]>
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 | ||
b87a2f91 FW |
75 | struct conntrack_gc_work { |
76 | struct delayed_work dwork; | |
77 | u32 last_bucket; | |
78 | bool exiting; | |
c6dd940b | 79 | bool early_drop; |
e0df8cae | 80 | long next_gc_run; |
b87a2f91 FW |
81 | }; |
82 | ||
0c5366b3 | 83 | static __read_mostly struct kmem_cache *nf_conntrack_cachep; |
b16c2919 | 84 | static __read_mostly spinlock_t nf_conntrack_locks_all_lock; |
70d72b7e | 85 | static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock); |
b16c2919 SL |
86 | static __read_mostly bool nf_conntrack_locks_all; |
87 | ||
e0df8cae | 88 | /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */ |
e5072053 FW |
89 | #define GC_MAX_BUCKETS_DIV 128u |
90 | /* upper bound of full table scan */ | |
91 | #define GC_MAX_SCAN_JIFFIES (16u * HZ) | |
92 | /* desired ratio of entries found to be expired */ | |
93 | #define GC_EVICT_RATIO 50u | |
b87a2f91 FW |
94 | |
95 | static struct conntrack_gc_work conntrack_gc_work; | |
96 | ||
b16c2919 SL |
97 | void nf_conntrack_lock(spinlock_t *lock) __acquires(lock) |
98 | { | |
99 | spin_lock(lock); | |
100 | while (unlikely(nf_conntrack_locks_all)) { | |
101 | spin_unlock(lock); | |
b316ff78 PZ |
102 | |
103 | /* | |
104 | * Order the 'nf_conntrack_locks_all' load vs. the | |
105 | * spin_unlock_wait() loads below, to ensure | |
106 | * that 'nf_conntrack_locks_all_lock' is indeed held: | |
107 | */ | |
108 | smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */ | |
e39365be | 109 | spin_unlock_wait(&nf_conntrack_locks_all_lock); |
b16c2919 SL |
110 | spin_lock(lock); |
111 | } | |
112 | } | |
113 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); | |
114 | ||
93bb0ceb JDB |
115 | static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2) |
116 | { | |
117 | h1 %= CONNTRACK_LOCKS; | |
118 | h2 %= CONNTRACK_LOCKS; | |
119 | spin_unlock(&nf_conntrack_locks[h1]); | |
120 | if (h1 != h2) | |
121 | spin_unlock(&nf_conntrack_locks[h2]); | |
122 | } | |
123 | ||
124 | /* return true if we need to recompute hashes (in case hash table was resized) */ | |
125 | static bool nf_conntrack_double_lock(struct net *net, unsigned int h1, | |
126 | unsigned int h2, unsigned int sequence) | |
127 | { | |
128 | h1 %= CONNTRACK_LOCKS; | |
129 | h2 %= CONNTRACK_LOCKS; | |
130 | if (h1 <= h2) { | |
b16c2919 | 131 | nf_conntrack_lock(&nf_conntrack_locks[h1]); |
93bb0ceb JDB |
132 | if (h1 != h2) |
133 | spin_lock_nested(&nf_conntrack_locks[h2], | |
134 | SINGLE_DEPTH_NESTING); | |
135 | } else { | |
b16c2919 | 136 | nf_conntrack_lock(&nf_conntrack_locks[h2]); |
93bb0ceb JDB |
137 | spin_lock_nested(&nf_conntrack_locks[h1], |
138 | SINGLE_DEPTH_NESTING); | |
139 | } | |
a3efd812 | 140 | if (read_seqcount_retry(&nf_conntrack_generation, sequence)) { |
93bb0ceb JDB |
141 | nf_conntrack_double_unlock(h1, h2); |
142 | return true; | |
143 | } | |
144 | return false; | |
145 | } | |
146 | ||
147 | static void nf_conntrack_all_lock(void) | |
148 | { | |
149 | int i; | |
150 | ||
b16c2919 SL |
151 | spin_lock(&nf_conntrack_locks_all_lock); |
152 | nf_conntrack_locks_all = true; | |
153 | ||
b316ff78 PZ |
154 | /* |
155 | * Order the above store of 'nf_conntrack_locks_all' against | |
156 | * the spin_unlock_wait() loads below, such that if | |
157 | * nf_conntrack_lock() observes 'nf_conntrack_locks_all' | |
158 | * we must observe nf_conntrack_locks[] held: | |
159 | */ | |
160 | smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */ | |
161 | ||
b16c2919 | 162 | for (i = 0; i < CONNTRACK_LOCKS; i++) { |
e39365be | 163 | spin_unlock_wait(&nf_conntrack_locks[i]); |
b16c2919 | 164 | } |
93bb0ceb JDB |
165 | } |
166 | ||
167 | static void nf_conntrack_all_unlock(void) | |
168 | { | |
b316ff78 PZ |
169 | /* |
170 | * All prior stores must be complete before we clear | |
171 | * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock() | |
172 | * might observe the false value but not the entire | |
173 | * critical section: | |
174 | */ | |
175 | smp_store_release(&nf_conntrack_locks_all, false); | |
b16c2919 | 176 | spin_unlock(&nf_conntrack_locks_all_lock); |
93bb0ceb JDB |
177 | } |
178 | ||
e2b7606c | 179 | unsigned int nf_conntrack_htable_size __read_mostly; |
2567c4ea PNA |
180 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
181 | ||
e478075c | 182 | unsigned int nf_conntrack_max __read_mostly; |
92e47ba8 | 183 | seqcount_t nf_conntrack_generation __read_mostly; |
141658fb | 184 | static unsigned int nf_conntrack_hash_rnd __read_mostly; |
9fb9cbb1 | 185 | |
1b8c8a9f FW |
186 | static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, |
187 | const struct net *net) | |
9fb9cbb1 | 188 | { |
0794935e | 189 | unsigned int n; |
1b8c8a9f | 190 | u32 seed; |
0794935e | 191 | |
141658fb FW |
192 | get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd)); |
193 | ||
0794935e PM |
194 | /* The direction must be ignored, so we hash everything up to the |
195 | * destination ports (which is a multiple of 4) and treat the last | |
196 | * three bytes manually. | |
197 | */ | |
1b8c8a9f | 198 | seed = nf_conntrack_hash_rnd ^ net_hash_mix(net); |
0794935e | 199 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); |
1b8c8a9f | 200 | return jhash2((u32 *)tuple, n, seed ^ |
99f07e91 CG |
201 | (((__force __u16)tuple->dst.u.all << 16) | |
202 | tuple->dst.protonum)); | |
203 | } | |
204 | ||
56d52d48 | 205 | static u32 scale_hash(u32 hash) |
99f07e91 | 206 | { |
56d52d48 | 207 | return reciprocal_scale(hash, nf_conntrack_htable_size); |
99f07e91 | 208 | } |
0794935e | 209 | |
1b8c8a9f FW |
210 | static u32 __hash_conntrack(const struct net *net, |
211 | const struct nf_conntrack_tuple *tuple, | |
212 | unsigned int size) | |
99f07e91 | 213 | { |
1b8c8a9f | 214 | return reciprocal_scale(hash_conntrack_raw(tuple, net), size); |
9fb9cbb1 YK |
215 | } |
216 | ||
1b8c8a9f FW |
217 | static u32 hash_conntrack(const struct net *net, |
218 | const struct nf_conntrack_tuple *tuple) | |
9fb9cbb1 | 219 | { |
56d52d48 | 220 | return scale_hash(hash_conntrack_raw(tuple, net)); |
9fb9cbb1 YK |
221 | } |
222 | ||
5f2b4c90 | 223 | bool |
9fb9cbb1 YK |
224 | nf_ct_get_tuple(const struct sk_buff *skb, |
225 | unsigned int nhoff, | |
226 | unsigned int dataoff, | |
227 | u_int16_t l3num, | |
228 | u_int8_t protonum, | |
a31f1adc | 229 | struct net *net, |
9fb9cbb1 YK |
230 | struct nf_conntrack_tuple *tuple, |
231 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 232 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 233 | { |
443a70d5 | 234 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
235 | |
236 | tuple->src.l3num = l3num; | |
237 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 238 | return false; |
9fb9cbb1 YK |
239 | |
240 | tuple->dst.protonum = protonum; | |
241 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
242 | ||
a31f1adc | 243 | return l4proto->pkt_to_tuple(skb, dataoff, net, tuple); |
9fb9cbb1 | 244 | } |
13b18339 | 245 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 246 | |
5f2b4c90 | 247 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
a31f1adc EB |
248 | u_int16_t l3num, |
249 | struct net *net, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
250 | { |
251 | struct nf_conntrack_l3proto *l3proto; | |
252 | struct nf_conntrack_l4proto *l4proto; | |
253 | unsigned int protoff; | |
254 | u_int8_t protonum; | |
255 | int ret; | |
256 | ||
257 | rcu_read_lock(); | |
258 | ||
259 | l3proto = __nf_ct_l3proto_find(l3num); | |
260 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
261 | if (ret != NF_ACCEPT) { | |
262 | rcu_read_unlock(); | |
5f2b4c90 | 263 | return false; |
e2a3123f YK |
264 | } |
265 | ||
266 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
267 | ||
a31f1adc | 268 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple, |
e2a3123f YK |
269 | l3proto, l4proto); |
270 | ||
271 | rcu_read_unlock(); | |
272 | return ret; | |
273 | } | |
274 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
275 | ||
5f2b4c90 | 276 | bool |
9fb9cbb1 YK |
277 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
278 | const struct nf_conntrack_tuple *orig, | |
279 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 280 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 281 | { |
443a70d5 | 282 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
283 | |
284 | inverse->src.l3num = orig->src.l3num; | |
285 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 286 | return false; |
9fb9cbb1 YK |
287 | |
288 | inverse->dst.dir = !orig->dst.dir; | |
289 | ||
290 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 291 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 292 | } |
13b18339 | 293 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 294 | |
9fb9cbb1 YK |
295 | static void |
296 | clean_from_lists(struct nf_conn *ct) | |
297 | { | |
0d53778e | 298 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
299 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
300 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
301 | |
302 | /* Destroy all pending expectations */ | |
c1d10adb | 303 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
304 | } |
305 | ||
b7779d06 JDB |
306 | /* must be called with local_bh_disable */ |
307 | static void nf_ct_add_to_dying_list(struct nf_conn *ct) | |
308 | { | |
309 | struct ct_pcpu *pcpu; | |
310 | ||
311 | /* add this conntrack to the (per cpu) dying list */ | |
312 | ct->cpu = smp_processor_id(); | |
313 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
314 | ||
315 | spin_lock(&pcpu->lock); | |
316 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
317 | &pcpu->dying); | |
318 | spin_unlock(&pcpu->lock); | |
319 | } | |
320 | ||
321 | /* must be called with local_bh_disable */ | |
322 | static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct) | |
323 | { | |
324 | struct ct_pcpu *pcpu; | |
325 | ||
326 | /* add this conntrack to the (per cpu) unconfirmed list */ | |
327 | ct->cpu = smp_processor_id(); | |
328 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
329 | ||
330 | spin_lock(&pcpu->lock); | |
331 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
332 | &pcpu->unconfirmed); | |
333 | spin_unlock(&pcpu->lock); | |
334 | } | |
335 | ||
