf_conntrack_tuple *repl, | |
958 | gfp_t gfp) | |
959 | { | |
960 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
961 | } | |
13b18339 | 962 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 963 | |
c88130bc | 964 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 965 | { |
1d45209d ED |
966 | struct net *net = nf_ct_net(ct); |
967 | ||
e53376be PNA |
968 | /* A freed object has refcnt == 0, that's |
969 | * the golden rule for SLAB_DESTROY_BY_RCU | |
970 | */ | |
971 | NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0); | |
972 | ||
ceeff754 | 973 | nf_ct_ext_destroy(ct); |
ea781f19 | 974 | nf_ct_ext_free(ct); |
0c5366b3 | 975 | kmem_cache_free(nf_conntrack_cachep, ct); |
4e857c58 | 976 | smp_mb__before_atomic(); |
0c3c6c00 | 977 | atomic_dec(&net->ct.count); |
76507f69 | 978 | } |
13b18339 | 979 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 | 980 | |
c539f017 | 981 | |
9fb9cbb1 YK |
982 | /* Allocate a new conntrack: we return -ENOMEM if classification |
983 | failed due to stress. Otherwise it really is unclassifiable. */ | |
984 | static struct nf_conntrack_tuple_hash * | |
b2a15a60 | 985 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 986 | const struct nf_conntrack_tuple *tuple, |
9fb9cbb1 | 987 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 988 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 989 | struct sk_buff *skb, |
60b5f8f7 | 990 | unsigned int dataoff, u32 hash) |
9fb9cbb1 | 991 | { |
c88130bc | 992 | struct nf_conn *ct; |
3c158f7f | 993 | struct nf_conn_help *help; |
9fb9cbb1 | 994 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 995 | struct nf_conntrack_ecache *ecache; |
ca7433df | 996 | struct nf_conntrack_expect *exp = NULL; |
308ac914 | 997 | const struct nf_conntrack_zone *zone; |
60b5f8f7 | 998 | struct nf_conn_timeout *timeout_ext; |
5e8018fc | 999 | struct nf_conntrack_zone tmp; |
60b5f8f7 | 1000 | unsigned int *timeouts; |
9fb9cbb1 | 1001 | |
605dcad6 | 1002 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 1003 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
1004 | return NULL; |
1005 | } | |
1006 | ||
5e8018fc | 1007 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
99f07e91 CG |
1008 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
1009 | hash); | |
0a9ee813 | 1010 | if (IS_ERR(ct)) |
c88130bc | 1011 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 | 1012 | |
48b1de4c PM |
1013 | if (tmpl && nfct_synproxy(tmpl)) { |
1014 | nfct_seqadj_ext_add(ct); | |
1015 | nfct_synproxy_ext_add(ct); | |
1016 | } | |
1017 | ||
60b5f8f7 | 1018 | timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; |
ae2d708e PNA |
1019 | if (timeout_ext) { |
1020 | timeouts = nf_ct_timeout_data(timeout_ext); | |
1021 | if (unlikely(!timeouts)) | |
1022 | timeouts = l4proto->get_timeouts(net); | |
1023 | } else { | |
60b5f8f7 | 1024 | timeouts = l4proto->get_timeouts(net); |
ae2d708e | 1025 | } |
60b5f8f7 | 1026 | |
2c8503f5 | 1027 | if (!l4proto->new(ct, skb, dataoff, timeouts)) { |
c88130bc | 1028 | nf_conntrack_free(ct); |
ccd63c20 | 1029 | pr_debug("can't track with proto module\n"); |
9fb9cbb1 YK |
1030 | return NULL; |
1031 | } | |
1032 | ||
60b5f8f7 | 1033 | if (timeout_ext) |
ae2d708e PNA |
1034 | nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout), |
1035 | GFP_ATOMIC); | |
60b5f8f7 | 1036 | |
58401572 | 1037 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a992ca2a | 1038 | nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); |
c539f017 | 1039 | nf_ct_labels_ext_add(ct); |
b2a15a60 PM |
1040 | |
1041 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
1042 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
1043 | ecache ? ecache->expmask : 0, | |
1044 | GFP_ATOMIC); | |
58401572 | 1045 | |
ca7433df JDB |
1046 | local_bh_disable(); |
1047 | if (net->ct.expect_count) { | |
1048 | spin_lock(&nf_conntrack_expect_lock); | |
1049 | exp = nf_ct_find_expectation(net, zone, tuple); | |
1050 | if (exp) { | |
ccd63c20 | 1051 | pr_debug("expectation arrives ct=%p exp=%p\n", |
ca7433df JDB |
1052 | ct, exp); |
1053 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
1054 | __set_bit(IPS_EXPECTED_BIT, &ct->status); | |
1055 | /* exp->master safe, refcnt bumped in nf_ct_find_expectation */ | |
1056 | ct->master = exp->master; | |
1057 | if (exp->helper) { | |
1058 | help = nf_ct_helper_ext_add(ct, exp->helper, | |
1059 | GFP_ATOMIC); | |
1060 | if (help) | |
1061 | rcu_assign_pointer(help->helper, exp->helper); | |
1062 | } | |
ceceae1b | 1063 | |
9fb9cbb1 | 1064 | #ifdef CONFIG_NF_CONNTRACK_MARK |
ca7433df | 1065 | ct->mark = exp->master->mark; |
7c9728c3 JM |
1066 | #endif |
