]>
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 | |
e2a75007 FW |
59 | #include "nf_internals.h" |
60 | ||
dc808fe2 | 61 | #define NF_CONNTRACK_VERSION "0.5.0" |
9fb9cbb1 | 62 | |
e17b666a PM |
63 | int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, |
64 | enum nf_nat_manip_type manip, | |
39938324 | 65 | const struct nlattr *attr) __read_mostly; |
e6a7d3c0 PNA |
66 | EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); |
67 | ||
93bb0ceb JDB |
68 | __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS]; |
69 | EXPORT_SYMBOL_GPL(nf_conntrack_locks); | |
9fb9cbb1 | 70 | |
ca7433df JDB |
71 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock); |
72 | EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock); | |
73 | ||
56d52d48 FW |
74 | struct hlist_nulls_head *nf_conntrack_hash __read_mostly; |
75 | EXPORT_SYMBOL_GPL(nf_conntrack_hash); | |
76 | ||
b87a2f91 FW |
77 | struct conntrack_gc_work { |
78 | struct delayed_work dwork; | |
79 | u32 last_bucket; | |
80 | bool exiting; | |
c6dd940b | 81 | bool early_drop; |
e0df8cae | 82 | long next_gc_run; |
b87a2f91 FW |
83 | }; |
84 | ||
0c5366b3 | 85 | static __read_mostly struct kmem_cache *nf_conntrack_cachep; |
b16c2919 | 86 | static __read_mostly spinlock_t nf_conntrack_locks_all_lock; |
70d72b7e | 87 | static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock); |
b16c2919 SL |
88 | static __read_mostly bool nf_conntrack_locks_all; |
89 | ||
e0df8cae | 90 | /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */ |
e5072053 FW |
91 | #define GC_MAX_BUCKETS_DIV 128u |
92 | /* upper bound of full table scan */ | |
93 | #define GC_MAX_SCAN_JIFFIES (16u * HZ) | |
94 | /* desired ratio of entries found to be expired */ | |
95 | #define GC_EVICT_RATIO 50u | |
b87a2f91 FW |
96 | |
97 | static struct conntrack_gc_work conntrack_gc_work; | |
98 | ||
b16c2919 SL |
99 | void nf_conntrack_lock(spinlock_t *lock) __acquires(lock) |
100 | { | |
3ef0c7a7 | 101 | /* 1) Acquire the lock */ |
b16c2919 | 102 | spin_lock(lock); |
b316ff78 | 103 | |
3ef0c7a7 MS |
104 | /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics |
105 | * It pairs with the smp_store_release() in nf_conntrack_all_unlock() | |
106 | */ | |
107 | if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false)) | |
108 | return; | |
109 | ||
110 | /* fast path failed, unlock */ | |
111 | spin_unlock(lock); | |
112 | ||
113 | /* Slow path 1) get global lock */ | |
114 | spin_lock(&nf_conntrack_locks_all_lock); | |
115 | ||
116 | /* Slow path 2) get the lock we want */ | |
117 | spin_lock(lock); | |
118 | ||
119 | /* Slow path 3) release the global lock */ | |
120 | spin_unlock(&nf_conntrack_locks_all_lock); | |
b16c2919 SL |
121 | } |
122 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); | |
123 | ||
93bb0ceb JDB |
124 | static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2) |
125 | { | |
126 | h1 %= CONNTRACK_LOCKS; | |
127 | h2 %= CONNTRACK_LOCKS; | |
128 | spin_unlock(&nf_conntrack_locks[h1]); | |
129 | if (h1 != h2) | |
130 | spin_unlock(&nf_conntrack_locks[h2]); | |
131 | } | |
132 | ||
133 | /* return true if we need to recompute hashes (in case hash table was resized) */ | |
134 | static bool nf_conntrack_double_lock(struct net *net, unsigned int h1, | |
135 | unsigned int h2, unsigned int sequence) | |
136 | { | |
137 | h1 %= CONNTRACK_LOCKS; | |
138 | h2 %= CONNTRACK_LOCKS; | |
139 | if (h1 <= h2) { | |
b16c2919 | 140 | nf_conntrack_lock(&nf_conntrack_locks[h1]); |
93bb0ceb JDB |
141 | if (h1 != h2) |
142 | spin_lock_nested(&nf_conntrack_locks[h2], | |
143 | SINGLE_DEPTH_NESTING); | |
144 | } else { | |
b16c2919 | 145 | nf_conntrack_lock(&nf_conntrack_locks[h2]); |
93bb0ceb JDB |
146 | spin_lock_nested(&nf_conntrack_locks[h1], |
147 | SINGLE_DEPTH_NESTING); | |
148 | } | |
a3efd812 | 149 | if (read_seqcount_retry(&nf_conntrack_generation, sequence)) { |
93bb0ceb JDB |
150 | nf_conntrack_double_unlock(h1, h2); |
151 | return true; | |
152 | } | |
153 | return false; | |
154 | } | |
155 | ||
156 | static void nf_conntrack_all_lock(void) | |