336 | /* must be called with local_bh_disable */ | |
337 | static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct) | |
338 | { | |
339 | struct ct_pcpu *pcpu; | |
340 | ||
341 | /* We overload first tuple to link into unconfirmed or dying list.*/ | |
342 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
343 | ||
344 | spin_lock(&pcpu->lock); | |
345 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); | |
346 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
347 | spin_unlock(&pcpu->lock); | |
348 | } | |
349 | ||
30322309 FW |
350 | #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK) |
351 | ||
0838aa7f | 352 | /* Released via destroy_conntrack() */ |
308ac914 DB |
353 | struct nf_conn *nf_ct_tmpl_alloc(struct net *net, |
354 | const struct nf_conntrack_zone *zone, | |
355 | gfp_t flags) | |
0838aa7f | 356 | { |
30322309 | 357 | struct nf_conn *tmpl, *p; |
0838aa7f | 358 | |
30322309 FW |
359 | if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) { |
360 | tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags); | |
361 | if (!tmpl) | |
362 | return NULL; | |
363 | ||
364 | p = tmpl; | |
365 | tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p); | |
366 | if (tmpl != p) { | |
367 | tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p); | |
368 | tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p; | |
369 | } | |
370 | } else { | |
371 | tmpl = kzalloc(sizeof(*tmpl), flags); | |
372 | if (!tmpl) | |
373 | return NULL; | |
374 | } | |
0838aa7f PNA |
375 | |
376 | tmpl->status = IPS_TEMPLATE; | |
377 | write_pnet(&tmpl->ct_net, net); | |
6c8dee98 | 378 | nf_ct_zone_add(tmpl, zone); |
0838aa7f PNA |
379 | atomic_set(&tmpl->ct_general.use, 0); |
380 | ||
381 | return tmpl; | |
0838aa7f PNA |
382 | } |
383 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc); | |
384 | ||
9cf94eab | 385 | void nf_ct_tmpl_free(struct nf_conn *tmpl) |
0838aa7f PNA |
386 | { |
387 | nf_ct_ext_destroy(tmpl); | |
388 | nf_ct_ext_free(tmpl); | |
30322309 FW |
389 | |
390 | if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) | |
391 | kfree((char *)tmpl - tmpl->proto.tmpl_padto); | |
392 | else | |
393 | kfree(tmpl); | |
0838aa7f | 394 | } |
9cf94eab | 395 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_free); |
0838aa7f | 396 | |
9fb9cbb1 YK |
397 | static void |
398 | destroy_conntrack(struct nf_conntrack *nfct) | |
399 | { | |
400 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 401 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 402 | |
0d53778e | 403 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 | 404 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
9fb9cbb1 | 405 | |
0838aa7f PNA |
406 | if (unlikely(nf_ct_is_template(ct))) { |
407 | nf_ct_tmpl_free(ct); | |
408 | return; | |
409 | } | |
923f4902 | 410 | rcu_read_lock(); |
5e8fbe2a | 411 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
4b4ceb9d | 412 | if (l4proto->destroy) |
605dcad6 | 413 | l4proto->destroy(ct); |
9fb9cbb1 | 414 | |
982d9a9c | 415 | rcu_read_unlock(); |
9fb9cbb1 | 416 | |
ca7433df | 417 | local_bh_disable(); |
9fb9cbb1 YK |
418 | /* Expectations will have been removed in clean_from_lists, |
419 | * except TFTP can create an expectation on the first packet, | |
420 | * before connection is in the list, so we need to clean here, | |
ca7433df JDB |
421 | * too. |
422 | */ | |
c1d10adb | 423 | nf_ct_remove_expectations(ct); |
9fb9cbb1 | 424 | |
b7779d06 | 425 | nf_ct_del_from_dying_or_unconfirmed_list(ct); |
9fb9cbb1 | 426 | |
ca7433df | 427 | local_bh_enable(); |
9fb9cbb1 YK |
428 | |
429 | if (ct->master) | |
430 | nf_ct_put(ct->master); | |
431 | ||
0d53778e | 432 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
433 | nf_conntrack_free(ct); |
434 | } | |
435 | ||
02982c27 | 436 | static void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 437 | { |
0d55af87 | 438 | struct net *net = nf_ct_net(ct); |
93bb0ceb | 439 | unsigned int hash, reply_hash; |
93bb0ceb | 440 | unsigned int sequence; |
9fb9cbb1 | 441 | |
9858a3ae | 442 | nf_ct_helper_destroy(ct); |
93bb0ceb JDB |
443 | |
444 | local_bh_disable(); | |
445 | do { | |
a3efd812 | 446 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 447 | hash = hash_conntrack(net, |
93bb0ceb | 448 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 449 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
450 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
451 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
452 | ||
9fb9cbb1 | 453 | clean_from_lists(ct); |
93bb0ceb JDB |
454 | nf_conntrack_double_unlock(hash, reply_hash); |
455 | ||
b7779d06 | 456 | nf_ct_add_to_dying_list(ct); |
93bb0ceb | 457 | |
93bb0ceb | 458 | local_bh_enable(); |
dd7669a9 | 459 | } |
dd7669a9 | 460 | |
02982c27 | 461 | bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report) |
dd7669a9 | 462 | { |
a992ca2a PNA |
463 | struct nf_conn_tstamp *tstamp; |
464 | ||
f330a7fd FW |
465 | if (test_and_set_bit(IPS_DYING_BIT, &ct->status)) |
466 | return false; | |
467 | ||
a992ca2a PNA |
468 | tstamp = nf_conn_tstamp_find(ct); |
469 | if (tstamp && tstamp->stop == 0) | |
d2de875c | 470 | tstamp->stop = ktime_get_real_ns(); |
dd7669a9 | 471 | |
9500507c FW |
472 | if (nf_conntrack_event_report(IPCT_DESTROY, ct, |
473 | portid, report) < 0) { | |
f330a7fd FW |
474 | /* destroy event was not delivered. nf_ct_put will |
475 | * be done by event cache worker on redelivery. | |
476 | */ | |
dd7669a9 | 477 | nf_ct_delete_from_lists(ct); |
9500507c | 478 | nf_conntrack_ecache_delayed_work(nf_ct_net(ct)); |
02982c27 | 479 | return false; |
dd7669a9 | 480 | } |
9500507c FW |
481 | |
482 | nf_conntrack_ecache_work(nf_ct_net(ct)); | |
dd7669a9 | 483 | nf_ct_delete_from_lists(ct); |
9fb9cbb1 | 484 | nf_ct_put(ct); |
02982c27 FW |
485 | return true; |
486 | } | |
487 | EXPORT_SYMBOL_GPL(nf_ct_delete); | |
488 | ||
c6825c09 AV |
489 | static inline bool |
490 | nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, | |
308ac914 | 491 | const struct nf_conntrack_tuple *tuple, |
e0c7d472 FW |
492 | const struct nf_conntrack_zone *zone, |
493 | const struct net *net) | |
c6825c09 AV |
494 | { |
495 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
496 | ||
497 | /* A conntrack can be recreated with the equal tuple, | |
498 | * so we need to check that the conntrack is confirmed | |
499 | */ | |
500 | return nf_ct_tuple_equal(tuple, &h->tuple) && | |
deedb590 | 501 | nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) && |
e0c7d472 FW |
502 | nf_ct_is_confirmed(ct) && |
503 | net_eq(net, nf_ct_net(ct)); | |
c6825c09 AV |
504 | } |
505 | ||
f330a7fd FW |
506 | /* caller must hold rcu readlock and none of the nf_conntrack_locks */ |
507 | static void nf_ct_gc_expired(struct nf_conn *ct) | |
508 | { | |
509 | if (!atomic_inc_not_zero(&ct->ct_general.use)) | |
510 | return; | |
511 | ||
512 | if (nf_ct_should_gc(ct)) | |
513 | nf_ct_kill(ct); | |
514 | ||
515 | nf_ct_put(ct); | |
516 | } | |
517 | ||
ea781f19 ED |
518 | /* |
519 | * Warning : | |
520 | * - Caller must take a reference on returned object | |
521 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
ea781f19 | 522 | */ |
99f07e91 | 523 | static struct nf_conntrack_tuple_hash * |
308ac914 | 524 | ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 525 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
526 | { |
527 | struct nf_conntrack_tuple_hash *h; | |
5e3c61f9 | 528 | struct hlist_nulls_head *ct_hash; |
ea781f19 | 529 | struct hlist_nulls_node *n; |
92e47ba8 | 530 | unsigned int bucket, hsize; |
9fb9cbb1 | 531 | |
ea781f19 | 532 | begin: |
92e47ba8 LZ |
533 | nf_conntrack_get_ht(&ct_hash, &hsize); |
534 | bucket = reciprocal_scale(hash, hsize); | |
5e3c61f9 FW |
535 | |
536 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) { | |
f330a7fd FW |
537 | struct nf_conn *ct; |
538 | ||
539 | ct = nf_ct_tuplehash_to_ctrack(h); | |
540 | if (nf_ct_is_expired(ct)) { | |
541 | nf_ct_gc_expired(ct); | |
542 | continue; | |
543 | } | |
544 | ||
545 | if (nf_ct_is_dying(ct)) | |
546 | continue; | |
547 | ||
8e8118f8 | 548 | if (nf_ct_key_equal(h, tuple, zone, net)) |
9fb9cbb1 | 549 | return h; |
9fb9cbb1 | 550 | } |
ea781f19 ED |
551 | /* |
552 | * if the nulls value we got at the end of this lookup is | |
553 | * not the expected one, we must restart lookup. | |
554 | * We probably met an item that was moved to another chain. | |
555 | */ | |
99f07e91 | 556 | if (get_nulls_value(n) != bucket) { |
2cf12348 | 557 | NF_CT_STAT_INC_ATOMIC(net, search_restart); |
ea781f19 | 558 | goto begin; |
af740b2c | 559 | } |
9fb9cbb1 YK |
560 | |
561 | return NULL; | |
562 | } | |
99f07e91 | 563 | |
9fb9cbb1 | 564 | /* Find a connection corresponding to a tuple. */ |
99f07e91 | 565 | static struct nf_conntrack_tuple_hash * |
308ac914 | 566 | __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 567 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
568 | { |
569 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 570 | struct nf_conn *ct; |
9fb9cbb1 | 571 | |
76507f69 | 572 | rcu_read_lock(); |
ea781f19 | 573 | begin: |
99f07e91 | 574 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
575 | if (h) { |
576 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
577 | if (unlikely(nf_ct_is_dying(ct) || |
578 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 579 | h = NULL; |
ea781f19 | 580 | else { |
e0c7d472 | 581 | if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) { |
ea781f19 ED |
582 | nf_ct_put(ct); |
583 | goto begin; | |
584 | } | |
585 | } | |
76507f69 PM |
586 | } |
587 | rcu_read_unlock(); | |
9fb9cbb1 YK |
588 | |
589 | return h; | |
590 | } | |
99f07e91 CG |
591 | |
592 | struct nf_conntrack_tuple_hash * | |
308ac914 | 593 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 CG |
594 | const struct nf_conntrack_tuple *tuple) |
595 | { | |
596 | return __nf_conntrack_find_get(net, zone, tuple, | |
1b8c8a9f | 597 | hash_conntrack_raw(tuple, net)); |
99f07e91 | 598 | } |
13b18339 | 599 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 600 | |
c1d10adb PNA |
601 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
602 | unsigned int hash, | |
b476b72a | 603 | unsigned int reply_hash) |
c1d10adb | 604 | { |
ea781f19 | 605 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