1067 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
ca7433df | 1068 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 1069 | #endif |
ca7433df JDB |
1070 | NF_CT_STAT_INC(net, expect_new); |
1071 | } | |
1072 | spin_unlock(&nf_conntrack_expect_lock); | |
1073 | } | |
1074 | if (!exp) { | |
b2a15a60 | 1075 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
0d55af87 | 1076 | NF_CT_STAT_INC(net, new); |
22e7410b | 1077 | } |
9fb9cbb1 | 1078 | |
e53376be PNA |
1079 | /* Now it is inserted into the unconfirmed list, bump refcount */ |
1080 | nf_conntrack_get(&ct->ct_general); | |
b7779d06 | 1081 | nf_ct_add_to_unconfirmed_list(ct); |
9fb9cbb1 | 1082 | |
ca7433df | 1083 | local_bh_enable(); |
9fb9cbb1 YK |
1084 | |
1085 | if (exp) { | |
1086 | if (exp->expectfn) | |
c88130bc | 1087 | exp->expectfn(ct, exp); |
6823645d | 1088 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
1089 | } |
1090 | ||
c88130bc | 1091 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
1092 | } |
1093 | ||
1094 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
1095 | static inline struct nf_conn * | |
b2a15a60 | 1096 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 1097 | struct sk_buff *skb, |
9fb9cbb1 YK |
1098 | unsigned int dataoff, |
1099 | u_int16_t l3num, | |
1100 | u_int8_t protonum, | |
1101 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 1102 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 1103 | int *set_reply, |
60b5f8f7 | 1104 | enum ip_conntrack_info *ctinfo) |
9fb9cbb1 | 1105 | { |
308ac914 | 1106 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 YK |
1107 | struct nf_conntrack_tuple tuple; |
1108 | struct nf_conntrack_tuple_hash *h; | |
5e8018fc | 1109 | struct nf_conntrack_zone tmp; |
9fb9cbb1 | 1110 | struct nf_conn *ct; |
99f07e91 | 1111 | u32 hash; |
9fb9cbb1 | 1112 | |
bbe735e4 | 1113 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
a31f1adc | 1114 | dataoff, l3num, protonum, net, &tuple, l3proto, |
605dcad6 | 1115 | l4proto)) { |
ccd63c20 | 1116 | pr_debug("Can't get tuple\n"); |
9fb9cbb1 YK |
1117 | return NULL; |
1118 | } | |
1119 | ||
1120 | /* look for tuple match */ | |
5e8018fc | 1121 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
1b8c8a9f | 1122 | hash = hash_conntrack_raw(&tuple, net); |
99f07e91 | 1123 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); |
9fb9cbb1 | 1124 | if (!h) { |
b2a15a60 | 1125 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
60b5f8f7 | 1126 | skb, dataoff, hash); |
9fb9cbb1 YK |
1127 | if (!h) |
1128 | return NULL; | |
1129 | if (IS_ERR(h)) | |
1130 | return (void *)h; | |
1131 | } | |
1132 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1133 | ||
1134 | /* It exists; we have (non-exclusive) reference. */ | |
1135 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
fb048833 | 1136 | *ctinfo = IP_CT_ESTABLISHED_REPLY; |
9fb9cbb1 YK |
1137 | /* Please set reply bit if this packet OK */ |
1138 | *set_reply = 1; | |
1139 | } else { | |
1140 | /* Once we've had two way comms, always ESTABLISHED. */ | |
1141 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
ccd63c20 | 1142 | pr_debug("normal packet for %p\n", ct); |
9fb9cbb1 YK |
1143 | *ctinfo = IP_CT_ESTABLISHED; |
1144 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
ccd63c20 | 1145 | pr_debug("related packet for %p\n", ct); |
9fb9cbb1 YK |
1146 | *ctinfo = IP_CT_RELATED; |
1147 | } else { | |
ccd63c20 | 1148 | pr_debug("new packet for %p\n", ct); |
9fb9cbb1 YK |
1149 | *ctinfo = IP_CT_NEW; |
1150 | } | |
1151 | *set_reply = 0; | |
1152 | } | |
1153 | skb->nfct = &ct->ct_general; | |
1154 | skb->nfctinfo = *ctinfo; | |
1155 | return ct; | |
1156 | } | |
1157 | ||
1158 | unsigned int | |
a702a65f AD |
1159 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
1160 | struct sk_buff *skb) | |
9fb9cbb1 | 1161 | { |
b2a15a60 | 1162 | struct nf_conn *ct, *tmpl = NULL; |
9fb9cbb1 YK |
1163 | enum ip_conntrack_info ctinfo; |
1164 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 1165 | struct nf_conntrack_l4proto *l4proto; |
2c8503f5 | 1166 | unsigned int *timeouts; |
9fb9cbb1 YK |
1167 | unsigned int dataoff; |
1168 | u_int8_t protonum; | |
1169 | int set_reply = 0; | |
1170 | int ret; | |
1171 | ||
3db05fea | 1172 | if (skb->nfct) { |
b2a15a60 PM |
1173 | /* Previously seen (loopback or untracked)? Ignore. */ |
1174 | tmpl = (struct nf_conn *)skb->nfct; | |
1175 | if (!nf_ct_is_template(tmpl)) { | |
1176 | NF_CT_STAT_INC_ATOMIC(net, ignore); | |
1177 | return NF_ACCEPT; | |
1178 | } | |
1179 | skb->nfct = NULL; | |
9fb9cbb1 YK |
1180 | } |
1181 | ||
923f4902 | 1182 | /* rcu_read_lock()ed by nf_hook_slow */ |