157 | { | |
158 | int i; | |
159 | ||
b16c2919 | 160 | spin_lock(&nf_conntrack_locks_all_lock); |
b16c2919 | 161 | |
3ef0c7a7 | 162 | nf_conntrack_locks_all = true; |
b316ff78 | 163 | |
b16c2919 | 164 | for (i = 0; i < CONNTRACK_LOCKS; i++) { |
3ef0c7a7 MS |
165 | spin_lock(&nf_conntrack_locks[i]); |
166 | ||
167 | /* This spin_unlock provides the "release" to ensure that | |
168 | * nf_conntrack_locks_all==true is visible to everyone that | |
169 | * acquired spin_lock(&nf_conntrack_locks[]). | |
170 | */ | |
171 | spin_unlock(&nf_conntrack_locks[i]); | |
b16c2919 | 172 | } |
93bb0ceb JDB |
173 | } |
174 | ||
175 | static void nf_conntrack_all_unlock(void) | |
176 | { | |
3ef0c7a7 | 177 | /* All prior stores must be complete before we clear |
b316ff78 PZ |
178 | * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock() |
179 | * might observe the false value but not the entire | |
3ef0c7a7 MS |
180 | * critical section. |
181 | * It pairs with the smp_load_acquire() in nf_conntrack_lock() | |
b316ff78 PZ |
182 | */ |
183 | smp_store_release(&nf_conntrack_locks_all, false); | |
b16c2919 | 184 | spin_unlock(&nf_conntrack_locks_all_lock); |
93bb0ceb JDB |
185 | } |
186 | ||
e2b7606c | 187 | unsigned int nf_conntrack_htable_size __read_mostly; |
2567c4ea PNA |
188 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
189 | ||
e478075c | 190 | unsigned int nf_conntrack_max __read_mostly; |
92e47ba8 | 191 | seqcount_t nf_conntrack_generation __read_mostly; |
141658fb | 192 | static unsigned int nf_conntrack_hash_rnd __read_mostly; |
9fb9cbb1 | 193 | |
1b8c8a9f FW |
194 | static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, |
195 | const struct net *net) | |
9fb9cbb1 | 196 | { |
0794935e | 197 | unsigned int n; |
1b8c8a9f | 198 | u32 seed; |
0794935e | 199 | |
141658fb FW |
200 | get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd)); |
201 | ||
0794935e PM |
202 | /* The direction must be ignored, so we hash everything up to the |
203 | * destination ports (which is a multiple of 4) and treat the last | |
204 | * three bytes manually. | |
205 | */ | |
1b8c8a9f | 206 | seed = nf_conntrack_hash_rnd ^ net_hash_mix(net); |
0794935e | 207 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); |
1b8c8a9f | 208 | return jhash2((u32 *)tuple, n, seed ^ |
99f07e91 CG |
209 | (((__force __u16)tuple->dst.u.all << 16) | |
210 | tuple->dst.protonum)); | |
211 | } | |
212 | ||
56d52d48 | 213 | static u32 scale_hash(u32 hash) |
99f07e91 | 214 | { |
56d52d48 | 215 | return reciprocal_scale(hash, nf_conntrack_htable_size); |
99f07e91 | 216 | } |
0794935e | 217 | |
1b8c8a9f FW |
218 | static u32 __hash_conntrack(const struct net *net, |
219 | const struct nf_conntrack_tuple *tuple, | |
220 | unsigned int size) | |
99f07e91 | 221 | { |
1b8c8a9f | 222 | return reciprocal_scale(hash_conntrack_raw(tuple, net), size); |
9fb9cbb1 YK |
223 | } |
224 | ||
1b8c8a9f FW |
225 | static u32 hash_conntrack(const struct net *net, |
226 | const struct nf_conntrack_tuple *tuple) | |
9fb9cbb1 | 227 | { |
56d52d48 | 228 | return scale_hash(hash_conntrack_raw(tuple, net)); |
9fb9cbb1 YK |
229 | } |
230 | ||
5f2b4c90 | 231 | bool |
9fb9cbb1 YK |
232 | nf_ct_get_tuple(const struct sk_buff *skb, |
233 | unsigned int nhoff, | |
234 | unsigned int dataoff, | |
235 | u_int16_t l3num, | |
236 | u_int8_t protonum, | |
a31f1adc | 237 | struct net *net, |
9fb9cbb1 YK |
238 | struct nf_conntrack_tuple *tuple, |
239 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 240 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 241 | { |
443a70d5 | 242 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
243 | |
244 | tuple->src.l3num = l3num; | |
245 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 246 | return false; |
9fb9cbb1 YK |
247 | |
248 | tuple->dst.protonum = protonum; | |
249 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
250 | ||
a31f1adc | 251 | return l4proto->pkt_to_tuple(skb, dataoff, net, tuple); |
9fb9cbb1 | 252 | } |
13b18339 | 253 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 254 | |