56d52d48 | 606 | &nf_conntrack_hash[hash]); |
ea781f19 | 607 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
56d52d48 | 608 | &nf_conntrack_hash[reply_hash]); |
c1d10adb PNA |
609 | } |
610 | ||
7d367e06 JK |
611 | int |
612 | nf_conntrack_hash_check_insert(struct nf_conn *ct) | |
c1d10adb | 613 | { |
308ac914 | 614 | const struct nf_conntrack_zone *zone; |
d696c7bd | 615 | struct net *net = nf_ct_net(ct); |
b476b72a | 616 | unsigned int hash, reply_hash; |
7d367e06 JK |
617 | struct nf_conntrack_tuple_hash *h; |
618 | struct hlist_nulls_node *n; | |
93bb0ceb | 619 | unsigned int sequence; |
c1d10adb | 620 | |
5d0aa2cc | 621 | zone = nf_ct_zone(ct); |
7d367e06 | 622 | |
93bb0ceb JDB |
623 | local_bh_disable(); |
624 | do { | |
a3efd812 | 625 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 626 | hash = hash_conntrack(net, |
93bb0ceb | 627 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 628 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
629 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
630 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
7d367e06 JK |
631 | |
632 | /* See if there's one in the list already, including reverse */ | |
56d52d48 | 633 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 634 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 635 | zone, net)) |
7d367e06 | 636 | goto out; |
86804348 | 637 | |
56d52d48 | 638 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 639 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 640 | zone, net)) |
7d367e06 | 641 | goto out; |
c1d10adb | 642 | |
e53376be PNA |
643 | smp_wmb(); |
644 | /* The caller holds a reference to this object */ | |
645 | atomic_set(&ct->ct_general.use, 2); | |
b476b72a | 646 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 647 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 648 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 649 | local_bh_enable(); |
7d367e06 JK |
650 | return 0; |
651 | ||
652 | out: | |
93bb0ceb | 653 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 654 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 655 | local_bh_enable(); |
7d367e06 | 656 | return -EEXIST; |
c1d10adb | 657 | } |
7d367e06 | 658 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); |
c1d10adb | 659 | |
ba76738c PNA |
660 | static inline void nf_ct_acct_update(struct nf_conn *ct, |
661 | enum ip_conntrack_info ctinfo, | |
662 | unsigned int len) | |
663 | { | |
664 | struct nf_conn_acct *acct; | |
665 | ||
666 | acct = nf_conn_acct_find(ct); | |
667 | if (acct) { | |
668 | struct nf_conn_counter *counter = acct->counter; | |
669 | ||
670 | atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); | |
671 | atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes); | |
672 | } | |
673 | } | |
674 | ||
71d8c47f PNA |
675 | static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo, |
676 | const struct nf_conn *loser_ct) | |
677 | { | |
678 | struct nf_conn_acct *acct; | |
679 | ||
680 | acct = nf_conn_acct_find(loser_ct); | |
681 | if (acct) { | |
682 | struct nf_conn_counter *counter = acct->counter; | |
71d8c47f PNA |
683 | unsigned int bytes; |
684 | ||
685 | /* u32 should be fine since we must have seen one packet. */ | |
686 | bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes); | |
687 | nf_ct_acct_update(ct, ctinfo, bytes); | |
688 | } | |
689 | } | |
690 | ||
691 | /* Resolve race on insertion if this protocol allows this. */ | |
692 | static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb, | |
693 | enum ip_conntrack_info ctinfo, | |
694 | struct nf_conntrack_tuple_hash *h) | |
695 | { | |
696 | /* This is the conntrack entry already in hashes that won race. */ | |
697 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
698 | struct nf_conntrack_l4proto *l4proto; | |
699 | ||
700 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); | |
701 | if (l4proto->allow_clash && | |
6e699867 | 702 | ((ct->status & IPS_NAT_DONE_MASK) == 0) && |
71d8c47f PNA |
703 | !nf_ct_is_dying(ct) && |
704 | atomic_inc_not_zero(&ct->ct_general.use)) { | |
97a6ad13 FW |
705 | enum ip_conntrack_info oldinfo; |
706 | struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo); | |
707 | ||
708 | nf_ct_acct_merge(ct, ctinfo, loser_ct); | |
709 | nf_conntrack_put(&loser_ct->ct_general); | |
c74454fa | 710 | nf_ct_set(skb, ct, oldinfo); |
71d8c47f PNA |
711 | return NF_ACCEPT; |
712 | } | |
713 | NF_CT_STAT_INC(net, drop); | |
714 | return NF_DROP; | |
715 | } | |
716 | ||
9fb9cbb1 YK |
717 | /* Confirm a connection given skb; places it in hash table */ |
718 | int | |
3db05fea | 719 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 | 720 | { |
308ac914 | 721 | const struct nf_conntrack_zone *zone; |
b476b72a | 722 | unsigned int hash, reply_hash; |
df0933dc | 723 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 724 | struct nf_conn *ct; |
df0933dc | 725 | struct nf_conn_help *help; |
a992ca2a | 726 | struct nf_conn_tstamp *tstamp; |
ea781f19 | 727 | struct hlist_nulls_node *n; |
9fb9cbb1 | 728 | enum ip_conntrack_info ctinfo; |
400dad39 | 729 | struct net *net; |
93bb0ceb | 730 | unsigned int sequence; |
71d8c47f | 731 | int ret = NF_DROP; |
9fb9cbb1 | 732 | |
3db05fea | 733 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 734 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
735 | |
736 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
737 | ICMP/TCP RST packets in other direction. Actual packet | |
738 | which created connection will be IP_CT_NEW or for an | |
739 | expected connection, IP_CT_RELATED. */ | |
740 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
741 | return NF_ACCEPT; | |
742 | ||
5d0aa2cc | 743 | zone = nf_ct_zone(ct); |
93bb0ceb JDB |
744 | local_bh_disable(); |
745 | ||
746 | do { | |
a3efd812 | 747 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb JDB |
748 | /* reuse the hash saved before */ |
749 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
56d52d48 | 750 | hash = scale_hash(hash); |
deedb590 | 751 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
752 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
753 | ||
754 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
9fb9cbb1 YK |
755 | |
756 | /* We're not in hash table, and we refuse to set up related | |
93bb0ceb JDB |
757 | * connections for unconfirmed conns. But packet copies and |
758 | * REJECT will give spurious warnings here. | |
759 | */ | |
9fb9cbb1 YK |
760 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ |
761 | ||
25985edc | 762 | /* No external references means no one else could have |
93bb0ceb JDB |
763 | * confirmed us. |
764 | */ | |
9fb9cbb1 | 765 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); |
0d53778e | 766 | pr_debug("Confirming conntrack %p\n", ct); |
8ca3f5e9 PNA |
767 | /* We have to check the DYING flag after unlink to prevent |
768 | * a race against nf_ct_get_next_corpse() possibly called from | |
769 | * user context, else we insert an already 'dead' hash, blocking | |
770 | * further use of that particular connection -JM. | |
771 | */ | |
772 | nf_ct_del_from_dying_or_unconfirmed_list(ct); | |
773 | ||
71d8c47f PNA |
774 | if (unlikely(nf_ct_is_dying(ct))) { |
775 | nf_ct_add_to_dying_list(ct); | |
776 | goto dying; | |
777 | } | |
fc350777 | 778 | |
9fb9cbb1 YK |
779 | /* See if there's one in the list already, including reverse: |
780 | NAT could have grabbed it without realizing, since we're | |
781 | not in the hash. If there is, we lost race. */ | |
56d52d48 | 782 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 783 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 784 | zone, net)) |
df0933dc | 785 | goto out; |
86804348 | 786 | |
56d52d48 | 787 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 788 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 789 | zone, net)) |
df0933dc | 790 | goto out; |
9fb9cbb1 | 791 | |
df0933dc PM |
792 | /* Timer relative to confirmation time, not original |
793 | setting time, otherwise we'd get timer wrap in | |
794 | weird delay cases. */ | |
f330a7fd | 795 | ct->timeout += nfct_time_stamp; |
df0933dc | 796 | atomic_inc(&ct->ct_general.use); |
45eec341 | 797 | ct->status |= IPS_CONFIRMED; |
5c8ec910 | 798 | |
a992ca2a PNA |
799 | /* set conntrack timestamp, if enabled. */ |
800 | tstamp = nf_conn_tstamp_find(ct); | |
801 | if (tstamp) { | |
2456e855 | 802 | if (skb->tstamp == 0) |
e3192690 | 803 | __net_timestamp(skb); |
a992ca2a PNA |
804 | |
805 | tstamp->start = ktime_to_ns(skb->tstamp); | |
806 | } | |
5c8ec910 PM |
807 | /* Since the lookup is lockless, hash insertion must be done after |
808 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
809 | * guarantee that no other CPU can find the conntrack before the above | |
810 | * stores are visible. | |
811 | */ | |
b476b72a | 812 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 813 | nf_conntrack_double_unlock(hash, reply_hash); |
93bb0ceb | 814 | local_bh_enable(); |
5c8ec910 | 815 | |
df0933dc PM |
816 | help = nfct_help(ct); |
817 | if (help && help->helper) | |
a71996fc | 818 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 819 | |
df0933dc | 820 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 821 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 822 | return NF_ACCEPT; |
9fb9cbb1 | 823 | |
df0933dc | 824 | out: |
8ca3f5e9 | 825 | nf_ct_add_to_dying_list(ct); |
71d8c47f PNA |
826 | ret = nf_ct_resolve_clash(net, skb, ctinfo, h); |
827 | dying: | |
93bb0ceb | 828 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 829 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 830 | local_bh_enable(); |
71d8c47f | 831 | return ret; |
9fb9cbb1 | 832 | } |
13b18339 | 833 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
834 | |
835 | /* Returns true if a connection correspondings to the tuple (required | |
836 | for NAT). */ | |
837 | int | |
838 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
839 | const struct nf_conn *ignored_conntrack) | |
840 | { | |
400dad39 | 841 | struct net *net = nf_ct_net(ignored_conntrack); |
308ac914 | 842 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 | 843 | struct nf_conntrack_tuple_hash *h; |
5e3c61f9 | 844 | struct hlist_nulls_head *ct_hash; |