76108cea | 1183 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 1184 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
1185 | &dataoff, &protonum); |
1186 | if (ret <= 0) { | |
25985edc | 1187 | pr_debug("not prepared to track yet or error occurred\n"); |
0d55af87 AD |
1188 | NF_CT_STAT_INC_ATOMIC(net, error); |
1189 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1190 | ret = -ret; |
1191 | goto out; | |
9fb9cbb1 YK |
1192 | } |
1193 | ||
76108cea | 1194 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
1195 | |
1196 | /* It may be an special packet, error, unclean... | |
1197 | * inverse of the return code tells to the netfilter | |
1198 | * core what to do with the packet. */ | |
74c51a14 | 1199 | if (l4proto->error != NULL) { |
8fea97ec PM |
1200 | ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo, |
1201 | pf, hooknum); | |
74c51a14 | 1202 | if (ret <= 0) { |
0d55af87 AD |
1203 | NF_CT_STAT_INC_ATOMIC(net, error); |
1204 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1205 | ret = -ret; |
1206 | goto out; | |
74c51a14 | 1207 | } |
88ed01d1 PNA |
1208 | /* ICMP[v6] protocol trackers may assign one conntrack. */ |
1209 | if (skb->nfct) | |
1210 | goto out; | |
9fb9cbb1 YK |
1211 | } |
1212 | ||
b2a15a60 | 1213 | ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
60b5f8f7 | 1214 | l3proto, l4proto, &set_reply, &ctinfo); |
9fb9cbb1 YK |
1215 | if (!ct) { |
1216 | /* Not valid part of a connection */ | |
0d55af87 | 1217 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
b2a15a60 PM |
1218 | ret = NF_ACCEPT; |
1219 | goto out; | |
9fb9cbb1 YK |
1220 | } |
1221 | ||
1222 | if (IS_ERR(ct)) { | |
1223 | /* Too stressed to deal. */ | |
0d55af87 | 1224 | NF_CT_STAT_INC_ATOMIC(net, drop); |
b2a15a60 PM |
1225 | ret = NF_DROP; |
1226 | goto out; | |
9fb9cbb1 YK |
1227 | } |
1228 | ||
3db05fea | 1229 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 1230 | |
60b5f8f7 | 1231 | /* Decide what timeout policy we want to apply to this flow. */ |
84b5ee93 | 1232 | timeouts = nf_ct_timeout_lookup(net, ct, l4proto); |
60b5f8f7 | 1233 | |
2c8503f5 | 1234 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); |
ec8d5409 | 1235 | if (ret <= 0) { |
9fb9cbb1 YK |
1236 | /* Invalid: inverse of the return code tells |
1237 | * the netfilter core what to do */ | |
0d53778e | 1238 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
1239 | nf_conntrack_put(skb->nfct); |
1240 | skb->nfct = NULL; | |
0d55af87 | 1241 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
1242 | if (ret == -NF_DROP) |
1243 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
b2a15a60 PM |
1244 | ret = -ret; |
1245 | goto out; | |
9fb9cbb1 YK |
1246 | } |
1247 | ||
1248 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 1249 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 | 1250 | out: |
c3174286 PNA |
1251 | if (tmpl) { |
1252 | /* Special case: we have to repeat this hook, assign the | |
1253 | * template again to this packet. We assume that this packet | |
1254 | * has no conntrack assigned. This is used by nf_ct_tcp. */ | |
1255 | if (ret == NF_REPEAT) | |
1256 | skb->nfct = (struct nf_conntrack *)tmpl; | |
1257 | else | |
1258 | nf_ct_put(tmpl); | |
1259 | } | |
9fb9cbb1 YK |
1260 | |
1261 | return ret; | |
1262 | } | |
13b18339 | 1263 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 1264 | |
5f2b4c90 JE |
1265 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
1266 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 1267 | { |
5f2b4c90 | 1268 | bool ret; |
923f4902 PM |
1269 | |
1270 | rcu_read_lock(); | |
1271 | ret = nf_ct_invert_tuple(inverse, orig, | |
1272 | __nf_ct_l3proto_find(orig->src.l3num), | |
1273 | __nf_ct_l4proto_find(orig->src.l3num, | |
1274 | orig->dst.protonum)); | |
1275 | rcu_read_unlock(); | |
1276 | return ret; | |
9fb9cbb1 | 1277 | } |
13b18339 | 1278 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 1279 | |
5b1158e9 JK |
1280 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
1281 | implicitly racy: see __nf_conntrack_confirm */ | |
1282 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
1283 | const struct nf_conntrack_tuple *newreply) | |
1284 | { | |
1285 | struct nf_conn_help *help = nfct_help(ct); | |
1286 | ||
5b1158e9 JK |
1287 | /* Should be unconfirmed, so not in hash table yet */ |
1288 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
1289 | ||
0d53778e | 1290 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 1291 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
1292 | |
1293 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 1294 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 1295 | return; |