92e47ba8 | 845 | unsigned int hash, hsize; |
ea781f19 | 846 | struct hlist_nulls_node *n; |
5d0aa2cc | 847 | struct nf_conn *ct; |
308ac914 DB |
848 | |
849 | zone = nf_ct_zone(ignored_conntrack); | |
9fb9cbb1 | 850 | |
2cf12348 | 851 | rcu_read_lock(); |
95a8d19f | 852 | begin: |
92e47ba8 LZ |
853 | nf_conntrack_get_ht(&ct_hash, &hsize); |
854 | hash = __hash_conntrack(net, tuple, hsize); | |
5e3c61f9 FW |
855 | |
856 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) { | |
5d0aa2cc | 857 | ct = nf_ct_tuplehash_to_ctrack(h); |
f330a7fd FW |
858 | |
859 | if (ct == ignored_conntrack) | |
860 | continue; | |
861 | ||
862 | if (nf_ct_is_expired(ct)) { | |
863 | nf_ct_gc_expired(ct); | |
864 | continue; | |
865 | } | |
866 | ||
867 | if (nf_ct_key_equal(h, tuple, zone, net)) { | |
2cf12348 FW |
868 | NF_CT_STAT_INC_ATOMIC(net, found); |
869 | rcu_read_unlock(); | |
ba419aff PM |
870 | return 1; |
871 | } | |
ba419aff | 872 | } |
95a8d19f FW |
873 | |
874 | if (get_nulls_value(n) != hash) { | |
875 | NF_CT_STAT_INC_ATOMIC(net, search_restart); | |
876 | goto begin; | |
877 | } | |
878 | ||
2cf12348 | 879 | rcu_read_unlock(); |
9fb9cbb1 | 880 | |
ba419aff | 881 | return 0; |
9fb9cbb1 | 882 | } |
13b18339 | 883 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 884 | |
7ae7730f PM |
885 | #define NF_CT_EVICTION_RANGE 8 |
886 | ||
9fb9cbb1 YK |
887 | /* There's a small race here where we may free a just-assured |
888 | connection. Too bad: we're in trouble anyway. */ | |
242922a0 FW |
889 | static unsigned int early_drop_list(struct net *net, |
890 | struct hlist_nulls_head *head) | |
9fb9cbb1 | 891 | { |
9fb9cbb1 | 892 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 893 | struct hlist_nulls_node *n; |
242922a0 FW |
894 | unsigned int drops = 0; |
895 | struct nf_conn *tmp; | |
3e86638e | 896 | |
242922a0 FW |
897 | hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) { |
898 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
9fb9cbb1 | 899 | |
f330a7fd FW |
900 | if (nf_ct_is_expired(tmp)) { |
901 | nf_ct_gc_expired(tmp); | |
902 | continue; | |
903 | } | |
904 | ||
242922a0 FW |
905 | if (test_bit(IPS_ASSURED_BIT, &tmp->status) || |
906 | !net_eq(nf_ct_net(tmp), net) || | |
907 | nf_ct_is_dying(tmp)) | |
908 | continue; | |
76507f69 | 909 | |
242922a0 FW |
910 | if (!atomic_inc_not_zero(&tmp->ct_general.use)) |
911 | continue; | |
76507f69 | 912 | |
242922a0 | 913 | /* kill only if still in same netns -- might have moved due to |
5f0d5a3a | 914 | * SLAB_TYPESAFE_BY_RCU rules. |
242922a0 FW |
915 | * |
916 | * We steal the timer reference. If that fails timer has | |
917 | * already fired or someone else deleted it. Just drop ref | |
918 | * and move to next entry. | |
919 | */ | |
920 | if (net_eq(nf_ct_net(tmp), net) && | |
921 | nf_ct_is_confirmed(tmp) && | |
242922a0 FW |
922 | nf_ct_delete(tmp, 0, 0)) |
923 | drops++; | |
924 | ||
925 | nf_ct_put(tmp); | |
9fb9cbb1 | 926 | } |
3e86638e | 927 | |
242922a0 FW |
928 | return drops; |
929 | } | |
9fb9cbb1 | 930 | |
242922a0 FW |
931 | static noinline int early_drop(struct net *net, unsigned int _hash) |
932 | { | |
933 | unsigned int i; | |
9fb9cbb1 | 934 | |
242922a0 FW |
935 | for (i = 0; i < NF_CT_EVICTION_RANGE; i++) { |
936 | struct hlist_nulls_head *ct_hash; | |
92e47ba8 | 937 | unsigned int hash, hsize, drops; |
242922a0 | 938 | |
3101e0fc | 939 | rcu_read_lock(); |
92e47ba8 LZ |
940 | nf_conntrack_get_ht(&ct_hash, &hsize); |
941 | hash = reciprocal_scale(_hash++, hsize); | |
242922a0 FW |
942 | |
943 | drops = early_drop_list(net, &ct_hash[hash]); | |
3101e0fc LZ |
944 | rcu_read_unlock(); |
945 | ||
242922a0 FW |
946 | if (drops) { |
947 | NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops); | |
948 | return true; | |
74138511 | 949 | } |
9fb9cbb1 | 950 | } |
3e86638e | 951 | |
242922a0 | 952 | return false; |
9fb9cbb1 YK |
953 | } |
954 | ||
c6dd940b FW |
955 | static bool gc_worker_skip_ct(const struct nf_conn *ct) |
956 | { | |
957 | return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct); | |
958 | } | |
959 | ||
960 | static bool gc_worker_can_early_drop(const struct nf_conn *ct) | |
961 | { | |
962 | const struct nf_conntrack_l4proto *l4proto; | |
963 | ||
964 | if (!test_bit(IPS_ASSURED_BIT, &ct->status)) | |
965 | return true; | |
966 | ||
967 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); | |
968 | if (l4proto->can_early_drop && l4proto->can_early_drop(ct)) | |
969 | return true; | |
970 | ||
971 | return false; | |
972 | } | |
973 | ||
b87a2f91 FW |
974 | static void gc_worker(struct work_struct *work) |
975 | { | |
e5072053 | 976 | unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u); |
b87a2f91 | 977 | unsigned int i, goal, buckets = 0, expired_count = 0; |
c6dd940b | 978 | unsigned int nf_conntrack_max95 = 0; |
b87a2f91 | 979 | struct conntrack_gc_work *gc_work; |
e0df8cae FW |
980 | unsigned int ratio, scanned = 0; |
981 | unsigned long next_run; | |
b87a2f91 FW |
982 | |
983 | gc_work = container_of(work, struct conntrack_gc_work, dwork.work); | |
984 | ||
e0df8cae | 985 | goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV; |
b87a2f91 | 986 | i = gc_work->last_bucket; |
c6dd940b FW |
987 | if (gc_work->early_drop) |
988 | nf_conntrack_max95 = nf_conntrack_max / 100u * 95u; | |
b87a2f91 FW |
989 | |
990 | do { | |
991 | struct nf_conntrack_tuple_hash *h; | |
992 | struct hlist_nulls_head *ct_hash; | |
993 | struct hlist_nulls_node *n; | |
994 | unsigned int hashsz; | |
995 | struct nf_conn *tmp; | |
996 | ||
997 | i++; | |
998 | rcu_read_lock(); | |
999 | ||
1000 | nf_conntrack_get_ht(&ct_hash, &hashsz); | |
1001 | if (i >= hashsz) | |
1002 | i = 0; | |
1003 | ||
1004 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) { | |
c6dd940b FW |
1005 | struct net *net; |
1006 | ||
b87a2f91 FW |
1007 | tmp = nf_ct_tuplehash_to_ctrack(h); |
1008 | ||
c023c0e4 | 1009 | scanned++; |
b87a2f91 FW |
1010 | if (nf_ct_is_expired(tmp)) { |
1011 | nf_ct_gc_expired(tmp); | |
1012 | expired_count++; | |
1013 | continue; | |
1014 | } | |
c6dd940b FW |
1015 | |
1016 | if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp)) | |
1017 | continue; | |
1018 | ||
1019 | net = nf_ct_net(tmp); | |
1020 | if (atomic_read(&net->ct.count) < nf_conntrack_max95) | |
1021 | continue; | |
1022 | ||
1023 | /* need to take reference to avoid possible races */ | |
1024 | if (!atomic_inc_not_zero(&tmp->ct_general.use)) | |
1025 | continue; | |
1026 | ||
1027 | if (gc_worker_skip_ct(tmp)) { | |
1028 | nf_ct_put(tmp); | |
1029 | continue; | |
1030 | } | |
1031 | ||
1032 | if (gc_worker_can_early_drop(tmp)) | |
1033 | nf_ct_kill(tmp); | |
1034 | ||
1035 | nf_ct_put(tmp); | |
b87a2f91 FW |
1036 | } |
1037 | ||
1038 | /* could check get_nulls_value() here and restart if ct | |
1039 | * was moved to another chain. But given gc is best-effort | |
1040 | * we will just continue with next hash slot. | |
1041 | */ | |
1042 | rcu_read_unlock(); | |
1043 | cond_resched_rcu_qs(); | |
524b698d | 1044 | } while (++buckets < goal); |
b87a2f91 FW |
1045 | |
1046 | if (gc_work->exiting) | |
1047 | return; | |
1048 | ||
e0df8cae FW |
1049 | /* |
1050 | * Eviction will normally happen from the packet path, and not | |
1051 | * from this gc worker. | |
1052 | * | |
1053 | * This worker is only here to reap expired entries when system went | |
1054 | * idle after a busy period. | |
1055 | * | |
1056 | * The heuristics below are supposed to balance conflicting goals: | |
1057 | * | |
1058 | * 1. Minimize time until we notice a stale entry | |
1059 | * 2. Maximize scan intervals to not waste cycles | |
1060 | * | |
e5072053 | 1061 | * Normally, expire ratio will be close to 0. |
e0df8cae | 1062 | * |
e5072053 FW |
1063 | * As soon as a sizeable fraction of the entries have expired |
1064 | * increase scan frequency. | |
e0df8cae | 1065 | */ |
c023c0e4 | 1066 | ratio = scanned ? expired_count * 100 / scanned : 0; |
e5072053 FW |
1067 | if (ratio > GC_EVICT_RATIO) { |
1068 | gc_work->next_gc_run = min_interval; | |
e0df8cae | 1069 | } else { |
e5072053 | 1070 | unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV; |
e0df8cae | 1071 | |
e5072053 FW |
1072 | BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0); |
1073 | ||
1074 | gc_work->next_gc_run += min_interval; | |
1075 | if (gc_work->next_gc_run > max) | |
1076 | gc_work->next_gc_run = max; | |
e0df8cae | 1077 | } |
c023c0e4 | 1078 | |
e5072053 | 1079 | next_run = gc_work->next_gc_run; |
b87a2f91 | 1080 | gc_work->last_bucket = i; |
c6dd940b | 1081 | gc_work->early_drop = false; |
e0df8cae | 1082 | queue_delayed_work(system_long_wq, &gc_work->dwork, next_run); |
b87a2f91 FW |
1083 | } |
1084 | ||
1085 | static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work) | |
1086 | { | |
1087 | INIT_DELAYED_WORK(&gc_work->dwork, gc_worker); | |
e5072053 | 1088 | gc_work->next_gc_run = HZ; |
b87a2f91 FW |
1089 | gc_work->exiting = false; |
1090 | } | |
1091 | ||
99f07e91 | 1092 | static struct nf_conn * |
308ac914 DB |
1093 | __nf_conntrack_alloc(struct net *net, |
1094 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1095 | const struct nf_conntrack_tuple *orig, |
1096 | const struct nf_conntrack_tuple *repl, | |
1097 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 1098 | { |
cd7fcbf1 | 1099 | struct nf_conn *ct; |
9fb9cbb1 | 1100 | |
5251e2d2 | 1101 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 1102 | atomic_inc(&net->ct.count); |
5251e2d2 | 1103 | |
76eb9460 | 1104 | if (nf_conntrack_max && |
49ac8713 | 1105 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
93bb0ceb | 1106 | if (!early_drop(net, hash)) { |
c6dd940b FW |
1107 | if (!conntrack_gc_work.early_drop) |
1108 | conntrack_gc_work.early_drop = true; | |
49ac8713 | 1109 | atomic_dec(&net->ct.count); |
e87cc472 | 1110 | net_warn_ratelimited("nf_conntrack: table full, dropping packet\n"); |
9fb9cbb1 YK |
1111 | return ERR_PTR(-ENOMEM); |
1112 | } | |
1113 | } | |
1114 | ||
941297f4 ED |
1115 | /* |
1116 | * Do not use kmem_cache_zalloc(), as this cache uses | |
5f0d5a3a | 1117 | * SLAB_TYPESAFE_BY_RCU. |
941297f4 | 1118 | */ |
0c5366b3 | 1119 | ct = kmem_cache_alloc(nf_conntrack_cachep, gfp); |
5e8018fc DB |
1120 | if (ct == NULL) |
1121 | goto out; | |
1122 | ||
440f0d58 | 1123 | spin_lock_init(&ct->lock); |
c88130bc | 1124 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 1125 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 1126 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
1127 | /* save hash for reusing when confirming */ |
1128 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
c41884ce | 1129 | ct->status = 0; |
c2d9ba9b | 1130 | write_pnet(&ct->ct_net, net); |
c41884ce FW |
1131 | memset(&ct->__nfct_init_offset[0], 0, |
1132 | offsetof(struct nf_conn, proto) - | |
1133 | offsetof(struct nf_conn, __nfct_init_offset[0])); | |
5e8018fc | 1134 | |
6c8dee98 | 1135 | nf_ct_zone_add(ct, zone); |
5e8018fc | 1136 | |
e53376be PNA |
1137 | /* Because we use RCU lookups, we set ct_general.use to zero before |
1138 | * this is inserted in any list. | |
941297f4 | 1139 | */ |
e53376be | 1140 | atomic_set(&ct->ct_general.use, 0); |
c88130bc | 1141 | return ct; |
5e8018fc DB |
1142 | out: |
1143 | atomic_dec(&net->ct.count); | |
5d0aa2cc | 1144 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 1145 | } |
99f07e91 | 1146 | |
308ac914 DB |
1147 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
1148 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1149 | const struct nf_conntrack_tuple *orig, |
1150 | const struct nf_conntrack_tuple *repl, | |
1151 | gfp_t gfp) | |
1152 | { | |
1153 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
1154 | } | |
13b18339 | 1155 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 1156 | |
c88130bc | 1157 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 1158 | { |
1d45209d ED |
1159 | struct net *net = nf_ct_net(ct); |
1160 | ||
e53376be | 1161 | /* A freed object has refcnt == 0, that's |
5f0d5a3a | 1162 | * the golden rule for SLAB_TYPESAFE_BY_RCU |
e53376be PNA |
1163 | */ |
1164 | NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0); | |
1165 | ||
ceeff754 | 1166 | nf_ct_ext_destroy(ct); |
ea781f19 | 1167 | nf_ct_ext_free(ct); |
0c5366b3 | 1168 | kmem_cache_free(nf_conntrack_cachep, ct); |
4e857c58 | 1169 | smp_mb__before_atomic(); |
0c3c6c00 | 1170 | atomic_dec(&net->ct.count); |
76507f69 | 1171 | } |
13b18339 | 1172 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 | 1173 | |
c539f017 | 1174 | |
9fb9cbb1 YK |
1175 | /* Allocate a new conntrack: we return -ENOMEM if classification |
1176 | failed due to stress. Otherwise it really is unclassifiable. */ | |
fc09e4a7 | 1177 | static noinline struct nf_conntrack_tuple_hash * |
b2a15a60 | 1178 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 1179 | const struct nf_conntrack_tuple *tuple, |
9fb9cbb1 | 1180 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 1181 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 1182 | struct sk_buff *skb, |
60b5f8f7 | 1183 | unsigned int dataoff, u32 hash) |
9fb9cbb1 | 1184 | { |
c88130bc | 1185 | struct nf_conn *ct; |
3c158f7f | 1186 | struct nf_conn_help *help; |
9fb9cbb1 | 1187 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 1188 | struct nf_conntrack_ecache *ecache; |
ca7433df | 1189 | struct nf_conntrack_expect *exp = NULL; |
308ac914 | 1190 | const struct nf_conntrack_zone *zone; |
60b5f8f7 | 1191 | struct nf_conn_timeout *timeout_ext; |
5e8018fc | 1192 | struct nf_conntrack_zone tmp; |
60b5f8f7 | 1193 | unsigned int *timeouts; |
9fb9cbb1 | 1194 | |
605dcad6 | 1195 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 1196 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
1197 | return NULL; |
1198 | } | |
1199 | ||
5e8018fc | 1200 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
99f07e91 CG |
1201 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
1202 | hash); | |
0a9ee813 | 1203 | if (IS_ERR(ct)) |
c88130bc | 1204 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 | 1205 | |
4440a2ab GF |
1206 | if (!nf_ct_add_synproxy(ct, tmpl)) { |
1207 | nf_conntrack_free(ct); | |
1208 | return ERR_PTR(-ENOMEM); | |
48b1de4c PM |
1209 | } |
1210 | ||
60b5f8f7 | 1211 | timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; |
ae2d708e PNA |
1212 | if (timeout_ext) { |
1213 | timeouts = nf_ct_timeout_data(timeout_ext); | |
1214 | if (unlikely(!timeouts)) | |
1215 | timeouts = l4proto->get_timeouts(net); | |
1216 | } else { | |
60b5f8f7 | 1217 | timeouts = l4proto->get_timeouts(net); |
ae2d708e | 1218 | } |
60b5f8f7 | 1219 | |
2c8503f5 | 1220 | if (!l4proto->new(ct, skb, dataoff, timeouts)) { |
c88130bc | 1221 | nf_conntrack_free(ct); |
ccd63c20 | 1222 | pr_debug("can't track with proto module\n"); |
9fb9cbb1 YK |
1223 | return NULL; |
1224 | } | |
1225 | ||
60b5f8f7 | 1226 | if (timeout_ext) |
ae2d708e PNA |
1227 | nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout), |
1228 | GFP_ATOMIC); | |
60b5f8f7 | 1229 | |
58401572 | 1230 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a992ca2a | 1231 | nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); |
c539f017 | 1232 | nf_ct_labels_ext_add(ct); |
b2a15a60 PM |
1233 | |
1234 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
1235 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
1236 | ecache ? ecache->expmask : 0, | |
1237 | GFP_ATOMIC); | |
58401572 | 1238 | |
ca7433df JDB |
1239 | local_bh_disable(); |
1240 | if (net->ct.expect_count) { | |
1241 | spin_lock(&nf_conntrack_expect_lock); | |
1242 | exp = nf_ct_find_expectation(net, zone, tuple); | |
1243 | if (exp) { | |
ccd63c20 | 1244 | pr_debug("expectation arrives ct=%p exp=%p\n", |
ca7433df JDB |
1245 | ct, exp); |
1246 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
1247 | __set_bit(IPS_EXPECTED_BIT, &ct->status); | |
1248 | /* exp->master safe, refcnt bumped in nf_ct_find_expectation */ | |
1249 | ct->master = exp->master; | |
1250 | if (exp->helper) { | |
1251 | help = nf_ct_helper_ext_add(ct, exp->helper, | |
1252 | GFP_ATOMIC); | |
1253 | if (help) | |
1254 | rcu_assign_pointer(help->helper, exp->helper); | |
1255 | } | |
ceceae1b | 1256 | |
9fb9cbb1 | 1257 | #ifdef CONFIG_NF_CONNTRACK_MARK |
ca7433df | 1258 | ct->mark = exp->master->mark; |
7c9728c3 JM |
1259 | #endif |
1260 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
ca7433df | 1261 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 1262 | #endif |
ca7433df JDB |
1263 | NF_CT_STAT_INC(net, expect_new); |
1264 | } | |
1265 | spin_unlock(&nf_conntrack_expect_lock); | |
1266 | } | |
8e8118f8 | 1267 | if (!exp) |
b2a15a60 | 1268 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
9fb9cbb1 | 1269 | |
e53376be PNA |
1270 | /* Now it is inserted into the unconfirmed list, bump refcount */ |
1271 | nf_conntrack_get(&ct->ct_general); | |
b7779d06 | 1272 | nf_ct_add_to_unconfirmed_list(ct); |
9fb9cbb1 | 1273 | |
ca7433df | 1274 | local_bh_enable(); |
9fb9cbb1 YK |
1275 | |
1276 | if (exp) { | |
1277 | if (exp->expectfn) | |
c88130bc | 1278 | exp->expectfn(ct, exp); |
6823645d | 1279 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
1280 | } |
1281 | ||
c88130bc | 1282 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
1283 | } |
1284 | ||
fc09e4a7 FW |
1285 | /* On success, returns 0, sets skb->_nfct | ctinfo */ |
1286 | static int | |
b2a15a60 | 1287 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 1288 | struct sk_buff *skb, |
9fb9cbb1 YK |
1289 | unsigned int dataoff, |
1290 | u_int16_t l3num, | |
1291 | u_int8_t protonum, | |
1292 | struct nf_conntrack_l3proto *l3proto, | |
fc09e4a7 | 1293 | struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 1294 | { |
308ac914 | 1295 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 YK |
1296 | struct nf_conntrack_tuple tuple; |
1297 | struct nf_conntrack_tuple_hash *h; | |
fc09e4a7 | 1298 | enum ip_conntrack_info ctinfo; |
5e8018fc | 1299 | struct nf_conntrack_zone tmp; |
9fb9cbb1 | 1300 | struct nf_conn *ct; |
99f07e91 | 1301 | u32 hash; |
9fb9cbb1 | 1302 | |
bbe735e4 | 1303 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
a31f1adc | 1304 | dataoff, l3num, protonum, net, &tuple, l3proto, |
605dcad6 | 1305 | l4proto)) { |
ccd63c20 | 1306 | pr_debug("Can't get tuple\n"); |
fc09e4a7 | 1307 | return 0; |
9fb9cbb1 YK |
1308 | } |
1309 | ||
1310 | /* look for tuple match */ | |
5e8018fc | 1311 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
1b8c8a9f | 1312 | hash = hash_conntrack_raw(&tuple, net); |
99f07e91 | 1313 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); |
9fb9cbb1 | 1314 | if (!h) { |
b2a15a60 | 1315 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
60b5f8f7 | 1316 | skb, dataoff, hash); |
9fb9cbb1 | 1317 | if (!h) |
fc09e4a7 | 1318 | return 0; |
9fb9cbb1 | 1319 | if (IS_ERR(h)) |
fc09e4a7 | 1320 | return PTR_ERR(h); |
9fb9cbb1 YK |
1321 | } |
1322 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1323 | ||
1324 | /* It exists; we have (non-exclusive) reference. */ | |
1325 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
fc09e4a7 | 1326 | ctinfo = IP_CT_ESTABLISHED_REPLY; |
9fb9cbb1 YK |
1327 | } else { |
1328 | /* Once we've had two way comms, always ESTABLISHED. */ | |
1329 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
ccd63c20 | 1330 | pr_debug("normal packet for %p\n", ct); |
fc09e4a7 | 1331 | ctinfo = IP_CT_ESTABLISHED; |
9fb9cbb1 | 1332 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { |
ccd63c20 | 1333 | pr_debug("related packet for %p\n", ct); |
fc09e4a7 | 1334 | ctinfo = IP_CT_RELATED; |
9fb9cbb1 | 1335 | } else { |
ccd63c20 | 1336 | pr_debug("new packet for %p\n", ct); |
fc09e4a7 | 1337 | ctinfo = IP_CT_NEW; |
9fb9cbb1 | 1338 | } |
9fb9cbb1 | 1339 | } |
fc09e4a7 FW |
1340 | nf_ct_set(skb, ct, ctinfo); |
1341 | return 0; | |
9fb9cbb1 YK |
1342 | } |
1343 | ||
1344 | unsigned int | |
a702a65f AD |
1345 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
1346 | struct sk_buff *skb) | |
9fb9cbb1 | 1347 | { |
97a6ad13 | 1348 | struct nf_conn *ct, *tmpl; |
9fb9cbb1 YK |
1349 | enum ip_conntrack_info ctinfo; |
1350 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 1351 | struct nf_conntrack_l4proto *l4proto; |
2c8503f5 | 1352 | unsigned int *timeouts; |
9fb9cbb1 YK |
1353 | unsigned int dataoff; |
1354 | u_int8_t protonum; | |
9fb9cbb1 YK |
1355 | int ret; |
1356 | ||
97a6ad13 | 1357 | tmpl = nf_ct_get(skb, &ctinfo); |
cc41c84b | 1358 | if (tmpl || ctinfo == IP_CT_UNTRACKED) { |
b2a15a60 | 1359 | /* Previously seen (loopback or untracked)? Ignore. */ |