ceceae1b | 1296 | |
c52fbb41 | 1297 | rcu_read_lock(); |
b2a15a60 | 1298 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 1299 | rcu_read_unlock(); |
5b1158e9 | 1300 | } |
13b18339 | 1301 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 1302 | |
9fb9cbb1 YK |
1303 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
1304 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
1305 | enum ip_conntrack_info ctinfo, | |
1306 | const struct sk_buff *skb, | |
1307 | unsigned long extra_jiffies, | |
1308 | int do_acct) | |
1309 | { | |
9fb9cbb1 YK |
1310 | NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); |
1311 | NF_CT_ASSERT(skb); | |
1312 | ||
997ae831 | 1313 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1314 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1315 | goto acct; | |
997ae831 | 1316 | |
9fb9cbb1 YK |
1317 | /* If not in hash table, timer will not be active yet */ |
1318 | if (!nf_ct_is_confirmed(ct)) { | |
1319 | ct->timeout.expires = extra_jiffies; | |
9fb9cbb1 | 1320 | } else { |
be00c8e4 MJ |
1321 | unsigned long newtime = jiffies + extra_jiffies; |
1322 | ||
1323 | /* Only update the timeout if the new timeout is at least | |
1324 | HZ jiffies from the old timeout. Need del_timer for race | |
1325 | avoidance (may already be dying). */ | |
65cb9fda PM |
1326 | if (newtime - ct->timeout.expires >= HZ) |
1327 | mod_timer_pending(&ct->timeout, newtime); | |
9fb9cbb1 YK |
1328 | } |
1329 | ||
47d95045 | 1330 | acct: |
ba76738c PNA |
1331 | if (do_acct) |
1332 | nf_ct_acct_update(ct, ctinfo, skb->len); | |
9fb9cbb1 | 1333 | } |
13b18339 | 1334 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1335 | |
4c889498 DM |
1336 | bool __nf_ct_kill_acct(struct nf_conn *ct, |
1337 | enum ip_conntrack_info ctinfo, | |
1338 | const struct sk_buff *skb, | |
1339 | int do_acct) | |
51091764 | 1340 | { |
ba76738c PNA |
1341 | if (do_acct) |
1342 | nf_ct_acct_update(ct, ctinfo, skb->len); | |
58401572 | 1343 | |
4c889498 | 1344 | if (del_timer(&ct->timeout)) { |
51091764 | 1345 | ct->timeout.function((unsigned long)ct); |
4c889498 DM |
1346 | return true; |
1347 | } | |
1348 | return false; | |
51091764 | 1349 | } |
718d4ad9 | 1350 | EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); |
51091764 | 1351 | |
c0cd1156 | 1352 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
c1d10adb PNA |
1353 | |
1354 | #include <linux/netfilter/nfnetlink.h> | |
1355 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1356 | #include <linux/mutex.h> |
1357 | ||
c1d10adb PNA |
1358 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1359 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1360 | * or depend on ctnetlink */ | |
fdf70832 | 1361 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1362 | const struct nf_conntrack_tuple *tuple) |
1363 | { | |
bae65be8 DM |
1364 | if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) || |
1365 | nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port)) | |
1366 | goto nla_put_failure; | |
c1d10adb PNA |
1367 | return 0; |
1368 | ||
df6fb868 | 1369 | nla_put_failure: |
c1d10adb PNA |
1370 | return -1; |
1371 | } | |
fdf70832 | 1372 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1373 | |
f73e924c PM |
1374 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1375 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1376 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1377 | }; |
f73e924c | 1378 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1379 | |
fdf70832 | 1380 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1381 | struct nf_conntrack_tuple *t) |
1382 | { | |
df6fb868 | 1383 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1384 | return -EINVAL; |
1385 | ||
77236b6e PM |
1386 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1387 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1388 | |
1389 | return 0; | |
1390 | } | |
fdf70832 | 1391 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1392 | |
1393 | int nf_ct_port_nlattr_tuple_size(void) | |
1394 | { | |
1395 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1396 | } | |
1397 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1398 | #endif |
1399 | ||
9fb9cbb1 | 1400 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
312a0c16 | 1401 | static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb) |
9fb9cbb1 YK |
1402 | { |
1403 | struct nf_conn *ct; | |
1404 | enum ip_conntrack_info ctinfo; | |
1405 | ||
1406 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1407 | ct = nf_ct_get(skb, &ctinfo); | |
1408 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
fb048833 | 1409 | ctinfo = IP_CT_RELATED_REPLY; |
9fb9cbb1 YK |