cc41c84b FW |
1360 | if ((tmpl && !nf_ct_is_template(tmpl)) || |
1361 | ctinfo == IP_CT_UNTRACKED) { | |
b2a15a60 PM |
1362 | NF_CT_STAT_INC_ATOMIC(net, ignore); |
1363 | return NF_ACCEPT; | |
1364 | } | |
a9e419dc | 1365 | skb->_nfct = 0; |
9fb9cbb1 YK |
1366 | } |
1367 | ||
e2361cb9 | 1368 | /* rcu_read_lock()ed by nf_hook_thresh */ |
76108cea | 1369 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 1370 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
1371 | &dataoff, &protonum); |
1372 | if (ret <= 0) { | |
25985edc | 1373 | pr_debug("not prepared to track yet or error occurred\n"); |
0d55af87 AD |
1374 | NF_CT_STAT_INC_ATOMIC(net, error); |
1375 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1376 | ret = -ret; |
1377 | goto out; | |
9fb9cbb1 YK |
1378 | } |
1379 | ||
76108cea | 1380 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
1381 | |
1382 | /* It may be an special packet, error, unclean... | |
1383 | * inverse of the return code tells to the netfilter | |
1384 | * core what to do with the packet. */ | |
74c51a14 | 1385 | if (l4proto->error != NULL) { |
11df4b76 | 1386 | ret = l4proto->error(net, tmpl, skb, dataoff, pf, hooknum); |
74c51a14 | 1387 | if (ret <= 0) { |
0d55af87 AD |
1388 | NF_CT_STAT_INC_ATOMIC(net, error); |
1389 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1390 | ret = -ret; |
1391 | goto out; | |
74c51a14 | 1392 | } |
88ed01d1 | 1393 | /* ICMP[v6] protocol trackers may assign one conntrack. */ |
a9e419dc | 1394 | if (skb->_nfct) |
88ed01d1 | 1395 | goto out; |
9fb9cbb1 | 1396 | } |
08733a0c | 1397 | repeat: |
fc09e4a7 FW |
1398 | ret = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
1399 | l3proto, l4proto); | |
1400 | if (ret < 0) { | |
9fb9cbb1 | 1401 | /* Too stressed to deal. */ |
0d55af87 | 1402 | NF_CT_STAT_INC_ATOMIC(net, drop); |
b2a15a60 PM |
1403 | ret = NF_DROP; |
1404 | goto out; | |
9fb9cbb1 YK |
1405 | } |
1406 | ||
fc09e4a7 FW |
1407 | ct = nf_ct_get(skb, &ctinfo); |
1408 | if (!ct) { | |
1409 | /* Not valid part of a connection */ | |
1410 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
1411 | ret = NF_ACCEPT; | |
1412 | goto out; | |
1413 | } | |
9fb9cbb1 | 1414 | |
60b5f8f7 | 1415 | /* Decide what timeout policy we want to apply to this flow. */ |
84b5ee93 | 1416 | timeouts = nf_ct_timeout_lookup(net, ct, l4proto); |
60b5f8f7 | 1417 | |
2c8503f5 | 1418 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); |
ec8d5409 | 1419 | if (ret <= 0) { |
9fb9cbb1 YK |
1420 | /* Invalid: inverse of the return code tells |
1421 | * the netfilter core what to do */ | |
0d53778e | 1422 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
97a6ad13 | 1423 | nf_conntrack_put(&ct->ct_general); |
a9e419dc | 1424 | skb->_nfct = 0; |
0d55af87 | 1425 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
1426 | if (ret == -NF_DROP) |
1427 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
56a62e22 AB |
1428 | /* Special case: TCP tracker reports an attempt to reopen a |
1429 | * closed/aborted connection. We have to go back and create a | |
1430 | * fresh conntrack. | |
1431 | */ | |
1432 | if (ret == -NF_REPEAT) | |
1433 | goto repeat; | |
b2a15a60 PM |
1434 | ret = -ret; |
1435 | goto out; | |
9fb9cbb1 YK |
1436 | } |
1437 | ||
fc09e4a7 FW |
1438 | if (ctinfo == IP_CT_ESTABLISHED_REPLY && |
1439 | !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 1440 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 | 1441 | out: |
56a62e22 AB |
1442 | if (tmpl) |
1443 | nf_ct_put(tmpl); | |
9fb9cbb1 YK |
1444 | |
1445 | return ret; | |
1446 | } | |
13b18339 | 1447 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 1448 | |
5f2b4c90 JE |
1449 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
1450 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 1451 | { |
5f2b4c90 | 1452 | bool ret; |
923f4902 PM |
1453 | |
1454 | rcu_read_lock(); | |
1455 | ret = nf_ct_invert_tuple(inverse, orig, | |
1456 | __nf_ct_l3proto_find(orig->src.l3num), | |
1457 | __nf_ct_l4proto_find(orig->src.l3num, | |
1458 | orig->dst.protonum)); | |
1459 | rcu_read_unlock(); | |
1460 | return ret; | |
9fb9cbb1 | 1461 | } |
13b18339 | 1462 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 1463 | |
5b1158e9 JK |
1464 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
1465 | implicitly racy: see __nf_conntrack_confirm */ | |
1466 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
1467 | const struct nf_conntrack_tuple *newreply) | |
1468 | { | |
1469 | struct nf_conn_help *help = nfct_help(ct); | |
1470 | ||
5b1158e9 JK |
1471 | /* Should be unconfirmed, so not in hash table yet */ |
1472 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
1473 | ||
0d53778e | 1474 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 1475 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
1476 | |
1477 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 1478 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 1479 | return; |
ceceae1b | 1480 | |
c52fbb41 | 1481 | rcu_read_lock(); |
b2a15a60 | 1482 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 1483 | rcu_read_unlock(); |
5b1158e9 | 1484 | } |
13b18339 | 1485 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 1486 | |
9fb9cbb1 YK |
1487 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
1488 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
1489 | enum ip_conntrack_info ctinfo, | |
1490 | const struct sk_buff *skb, | |
1491 | unsigned long extra_jiffies, | |
1492 | int do_acct) | |
1493 | { | |
9fb9cbb1 YK |
1494 | NF_CT_ASSERT(skb); |
1495 | ||
997ae831 | 1496 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1497 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1498 | goto acct; | |
997ae831 | 1499 | |
9fb9cbb1 | 1500 | /* If not in hash table, timer will not be active yet */ |
f330a7fd FW |
1501 | if (nf_ct_is_confirmed(ct)) |
1502 | extra_jiffies += nfct_time_stamp; | |
9fb9cbb1 | 1503 | |
f330a7fd | 1504 | ct->timeout = extra_jiffies; |
47d95045 | 1505 | acct: |
ba76738c PNA |
1506 | if (do_acct) |
1507 | nf_ct_acct_update(ct, ctinfo, skb->len); | |
9fb9cbb1 | 1508 | } |
13b18339 | 1509 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1510 | |
ad66713f FW |
1511 | bool nf_ct_kill_acct(struct nf_conn *ct, |
1512 | enum ip_conntrack_info ctinfo, | |
1513 | const struct sk_buff *skb) | |
51091764 | 1514 | { |
ad66713f | 1515 | nf_ct_acct_update(ct, ctinfo, skb->len); |
58401572 | 1516 | |
f330a7fd | 1517 | return nf_ct_delete(ct, 0, 0); |
51091764 | 1518 | } |
ad66713f | 1519 | EXPORT_SYMBOL_GPL(nf_ct_kill_acct); |
51091764 | 1520 | |
c0cd1156 | 1521 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
c1d10adb PNA |
1522 | |
1523 | #include <linux/netfilter/nfnetlink.h> | |
1524 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1525 | #include <linux/mutex.h> |
1526 | ||
c1d10adb PNA |
1527 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1528 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1529 | * or depend on ctnetlink */ | |
fdf70832 | 1530 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1531 | const struct nf_conntrack_tuple *tuple) |
1532 | { | |
bae65be8 DM |
1533 | if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) || |
1534 | nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port)) | |
1535 | goto nla_put_failure; | |
c1d10adb PNA |
1536 | return 0; |
1537 | ||
df6fb868 | 1538 | nla_put_failure: |
c1d10adb PNA |
1539 | return -1; |
1540 | } | |
fdf70832 | 1541 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1542 | |
f73e924c PM |
1543 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1544 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1545 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1546 | }; |
f73e924c | 1547 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1548 | |
fdf70832 | 1549 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1550 | struct nf_conntrack_tuple *t) |
1551 | { | |
df6fb868 | 1552 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1553 | return -EINVAL; |
1554 | ||
77236b6e PM |
1555 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1556 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1557 | |
1558 | return 0; | |
1559 | } | |
fdf70832 | 1560 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1561 | |
1562 | int nf_ct_port_nlattr_tuple_size(void) | |
1563 | { | |
1564 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1565 | } | |
1566 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1567 | #endif |
1568 | ||
9fb9cbb1 | 1569 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
312a0c16 | 1570 | static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb) |
9fb9cbb1 YK |
1571 | { |
1572 | struct nf_conn *ct; | |
1573 | enum ip_conntrack_info ctinfo; | |
1574 | ||
1575 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1576 | ct = nf_ct_get(skb, &ctinfo); | |
1577 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
fb048833 | 1578 | ctinfo = IP_CT_RELATED_REPLY; |
9fb9cbb1 YK |
1579 | else |
1580 | ctinfo = IP_CT_RELATED; | |
1581 | ||
1582 | /* Attach to new skbuff, and increment count */ | |
c74454fa | 1583 | nf_ct_set(nskb, ct, ctinfo); |
cb9c6836 | 1584 | nf_conntrack_get(skb_nfct(nskb)); |
9fb9cbb1 YK |
1585 | } |
1586 | ||
9fb9cbb1 | 1587 | /* Bring out ya dead! */ |
df0933dc | 1588 | static struct nf_conn * |
2843fb69 | 1589 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1590 | void *data, unsigned int *bucket) |
1591 | { | |
df0933dc PM |
1592 | struct nf_conntrack_tuple_hash *h; |
1593 | struct nf_conn *ct; | |
ea781f19 | 1594 | struct hlist_nulls_node *n; |
93bb0ceb | 1595 | spinlock_t *lockp; |
9fb9cbb1 | 1596 | |
56d52d48 | 1597 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
93bb0ceb JDB |
1598 | lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS]; |
1599 | local_bh_disable(); | |
b16c2919 | 1600 | nf_conntrack_lock(lockp); |
56d52d48 FW |
1601 | if (*bucket < nf_conntrack_htable_size) { |
1602 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) { | |
93bb0ceb JDB |
1603 | if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL) |
1604 | continue; | |
1605 | ct = nf_ct_tuplehash_to_ctrack(h); | |
2843fb69 | 1606 | if (iter(ct, data)) |
93bb0ceb JDB |
1607 | goto found; |
1608 | } | |
df0933dc | 1609 | } |
93bb0ceb JDB |
1610 | spin_unlock(lockp); |
1611 | local_bh_enable(); | |
d93c6258 | 1612 | cond_resched(); |
601e68e1 | 1613 | } |
b7779d06 | 1614 | |
b0feacaa FW |
1615 | return NULL; |
1616 | found: | |
1617 | atomic_inc(&ct->ct_general.use); | |
1618 | spin_unlock(lockp); | |
1619 | local_bh_enable(); | |
1620 | return ct; | |
1621 | } | |
1622 | ||
2843fb69 FW |
1623 | static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), |
1624 | void *data, u32 portid, int report) | |
1625 | { | |
1626 | struct nf_conn *ct; | |
1627 | unsigned int bucket = 0; | |
1628 | ||
1629 | might_sleep(); | |
1630 | ||
1631 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { | |
1632 | /* Time to push up daises... */ | |
1633 | ||
1634 | nf_ct_delete(ct, portid, report); | |
1635 | nf_ct_put(ct); | |
1636 | cond_resched(); | |
1637 | } | |
1638 | } | |
1639 | ||
1640 | struct iter_data { | |
1641 | int (*iter)(struct nf_conn *i, void *data); | |
1642 | void *data; | |
1643 | struct net *net; | |
1644 | }; | |
1645 | ||
1646 | static int iter_net_only(struct nf_conn *i, void *data) | |
1647 | { | |
1648 | struct iter_data *d = data; | |
1649 | ||
1650 | if (!net_eq(d->net, nf_ct_net(i))) | |
1651 | return 0; | |
1652 | ||
1653 | return d->iter(i, d->data); | |
1654 | } | |
1655 | ||
b0feacaa FW |
1656 | static void |
1657 | __nf_ct_unconfirmed_destroy(struct net *net) | |
1658 | { | |
1659 | int cpu; | |
1660 | ||
b7779d06 | 1661 | for_each_possible_cpu(cpu) { |
b0feacaa FW |
1662 | struct nf_conntrack_tuple_hash *h; |
1663 | struct hlist_nulls_node *n; | |
1664 | struct ct_pcpu *pcpu; | |
1665 | ||
1666 | pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
b7779d06 JDB |
1667 | |
1668 | spin_lock_bh(&pcpu->lock); | |
1669 | hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) { | |
b0feacaa FW |
1670 | struct nf_conn *ct; |
1671 | ||
b7779d06 | 1672 | ct = nf_ct_tuplehash_to_ctrack(h); |
b0feacaa FW |
1673 | |
1674 | /* we cannot call iter() on unconfirmed list, the | |
1675 | * owning cpu can reallocate ct->ext at any time. | |
1676 | */ | |
1677 | set_bit(IPS_DYING_BIT, &ct->status); | |
b7779d06 JDB |
1678 | } |
1679 | spin_unlock_bh(&pcpu->lock); | |
d93c6258 | 1680 | cond_resched(); |
b7779d06 | 1681 | } |
9fb9cbb1 YK |
1682 | } |
1683 | ||
9fd6452d FW |
1684 | void nf_ct_iterate_cleanup_net(struct net *net, |
1685 | int (*iter)(struct nf_conn *i, void *data), | |
1686 | void *data, u32 portid, int report) | |
9fb9cbb1 | 1687 | { |
2843fb69 | 1688 | struct iter_data d; |
9fb9cbb1 | 1689 | |
d93c6258 FW |
1690 | might_sleep(); |
1691 | ||
88b68bc5 FW |
1692 | if (atomic_read(&net->ct.count) == 0) |
1693 | return; | |
1694 | ||
b0feacaa FW |
1695 | __nf_ct_unconfirmed_destroy(net); |
1696 | ||
2843fb69 FW |
1697 | d.iter = iter; |
1698 | d.data = data; | |
1699 | d.net = net; | |
1700 | ||
b0feacaa FW |
1701 | synchronize_net(); |
1702 | ||
2843fb69 FW |
1703 | nf_ct_iterate_cleanup(iter_net_only, &d, portid, report); |
1704 | } | |
1705 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net); | |
9fb9cbb1 | 1706 | |
2843fb69 FW |
1707 | /** |
1708 | * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table | |
1709 | * @iter: callback to invoke for each conntrack | |
1710 | * @data: data to pass to @iter | |
1711 | * | |
1712 | * Like nf_ct_iterate_cleanup, but first marks conntracks on the | |
1713 | * unconfirmed list as dying (so they will not be inserted into | |
1714 | * main table). | |
1715 | */ | |
1716 | void | |
1717 | nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data) | |
1718 | { | |
1719 | struct net *net; | |
1720 | ||
1721 | rtnl_lock(); | |
1722 | for_each_net(net) { | |
1723 | if (atomic_read(&net->ct.count) == 0) | |
1724 | continue; | |
1725 | __nf_ct_unconfirmed_destroy(net); | |
9fb9cbb1 | 1726 | } |
2843fb69 FW |
1727 | rtnl_unlock(); |
1728 | ||
1729 | /* a conntrack could have been unlinked from unconfirmed list | |
1730 | * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy(). | |
1731 | * This makes sure its inserted into conntrack table. | |
1732 | */ | |
1733 | synchronize_net(); | |
1734 | ||
1735 | nf_ct_iterate_cleanup(iter, data, 0, 0); | |
9fb9cbb1 | 1736 | } |
2843fb69 | 1737 | EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy); |
9fb9cbb1 | 1738 | |
274d383b PNA |
1739 | static int kill_all(struct nf_conn *i, void *data) |
1740 | { | |
2843fb69 | 1741 | return net_eq(nf_ct_net(i), data); |
274d383b PNA |
1742 | } |
1743 | ||
d862a662 | 1744 | void nf_ct_free_hashtable(void *hash, unsigned int size) |
9fb9cbb1 | 1745 | { |
d862a662 | 1746 | if (is_vmalloc_addr(hash)) |
9fb9cbb1 YK |
1747 | vfree(hash); |
1748 | else | |
601e68e1 | 1749 | free_pages((unsigned long)hash, |
f205c5e0 | 1750 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1751 | } |
ac565e5f | 1752 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1753 | |
f94161c1 | 1754 | void nf_conntrack_cleanup_start(void) |
9fb9cbb1 | 1755 | { |
b87a2f91 | 1756 | conntrack_gc_work.exiting = true; |
f94161c1 G |
1757 | RCU_INIT_POINTER(ip_ct_attach, NULL); |
1758 | } | |
1759 | ||
1760 | void nf_conntrack_cleanup_end(void) | |
1761 | { | |
1762 | RCU_INIT_POINTER(nf_ct_destroy, NULL); | |
9edd7ca0 | 1763 | |
b87a2f91 | 1764 | cancel_delayed_work_sync(&conntrack_gc_work.dwork); |
56d52d48 FW |
1765 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
1766 | ||
04d87001 | 1767 | nf_conntrack_proto_fini(); |
41d73ec0 | 1768 | nf_conntrack_seqadj_fini(); |
5f69b8f5 | 1769 | nf_conntrack_labels_fini(); |
5e615b22 | 1770 | nf_conntrack_helper_fini(); |
8684094c | 1771 | nf_conntrack_timeout_fini(); |
3fe0f943 | 1772 | nf_conntrack_ecache_fini(); |
73f4001a | 1773 | nf_conntrack_tstamp_fini(); |
b7ff3a1f | 1774 | nf_conntrack_acct_fini(); |
83b4dbe1 | 1775 | nf_conntrack_expect_fini(); |
77571149 FW |
1776 | |
1777 | kmem_cache_destroy(nf_conntrack_cachep); | |
08f6547d | 1778 | } |
9fb9cbb1 | 1779 | |
f94161c1 G |
1780 | /* |
1781 | * Mishearing the voices in his head, our hero wonders how he's | |
1782 | * supposed to kill the mall. | |
1783 | */ | |
1784 | void nf_conntrack_cleanup_net(struct net *net) | |
08f6547d | 1785 | { |
dece40e8 VD |
1786 | LIST_HEAD(single); |
1787 | ||
1788 | list_add(&net->exit_list, &single); | |
1789 | nf_conntrack_cleanup_net_list(&single); | |
1790 | } | |
1791 | ||
1792 | void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list) | |
1793 | { | |
1794 | int busy; | |
1795 | struct net *net; | |
1796 | ||
f94161c1 G |
1797 | /* |
1798 | * This makes sure all current packets have passed through | |
1799 | * netfilter framework. Roll on, two-stage module | |
1800 | * delete... | |
1801 | */ | |
1802 | synchronize_net(); | |
dece40e8 VD |
1803 | i_see_dead_people: |
1804 | busy = 0; | |
1805 | list_for_each_entry(net, net_exit_list, exit_list) { | |
2843fb69 | 1806 | nf_ct_iterate_cleanup(kill_all, net, 0, 0); |
dece40e8 VD |
1807 | if (atomic_read(&net->ct.count) != 0) |
1808 | busy = 1; | |
1809 | } | |
1810 | if (busy) { | |
9fb9cbb1 YK |
1811 | schedule(); |
1812 | goto i_see_dead_people; | |
1813 | } | |
1814 | ||
dece40e8 | 1815 | list_for_each_entry(net, net_exit_list, exit_list) { |
dece40e8 VD |
1816 | nf_conntrack_proto_pernet_fini(net); |
1817 | nf_conntrack_helper_pernet_fini(net); | |
1818 | nf_conntrack_ecache_pernet_fini(net); | |
1819 | nf_conntrack_tstamp_pernet_fini(net); | |
1820 | nf_conntrack_acct_pernet_fini(net); | |
1821 | nf_conntrack_expect_pernet_fini(net); | |
dece40e8 | 1822 | free_percpu(net->ct.stat); |
b7779d06 | 1823 | free_percpu(net->ct.pcpu_lists); |
dece40e8 | 1824 | } |
08f6547d AD |
1825 | } |
1826 | ||
d862a662 | 1827 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) |
9fb9cbb1 | 1828 | { |
ea781f19 ED |
1829 | struct hlist_nulls_head *hash; |
1830 | unsigned int nr_slots, i; | |
1831 | size_t sz; | |
9fb9cbb1 | 1832 | |
9cc1c73a FW |
1833 | if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head))) |
1834 | return NULL; | |
1835 | ||
ea781f19 ED |
1836 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1837 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
9cc1c73a FW |
1838 | |
1839 | if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head))) | |
1840 | return NULL; | |
1841 | ||
ea781f19 ED |
1842 | sz = nr_slots * sizeof(struct hlist_nulls_head); |
1843 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1844 | get_order(sz)); | |
f0ad4621 | 1845 | if (!hash) |
966567b7 | 1846 | hash = vzalloc(sz); |
9fb9cbb1 | 1847 | |
ea781f19 ED |
1848 | if (hash && nulls) |
1849 | for (i = 0; i < nr_slots; i++) | |
1850 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1851 | |
1852 | return hash; | |
1853 | } | |
ac565e5f | 1854 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1855 | |
3183ab89 | 1856 | int nf_conntrack_hash_resize(unsigned int hashsize) |
9fb9cbb1 | 1857 | { |
3183ab89 FW |
1858 | int i, bucket; |
1859 | unsigned int old_size; | |
ea781f19 | 1860 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1861 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1862 | struct nf_conn *ct; |
9fb9cbb1 | 1863 | |
9fb9cbb1 YK |
1864 | if (!hashsize) |
1865 | return -EINVAL; | |
1866 | ||
d862a662 | 1867 | hash = nf_ct_alloc_hashtable(&hashsize, 1); |
9fb9cbb1 YK |
1868 | if (!hash) |
1869 | return -ENOMEM; | |
1870 | ||
3183ab89 FW |
1871 | old_size = nf_conntrack_htable_size; |
1872 | if (old_size == hashsize) { | |
1873 | nf_ct_free_hashtable(hash, hashsize); | |
1874 | return 0; | |
1875 | } | |
1876 | ||
93bb0ceb JDB |
1877 | local_bh_disable(); |
1878 | nf_conntrack_all_lock(); | |
a3efd812 | 1879 | write_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb | 1880 | |
76507f69 PM |
1881 | /* Lookups in the old hash might happen in parallel, which means we |
1882 | * might get false negatives during connection lookup. New connections | |
1883 | * created because of a false negative won't make it into the hash | |
93bb0ceb | 1884 | * though since that required taking the locks. |
76507f69 | 1885 | */ |
93bb0ceb | 1886 | |
56d52d48 FW |
1887 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
1888 | while (!hlist_nulls_empty(&nf_conntrack_hash[i])) { | |
1889 | h = hlist_nulls_entry(nf_conntrack_hash[i].first, | |