1410 | else |
1411 | ctinfo = IP_CT_RELATED; | |
1412 | ||
1413 | /* Attach to new skbuff, and increment count */ | |
1414 | nskb->nfct = &ct->ct_general; | |
1415 | nskb->nfctinfo = ctinfo; | |
1416 | nf_conntrack_get(nskb->nfct); | |
1417 | } | |
1418 | ||
9fb9cbb1 | 1419 | /* Bring out ya dead! */ |
df0933dc | 1420 | static struct nf_conn * |
400dad39 | 1421 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1422 | void *data, unsigned int *bucket) |
1423 | { | |
df0933dc PM |
1424 | struct nf_conntrack_tuple_hash *h; |
1425 | struct nf_conn *ct; | |
ea781f19 | 1426 | struct hlist_nulls_node *n; |
b7779d06 | 1427 | int cpu; |
93bb0ceb | 1428 | spinlock_t *lockp; |
9fb9cbb1 | 1429 | |
56d52d48 | 1430 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
93bb0ceb JDB |
1431 | lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS]; |
1432 | local_bh_disable(); | |
b16c2919 | 1433 | nf_conntrack_lock(lockp); |
56d52d48 FW |
1434 | if (*bucket < nf_conntrack_htable_size) { |
1435 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) { | |
93bb0ceb JDB |
1436 | if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL) |
1437 | continue; | |
1438 | ct = nf_ct_tuplehash_to_ctrack(h); | |
e0c7d472 FW |
1439 | if (net_eq(nf_ct_net(ct), net) && |
1440 | iter(ct, data)) | |
93bb0ceb JDB |
1441 | goto found; |
1442 | } | |
df0933dc | 1443 | } |
93bb0ceb JDB |
1444 | spin_unlock(lockp); |
1445 | local_bh_enable(); | |
d93c6258 | 1446 | cond_resched(); |
601e68e1 | 1447 | } |
b7779d06 JDB |
1448 | |
1449 | for_each_possible_cpu(cpu) { | |
1450 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1451 | ||
1452 | spin_lock_bh(&pcpu->lock); | |
1453 | hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) { | |
1454 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1455 | if (iter(ct, data)) | |
1456 | set_bit(IPS_DYING_BIT, &ct->status); | |
1457 | } | |
1458 | spin_unlock_bh(&pcpu->lock); | |
d93c6258 | 1459 | cond_resched(); |
b7779d06 | 1460 | } |
df0933dc PM |
1461 | return NULL; |
1462 | found: | |
c073e3fa | 1463 | atomic_inc(&ct->ct_general.use); |
93bb0ceb JDB |
1464 | spin_unlock(lockp); |
1465 | local_bh_enable(); | |
df0933dc | 1466 | return ct; |
9fb9cbb1 YK |
1467 | } |
1468 | ||
400dad39 AD |
1469 | void nf_ct_iterate_cleanup(struct net *net, |
1470 | int (*iter)(struct nf_conn *i, void *data), | |
c655bc68 | 1471 | void *data, u32 portid, int report) |
9fb9cbb1 | 1472 | { |
df0933dc | 1473 | struct nf_conn *ct; |
9fb9cbb1 YK |
1474 | unsigned int bucket = 0; |
1475 | ||
d93c6258 FW |
1476 | might_sleep(); |
1477 | ||
88b68bc5 FW |
1478 | if (atomic_read(&net->ct.count) == 0) |
1479 | return; | |
1480 | ||
400dad39 | 1481 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 YK |
1482 | /* Time to push up daises... */ |
1483 | if (del_timer(&ct->timeout)) | |
c655bc68 | 1484 | nf_ct_delete(ct, portid, report); |
02982c27 | 1485 | |
9fb9cbb1 YK |
1486 | /* ... else the timer will get him soon. */ |
1487 | ||
1488 | nf_ct_put(ct); | |
d93c6258 | 1489 | cond_resched(); |
9fb9cbb1 YK |
1490 | } |
1491 | } | |
13b18339 | 1492 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1493 | |
274d383b PNA |
1494 | static int kill_all(struct nf_conn *i, void *data) |
1495 | { | |
1496 | return 1; | |
1497 | } | |
1498 | ||
d862a662 | 1499 | void nf_ct_free_hashtable(void *hash, unsigned int size) |
9fb9cbb1 | 1500 | { |
d862a662 | 1501 | if (is_vmalloc_addr(hash)) |
9fb9cbb1 YK |
1502 | vfree(hash); |
1503 | else | |
601e68e1 | 1504 | free_pages((unsigned long)hash, |
f205c5e0 | 1505 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1506 | } |
ac565e5f | 1507 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1508 | |
b3c5163f ED |
1509 | static int untrack_refs(void) |
1510 | { | |
1511 | int cnt = 0, cpu; | |
1512 | ||
1513 | for_each_possible_cpu(cpu) { | |
1514 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
1515 | ||
1516 | cnt += atomic_read(&ct->ct_general.use) - 1; | |
1517 | } | |
1518 | return cnt; | |
1519 | } | |
1520 | ||
f94161c1 | 1521 | void nf_conntrack_cleanup_start(void) |
9fb9cbb1 | 1522 | { |
f94161c1 G |
1523 | RCU_INIT_POINTER(ip_ct_attach, NULL); |
1524 | } | |
1525 | ||
1526 | void nf_conntrack_cleanup_end(void) | |
1527 | { | |
1528 | RCU_INIT_POINTER(nf_ct_destroy, NULL); | |
b3c5163f | 1529 | while (untrack_refs() > 0) |
9edd7ca0 PM |
1530 | schedule(); |
1531 | ||
56d52d48 FW |
1532 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
1533 | ||
04d87001 | 1534 | nf_conntrack_proto_fini(); |
41d73ec0 | 1535 | nf_conntrack_seqadj_fini(); |
5f69b8f5 | 1536 | nf_conntrack_labels_fini(); |
5e615b22 | 1537 | nf_conntrack_helper_fini(); |
8684094c | 1538 | nf_conntrack_timeout_fini(); |
3fe0f943 | 1539 | nf_conntrack_ecache_fini(); |
73f4001a | 1540 | nf_conntrack_tstamp_fini(); |
b7ff3a1f | 1541 | nf_conntrack_acct_fini(); |
83b4dbe1 | 1542 | nf_conntrack_expect_fini(); |
77571149 FW |
1543 | |
1544 | kmem_cache_destroy(nf_conntrack_cachep); | |
08f6547d | 1545 | } |
9fb9cbb1 | 1546 | |
f94161c1 G |
1547 | /* |
1548 | * Mishearing the voices in his head, our hero wonders how he's | |
1549 | * supposed to kill the mall. | |
1550 | */ | |
1551 | void nf_conntrack_cleanup_net(struct net *net) | |
08f6547d | 1552 | { |
dece40e8 VD |
1553 | LIST_HEAD(single); |
1554 | ||
1555 | list_add(&net->exit_list, &single); | |
1556 | nf_conntrack_cleanup_net_list(&single); | |
1557 | } | |
1558 | ||
1559 | void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list) | |
1560 | { | |
1561 | int busy; | |
1562 | struct net *net; | |
1563 | ||
f94161c1 G |
1564 | /* |
1565 | * This makes sure all current packets have passed through | |
1566 | * netfilter framework. Roll on, two-stage module | |
1567 | * delete... | |
1568 | */ | |
1569 | synchronize_net(); | |
dece40e8 VD |
1570 | i_see_dead_people: |
1571 | busy = 0; | |
1572 | list_for_each_entry(net, net_exit_list, exit_list) { | |
c655bc68 | 1573 | nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0); |
dece40e8 VD |
1574 | if (atomic_read(&net->ct.count) != 0) |
1575 | busy = 1; | |
1576 | } | |
1577 | if (busy) { | |
9fb9cbb1 YK |
1578 | schedule(); |
1579 | goto i_see_dead_people; | |
1580 | } | |
1581 | ||
dece40e8 | 1582 | list_for_each_entry(net, net_exit_list, exit_list) { |
dece40e8 VD |
1583 | nf_conntrack_proto_pernet_fini(net); |
1584 | nf_conntrack_helper_pernet_fini(net); | |
1585 | nf_conntrack_ecache_pernet_fini(net); | |
1586 | nf_conntrack_tstamp_pernet_fini(net); | |
1587 | nf_conntrack_acct_pernet_fini(net); | |
1588 | nf_conntrack_expect_pernet_fini(net); | |
dece40e8 | 1589 | free_percpu(net->ct.stat); |
b7779d06 | 1590 | free_percpu(net->ct.pcpu_lists); |
dece40e8 | 1591 | } |
08f6547d AD |
1592 | } |
1593 | ||
d862a662 | 1594 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) |
9fb9cbb1 | 1595 | { |
ea781f19 ED |
1596 | struct hlist_nulls_head *hash; |
1597 | unsigned int nr_slots, i; | |
1598 | size_t sz; | |
9fb9cbb1 | 1599 | |
ea781f19 ED |
1600 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1601 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
1602 | sz = nr_slots * sizeof(struct hlist_nulls_head); | |
1603 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1604 | get_order(sz)); | |
f0ad4621 | 1605 | if (!hash) |
966567b7 | 1606 | hash = vzalloc(sz); |
9fb9cbb1 | 1607 | |
ea781f19 ED |
1608 | if (hash && nulls) |
1609 | for (i = 0; i < nr_slots; i++) | |
1610 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1611 | |
1612 | return hash; | |
1613 | } | |
ac565e5f | 1614 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1615 | |
3183ab89 | 1616 | int nf_conntrack_hash_resize(unsigned int hashsize) |
9fb9cbb1 | 1617 | { |
3183ab89 FW |
1618 | int i, bucket; |
1619 | unsigned int old_size; | |
ea781f19 | 1620 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1621 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1622 | struct nf_conn *ct; |
9fb9cbb1 | 1623 | |
9fb9cbb1 YK |
1624 | if (!hashsize) |
1625 | return -EINVAL; | |
1626 | ||
d862a662 | 1627 | hash = nf_ct_alloc_hashtable(&hashsize, 1); |
9fb9cbb1 YK |
1628 | if (!hash) |
1629 | return -ENOMEM; | |
1630 | ||
3183ab89 FW |
1631 | old_size = nf_conntrack_htable_size; |
1632 | if (old_size == hashsize) { | |
1633 | nf_ct_free_hashtable(hash, hashsize); | |
1634 | return 0; | |
1635 | } | |
1636 | ||
93bb0ceb JDB |
1637 | local_bh_disable(); |
1638 | nf_conntrack_all_lock(); | |
a3efd812 | 1639 | write_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb | 1640 | |
76507f69 PM |
1641 | /* Lookups in the old hash might happen in parallel, which means we |
1642 | * might get false negatives during connection lookup. New connections | |
1643 | * created because of a false negative won't make it into the hash | |
93bb0ceb | 1644 | * though since that required taking the locks. |
76507f69 | 1645 | */ |
93bb0ceb | 1646 | |
56d52d48 FW |
1647 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
1648 | while (!hlist_nulls_empty(&nf_conntrack_hash[i])) { | |
1649 | h = hlist_nulls_entry(nf_conntrack_hash[i].first, | |
1650 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1651 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1652 | hlist_nulls_del_rcu(&h->hnnode); |
1b8c8a9f FW |
1653 | bucket = __hash_conntrack(nf_ct_net(ct), |
1654 | &h->tuple, hashsize); | |