1890 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1891 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1892 | hlist_nulls_del_rcu(&h->hnnode); |
1b8c8a9f FW |
1893 | bucket = __hash_conntrack(nf_ct_net(ct), |
1894 | &h->tuple, hashsize); | |
ea781f19 | 1895 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1896 | } |
1897 | } | |
56d52d48 FW |
1898 | old_size = nf_conntrack_htable_size; |
1899 | old_hash = nf_conntrack_hash; | |
9fb9cbb1 | 1900 | |
56d52d48 FW |
1901 | nf_conntrack_hash = hash; |
1902 | nf_conntrack_htable_size = hashsize; | |
93bb0ceb | 1903 | |
a3efd812 | 1904 | write_seqcount_end(&nf_conntrack_generation); |
93bb0ceb JDB |
1905 | nf_conntrack_all_unlock(); |
1906 | local_bh_enable(); | |
9fb9cbb1 | 1907 | |
5e3c61f9 | 1908 | synchronize_net(); |
d862a662 | 1909 | nf_ct_free_hashtable(old_hash, old_size); |
9fb9cbb1 YK |
1910 | return 0; |
1911 | } | |
3183ab89 FW |
1912 | |
1913 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) | |
1914 | { | |
1915 | unsigned int hashsize; | |
1916 | int rc; | |
1917 | ||
1918 | if (current->nsproxy->net_ns != &init_net) | |
1919 | return -EOPNOTSUPP; | |
1920 | ||
1921 | /* On boot, we can set this without any fancy locking. */ | |
1922 | if (!nf_conntrack_htable_size) | |
1923 | return param_set_uint(val, kp); | |
1924 | ||
1925 | rc = kstrtouint(val, 0, &hashsize); | |
1926 | if (rc) | |
1927 | return rc; | |
1928 | ||
1929 | return nf_conntrack_hash_resize(hashsize); | |
1930 | } | |
fae718dd | 1931 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1932 | |
fae718dd | 1933 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1934 | &nf_conntrack_htable_size, 0600); |
1935 | ||
ab71632c | 1936 | static __always_inline unsigned int total_extension_size(void) |
b3a5db10 FW |
1937 | { |
1938 | /* remember to add new extensions below */ | |
1939 | BUILD_BUG_ON(NF_CT_EXT_NUM > 9); | |
1940 | ||
1941 | return sizeof(struct nf_ct_ext) + | |
1942 | sizeof(struct nf_conn_help) | |
1943 | #if IS_ENABLED(CONFIG_NF_NAT) | |
1944 | + sizeof(struct nf_conn_nat) | |
1945 | #endif | |
1946 | + sizeof(struct nf_conn_seqadj) | |
1947 | + sizeof(struct nf_conn_acct) | |
1948 | #ifdef CONFIG_NF_CONNTRACK_EVENTS | |
1949 | + sizeof(struct nf_conntrack_ecache) | |
1950 | #endif | |
1951 | #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP | |
1952 | + sizeof(struct nf_conn_tstamp) | |
1953 | #endif | |
1954 | #ifdef CONFIG_NF_CONNTRACK_TIMEOUT | |
1955 | + sizeof(struct nf_conn_timeout) | |
1956 | #endif | |
1957 | #ifdef CONFIG_NF_CONNTRACK_LABELS | |
1958 | + sizeof(struct nf_conn_labels) | |
1959 | #endif | |
1960 | #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY) | |
1961 | + sizeof(struct nf_conn_synproxy) | |
1962 | #endif | |
1963 | ; | |
1964 | }; | |
1965 | ||
f94161c1 | 1966 | int nf_conntrack_init_start(void) |
9fb9cbb1 | 1967 | { |
f205c5e0 | 1968 | int max_factor = 8; |
0c5366b3 | 1969 | int ret = -ENOMEM; |
cc41c84b | 1970 | int i; |
93bb0ceb | 1971 | |
b3a5db10 FW |
1972 | /* struct nf_ct_ext uses u8 to store offsets/size */ |
1973 | BUILD_BUG_ON(total_extension_size() > 255u); | |
1974 | ||
a3efd812 FW |
1975 | seqcount_init(&nf_conntrack_generation); |
1976 | ||
d5d20912 | 1977 | for (i = 0; i < CONNTRACK_LOCKS; i++) |
93bb0ceb | 1978 | spin_lock_init(&nf_conntrack_locks[i]); |
9fb9cbb1 | 1979 | |
9fb9cbb1 | 1980 | if (!nf_conntrack_htable_size) { |
88eab472 ML |
1981 | /* Idea from tcp.c: use 1/16384 of memory. |
1982 | * On i386: 32MB machine has 512 buckets. | |
1983 | * >= 1GB machines have 16384 buckets. | |
1984 | * >= 4GB machines have 65536 buckets. | |
1985 | */ | |
9fb9cbb1 | 1986 | nf_conntrack_htable_size |
4481374c | 1987 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1988 | / sizeof(struct hlist_head)); |
88eab472 ML |
1989 | if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE))) |
1990 | nf_conntrack_htable_size = 65536; | |
1991 | else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
f205c5e0 PM |
1992 | nf_conntrack_htable_size = 16384; |
1993 | if (nf_conntrack_htable_size < 32) | |
1994 | nf_conntrack_htable_size = 32; | |
1995 | ||
1996 | /* Use a max. factor of four by default to get the same max as | |
1997 | * with the old struct list_heads. When a table size is given | |
1998 | * we use the old value of 8 to avoid reducing the max. | |
1999 | * entries. */ | |
2000 | max_factor = 4; | |
9fb9cbb1 | 2001 | } |
56d52d48 FW |
2002 | |
2003 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1); | |
2004 | if (!nf_conntrack_hash) | |
2005 | return -ENOMEM; | |
2006 | ||
f205c5e0 | 2007 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 2008 | |
0c5366b3 | 2009 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
a9e419dc FW |
2010 | sizeof(struct nf_conn), |
2011 | NFCT_INFOMASK + 1, | |
5f0d5a3a | 2012 | SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL); |
0c5366b3 FW |
2013 | if (!nf_conntrack_cachep) |
2014 | goto err_cachep; | |
2015 | ||
654d0fbd | 2016 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
2017 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
2018 | nf_conntrack_max); | |
83b4dbe1 G |
2019 | |
2020 | ret = nf_conntrack_expect_init(); | |
2021 | if (ret < 0) | |
2022 | goto err_expect; | |
2023 | ||
b7ff3a1f G |
2024 | ret = nf_conntrack_acct_init(); |
2025 | if (ret < 0) | |
2026 | goto err_acct; | |
2027 | ||
73f4001a G |
2028 | ret = nf_conntrack_tstamp_init(); |
2029 | if (ret < 0) | |
2030 | goto err_tstamp; | |
2031 | ||
3fe0f943 G |
2032 | ret = nf_conntrack_ecache_init(); |
2033 | if (ret < 0) | |
2034 | goto err_ecache; | |
2035 | ||
8684094c G |
2036 | ret = nf_conntrack_timeout_init(); |
2037 | if (ret < 0) | |
2038 | goto err_timeout; | |
2039 | ||
5e615b22 G |
2040 | ret = nf_conntrack_helper_init(); |
2041 | if (ret < 0) | |
2042 | goto err_helper; | |
2043 | ||
5f69b8f5 G |
2044 | ret = nf_conntrack_labels_init(); |
2045 | if (ret < 0) | |
2046 | goto err_labels; | |
2047 | ||
41d73ec0 PM |
2048 | ret = nf_conntrack_seqadj_init(); |
2049 | if (ret < 0) | |
2050 | goto err_seqadj; | |
2051 | ||
04d87001 G |
2052 | ret = nf_conntrack_proto_init(); |
2053 | if (ret < 0) | |
2054 | goto err_proto; | |
2055 | ||
b87a2f91 | 2056 | conntrack_gc_work_init(&conntrack_gc_work); |
e5072053 | 2057 | queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, HZ); |
b87a2f91 | 2058 | |
08f6547d AD |
2059 | return 0; |
2060 | ||
04d87001 | 2061 | err_proto: |
41d73ec0 PM |
2062 | nf_conntrack_seqadj_fini(); |
2063 | err_seqadj: | |
04d87001 | 2064 | nf_conntrack_labels_fini(); |
5f69b8f5 G |
2065 | err_labels: |
2066 | nf_conntrack_helper_fini(); | |
5e615b22 G |
2067 | err_helper: |
2068 | nf_conntrack_timeout_fini(); | |
8684094c G |
2069 | err_timeout: |
2070 | nf_conntrack_ecache_fini(); | |
3fe0f943 G |
2071 | err_ecache: |
2072 | nf_conntrack_tstamp_fini(); | |
73f4001a G |
2073 | err_tstamp: |
2074 | nf_conntrack_acct_fini(); | |
b7ff3a1f G |
2075 | err_acct: |
2076 | nf_conntrack_expect_fini(); | |
83b4dbe1 | 2077 | err_expect: |
0c5366b3 FW |
2078 | kmem_cache_destroy(nf_conntrack_cachep); |
2079 | err_cachep: | |
56d52d48 | 2080 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
08f6547d AD |
2081 | return ret; |
2082 | } | |
2083 | ||
f94161c1 G |
2084 | void nf_conntrack_init_end(void) |
2085 | { | |
2086 | /* For use by REJECT target */ | |
2087 | RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); | |
2088 | RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); | |
f94161c1 G |
2089 | } |
2090 | ||
8cc20198 ED |
2091 | /* |
2092 | * We need to use special "null" values, not used in hash table | |
2093 | */ | |
2094 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
2095 | #define DYING_NULLS_VAL ((1<<30)+1) | |
252b3e8c | 2096 | #define TEMPLATE_NULLS_VAL ((1<<30)+2) |
8cc20198 | 2097 | |
f94161c1 | 2098 | int nf_conntrack_init_net(struct net *net) |
08f6547d | 2099 | { |
b7779d06 JDB |
2100 | int ret = -ENOMEM; |
2101 | int cpu; | |
ceceae1b | 2102 | |
cc41c84b | 2103 | BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER); |
08f6547d | 2104 | atomic_set(&net->ct.count, 0); |
b7779d06 JDB |
2105 | |
2106 | net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu); | |
2107 | if (!net->ct.pcpu_lists) | |
08f6547d | 2108 | goto err_stat; |
b7779d06 JDB |
2109 | |
2110 | for_each_possible_cpu(cpu) { | |
2111 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
2112 | ||
2113 | spin_lock_init(&pcpu->lock); | |
2114 | INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL); | |
2115 | INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL); | |
08f6547d | 2116 | } |
5b3501fa | 2117 | |
b7779d06 JDB |
2118 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
2119 | if (!net->ct.stat) | |
2120 | goto err_pcpu_lists; | |
2121 | ||
83b4dbe1 | 2122 | ret = nf_conntrack_expect_pernet_init(net); |
08f6547d AD |
2123 | if (ret < 0) |
2124 | goto err_expect; | |
b7ff3a1f | 2125 | ret = nf_conntrack_acct_pernet_init(net); |
58401572 | 2126 | if (ret < 0) |
08f6547d | 2127 | goto err_acct; |
73f4001a | 2128 | ret = nf_conntrack_tstamp_pernet_init(net); |
a992ca2a PNA |
2129 | if (ret < 0) |
2130 | goto err_tstamp; | |
3fe0f943 | 2131 | ret = nf_conntrack_ecache_pernet_init(net); |
a0891aa6 PNA |
2132 | if (ret < 0) |
2133 | goto err_ecache; | |
5e615b22 | 2134 | ret = nf_conntrack_helper_pernet_init(net); |
a9006892 EL |
2135 | if (ret < 0) |
2136 | goto err_helper; | |
04d87001 | 2137 | ret = nf_conntrack_proto_pernet_init(net); |
f94161c1 G |
2138 | if (ret < 0) |
2139 | goto err_proto; | |
08f6547d | 2140 | return 0; |
c539f017 | 2141 | |
f94161c1 | 2142 | err_proto: |
5e615b22 | 2143 | nf_conntrack_helper_pernet_fini(net); |
a9006892 | 2144 | err_helper: |
3fe0f943 | 2145 | nf_conntrack_ecache_pernet_fini(net); |
a0891aa6 | 2146 | err_ecache: |
73f4001a | 2147 | nf_conntrack_tstamp_pernet_fini(net); |
a992ca2a | 2148 | err_tstamp: |
b7ff3a1f | 2149 | nf_conntrack_acct_pernet_fini(net); |
08f6547d | 2150 | err_acct: |
83b4dbe1 | 2151 | nf_conntrack_expect_pernet_fini(net); |
08f6547d | 2152 | err_expect: |
0d55af87 | 2153 | free_percpu(net->ct.stat); |
b7779d06 JDB |
2154 | err_pcpu_lists: |
2155 | free_percpu(net->ct.pcpu_lists); | |
0d55af87 | 2156 | err_stat: |
08f6547d AD |
2157 | return ret; |
2158 | } |