ea781f19 | 1655 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1656 | } |
1657 | } | |
56d52d48 FW |
1658 | old_size = nf_conntrack_htable_size; |
1659 | old_hash = nf_conntrack_hash; | |
9fb9cbb1 | 1660 | |
56d52d48 FW |
1661 | nf_conntrack_hash = hash; |
1662 | nf_conntrack_htable_size = hashsize; | |
93bb0ceb | 1663 | |
a3efd812 | 1664 | write_seqcount_end(&nf_conntrack_generation); |
93bb0ceb JDB |
1665 | nf_conntrack_all_unlock(); |
1666 | local_bh_enable(); | |
9fb9cbb1 | 1667 | |
5e3c61f9 | 1668 | synchronize_net(); |
d862a662 | 1669 | nf_ct_free_hashtable(old_hash, old_size); |
9fb9cbb1 YK |
1670 | return 0; |
1671 | } | |
3183ab89 FW |
1672 | |
1673 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) | |
1674 | { | |
1675 | unsigned int hashsize; | |
1676 | int rc; | |
1677 | ||
1678 | if (current->nsproxy->net_ns != &init_net) | |
1679 | return -EOPNOTSUPP; | |
1680 | ||
1681 | /* On boot, we can set this without any fancy locking. */ | |
1682 | if (!nf_conntrack_htable_size) | |
1683 | return param_set_uint(val, kp); | |
1684 | ||
1685 | rc = kstrtouint(val, 0, &hashsize); | |
1686 | if (rc) | |
1687 | return rc; | |
1688 | ||
1689 | return nf_conntrack_hash_resize(hashsize); | |
1690 | } | |
fae718dd | 1691 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1692 | |
fae718dd | 1693 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1694 | &nf_conntrack_htable_size, 0600); |
1695 | ||
5bfddbd4 ED |
1696 | void nf_ct_untracked_status_or(unsigned long bits) |
1697 | { | |
b3c5163f ED |
1698 | int cpu; |
1699 | ||
1700 | for_each_possible_cpu(cpu) | |
1701 | per_cpu(nf_conntrack_untracked, cpu).status |= bits; | |
5bfddbd4 ED |
1702 | } |
1703 | EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); | |
1704 | ||
f94161c1 | 1705 | int nf_conntrack_init_start(void) |
9fb9cbb1 | 1706 | { |
f205c5e0 | 1707 | int max_factor = 8; |
0c5366b3 FW |
1708 | int ret = -ENOMEM; |
1709 | int i, cpu; | |
93bb0ceb | 1710 | |
a3efd812 FW |
1711 | seqcount_init(&nf_conntrack_generation); |
1712 | ||
d5d20912 | 1713 | for (i = 0; i < CONNTRACK_LOCKS; i++) |
93bb0ceb | 1714 | spin_lock_init(&nf_conntrack_locks[i]); |
9fb9cbb1 | 1715 | |
9fb9cbb1 | 1716 | if (!nf_conntrack_htable_size) { |
88eab472 ML |
1717 | /* Idea from tcp.c: use 1/16384 of memory. |
1718 | * On i386: 32MB machine has 512 buckets. | |
1719 | * >= 1GB machines have 16384 buckets. | |
1720 | * >= 4GB machines have 65536 buckets. | |
1721 | */ | |
9fb9cbb1 | 1722 | nf_conntrack_htable_size |
4481374c | 1723 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1724 | / sizeof(struct hlist_head)); |
88eab472 ML |
1725 | if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE))) |
1726 | nf_conntrack_htable_size = 65536; | |
1727 | else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
f205c5e0 PM |
1728 | nf_conntrack_htable_size = 16384; |
1729 | if (nf_conntrack_htable_size < 32) | |
1730 | nf_conntrack_htable_size = 32; | |
1731 | ||
1732 | /* Use a max. factor of four by default to get the same max as | |
1733 | * with the old struct list_heads. When a table size is given | |
1734 | * we use the old value of 8 to avoid reducing the max. | |
1735 | * entries. */ | |
1736 | max_factor = 4; | |
9fb9cbb1 | 1737 | } |
56d52d48 FW |
1738 | |
1739 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1); | |
1740 | if (!nf_conntrack_hash) | |
1741 | return -ENOMEM; | |
1742 | ||
f205c5e0 | 1743 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 1744 | |
0c5366b3 FW |
1745 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1746 | sizeof(struct nf_conn), 0, | |
5a75cdeb | 1747 | SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL); |
0c5366b3 FW |
1748 | if (!nf_conntrack_cachep) |
1749 | goto err_cachep; | |
1750 | ||
654d0fbd | 1751 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
1752 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
1753 | nf_conntrack_max); | |
83b4dbe1 G |
1754 | |
1755 | ret = nf_conntrack_expect_init(); | |
1756 | if (ret < 0) | |
1757 | goto err_expect; | |
1758 | ||
b7ff3a1f G |
1759 | ret = nf_conntrack_acct_init(); |
1760 | if (ret < 0) | |
1761 | goto err_acct; | |
1762 | ||
73f4001a G |
1763 | ret = nf_conntrack_tstamp_init(); |
1764 | if (ret < 0) | |
1765 | goto err_tstamp; | |
1766 | ||
3fe0f943 G |
1767 | ret = nf_conntrack_ecache_init(); |
1768 | if (ret < 0) | |
1769 | goto err_ecache; | |
1770 | ||
8684094c G |
1771 | ret = nf_conntrack_timeout_init(); |
1772 | if (ret < 0) | |
1773 | goto err_timeout; | |
1774 | ||
5e615b22 G |
1775 | ret = nf_conntrack_helper_init(); |
1776 | if (ret < 0) | |
1777 | goto err_helper; | |
1778 | ||
5f69b8f5 G |
1779 | ret = nf_conntrack_labels_init(); |
1780 | if (ret < 0) | |
1781 | goto err_labels; | |
1782 | ||
41d73ec0 PM |
1783 | ret = nf_conntrack_seqadj_init(); |
1784 | if (ret < 0) | |
1785 | goto err_seqadj; | |
1786 | ||
04d87001 G |
1787 | ret = nf_conntrack_proto_init(); |
1788 | if (ret < 0) | |
1789 | goto err_proto; | |
1790 | ||
9edd7ca0 | 1791 | /* Set up fake conntrack: to never be deleted, not in any hashes */ |
b3c5163f ED |
1792 | for_each_possible_cpu(cpu) { |
1793 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
b3c5163f ED |
1794 | write_pnet(&ct->ct_net, &init_net); |
1795 | atomic_set(&ct->ct_general.use, 1); | |
1796 | } | |
9edd7ca0 | 1797 | /* - and look it like as a confirmed connection */ |
5bfddbd4 | 1798 | nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); |
08f6547d AD |
1799 | return 0; |
1800 | ||
04d87001 | 1801 | err_proto: |
41d73ec0 PM |
1802 | nf_conntrack_seqadj_fini(); |
1803 | err_seqadj: | |
04d87001 | 1804 | nf_conntrack_labels_fini(); |
5f69b8f5 G |
1805 | err_labels: |
1806 | nf_conntrack_helper_fini(); | |
5e615b22 G |
1807 | err_helper: |
1808 | nf_conntrack_timeout_fini(); | |
8684094c G |
1809 | err_timeout: |
1810 | nf_conntrack_ecache_fini(); | |
3fe0f943 G |
1811 | err_ecache: |
1812 | nf_conntrack_tstamp_fini(); | |
73f4001a G |
1813 | err_tstamp: |
1814 | nf_conntrack_acct_fini(); | |
b7ff3a1f G |
1815 | err_acct: |
1816 | nf_conntrack_expect_fini(); | |
83b4dbe1 | 1817 | err_expect: |
0c5366b3 FW |
1818 | kmem_cache_destroy(nf_conntrack_cachep); |
1819 | err_cachep: | |
56d52d48 | 1820 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
08f6547d AD |
1821 | return ret; |
1822 | } | |
1823 | ||
f94161c1 G |
1824 | void nf_conntrack_init_end(void) |
1825 | { | |
1826 | /* For use by REJECT target */ | |
1827 | RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); | |
1828 | RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); | |
f94161c1 G |
1829 | } |
1830 | ||
8cc20198 ED |
1831 | /* |
1832 | * We need to use special "null" values, not used in hash table | |
1833 | */ | |
1834 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1835 | #define DYING_NULLS_VAL ((1<<30)+1) | |
252b3e8c | 1836 | #define TEMPLATE_NULLS_VAL ((1<<30)+2) |
8cc20198 | 1837 | |
f94161c1 | 1838 | int nf_conntrack_init_net(struct net *net) |
08f6547d | 1839 | { |
b7779d06 JDB |
1840 | int ret = -ENOMEM; |
1841 | int cpu; | |
ceceae1b | 1842 | |
08f6547d | 1843 | atomic_set(&net->ct.count, 0); |
b7779d06 JDB |
1844 | |
1845 | net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu); | |
1846 | if (!net->ct.pcpu_lists) | |
08f6547d | 1847 | goto err_stat; |
b7779d06 JDB |
1848 | |
1849 | for_each_possible_cpu(cpu) { | |
1850 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1851 | ||
1852 | spin_lock_init(&pcpu->lock); | |
1853 | INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL); | |
1854 | INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL); | |
08f6547d | 1855 | } |
5b3501fa | 1856 | |
b7779d06 JDB |
1857 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1858 | if (!net->ct.stat) | |
1859 | goto err_pcpu_lists; | |
1860 | ||
83b4dbe1 | 1861 | ret = nf_conntrack_expect_pernet_init(net); |
08f6547d AD |
1862 | if (ret < 0) |
1863 | goto err_expect; | |
b7ff3a1f | 1864 | ret = nf_conntrack_acct_pernet_init(net); |
58401572 | 1865 | if (ret < 0) |
08f6547d | 1866 | goto err_acct; |
73f4001a | 1867 | ret = nf_conntrack_tstamp_pernet_init(net); |
a992ca2a PNA |
1868 | if (ret < 0) |
1869 | goto err_tstamp; | |
3fe0f943 | 1870 | ret = nf_conntrack_ecache_pernet_init(net); |
a0891aa6 PNA |
1871 | if (ret < 0) |
1872 | goto err_ecache; | |
5e615b22 | 1873 | ret = nf_conntrack_helper_pernet_init(net); |
a9006892 EL |
1874 | if (ret < 0) |
1875 | goto err_helper; | |
04d87001 | 1876 | ret = nf_conntrack_proto_pernet_init(net); |
f94161c1 G |
1877 | if (ret < 0) |
1878 | goto err_proto; | |
08f6547d | 1879 | return 0; |
c539f017 | 1880 | |
f94161c1 | 1881 | err_proto: |
5e615b22 | 1882 | nf_conntrack_helper_pernet_fini(net); |
a9006892 | 1883 | err_helper: |
3fe0f943 | 1884 | nf_conntrack_ecache_pernet_fini(net); |
a0891aa6 | 1885 | err_ecache: |
73f4001a | 1886 | nf_conntrack_tstamp_pernet_fini(net); |
a992ca2a | 1887 | err_tstamp: |
b7ff3a1f | 1888 | nf_conntrack_acct_pernet_fini(net); |
08f6547d | 1889 | err_acct: |
83b4dbe1 | 1890 | nf_conntrack_expect_pernet_fini(net); |
08f6547d | 1891 | err_expect: |
0d55af87 | 1892 | free_percpu(net->ct.stat); |
b7779d06 JDB |
1893 | err_pcpu_lists: |
1894 | free_percpu(net->ct.pcpu_lists); | |
0d55af87 | 1895 | err_stat: |
08f6547d AD |
1896 | return ret; |
1897 | } |