5f2b4c90 | 255 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
a31f1adc EB |
256 | u_int16_t l3num, |
257 | struct net *net, struct nf_conntrack_tuple *tuple) | |
e2a3123f | 258 | { |
b3480fe0 FW |
259 | const struct nf_conntrack_l3proto *l3proto; |
260 | const struct nf_conntrack_l4proto *l4proto; | |
e2a3123f YK |
261 | unsigned int protoff; |
262 | u_int8_t protonum; | |
263 | int ret; | |
264 | ||
265 | rcu_read_lock(); | |
266 | ||
267 | l3proto = __nf_ct_l3proto_find(l3num); | |
268 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
269 | if (ret != NF_ACCEPT) { | |
270 | rcu_read_unlock(); | |
5f2b4c90 | 271 | return false; |
e2a3123f YK |
272 | } |
273 | ||
274 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
275 | ||
a31f1adc | 276 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple, |
e2a3123f YK |
277 | l3proto, l4proto); |
278 | ||
279 | rcu_read_unlock(); | |
280 | return ret; | |
281 | } | |
282 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
283 | ||
5f2b4c90 | 284 | bool |
9fb9cbb1 YK |
285 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
286 | const struct nf_conntrack_tuple *orig, | |
287 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 288 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 289 | { |
443a70d5 | 290 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
291 | |
292 | inverse->src.l3num = orig->src.l3num; | |
293 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 294 | return false; |
9fb9cbb1 YK |
295 | |
296 | inverse->dst.dir = !orig->dst.dir; | |
297 | ||
298 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 299 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 300 | } |
13b18339 | 301 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 302 | |
9fb9cbb1 YK |
303 | static void |
304 | clean_from_lists(struct nf_conn *ct) | |
305 | { | |
0d53778e | 306 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
307 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
308 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
309 | |
310 | /* Destroy all pending expectations */ | |
c1d10adb | 311 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
312 | } |
313 | ||
b7779d06 JDB |
314 | /* must be called with local_bh_disable */ |
315 | static void nf_ct_add_to_dying_list(struct nf_conn *ct) | |
316 | { | |
317 | struct ct_pcpu *pcpu; | |
318 | ||
319 | /* add this conntrack to the (per cpu) dying list */ | |
320 | ct->cpu = smp_processor_id(); | |
321 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
322 | ||
323 | spin_lock(&pcpu->lock); | |
324 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
325 | &pcpu->dying); | |
326 | spin_unlock(&pcpu->lock); | |
327 | } | |
328 | ||
329 | /* must be called with local_bh_disable */ | |
330 | static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct) | |
331 | { | |
332 | struct ct_pcpu *pcpu; | |
333 | ||
334 | /* add this conntrack to the (per cpu) unconfirmed list */ | |
335 | ct->cpu = smp_processor_id(); | |
336 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
337 | ||
338 | spin_lock(&pcpu->lock); | |
339 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
340 | &pcpu->unconfirmed); | |
341 | spin_unlock(&pcpu->lock); | |
342 | } | |
343 | ||
344 | /* must be called with local_bh_disable */ | |
345 | static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct) | |
346 | { | |
347 | struct ct_pcpu *pcpu; | |
348 | ||
349 | /* We overload first tuple to link into unconfirmed or dying list.*/ | |
350 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
351 | ||
352 | spin_lock(&pcpu->lock); | |
353 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); | |
354 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
355 | spin_unlock(&pcpu->lock); | |
356 | } | |
357 | ||
30322309 FW |
358 | #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK) |
359 | ||
0838aa7f | 360 | /* Released via destroy_conntrack() */ |
308ac914 DB |
361 | struct nf_conn *nf_ct_tmpl_alloc(struct net *net, |
362 | const struct nf_conntrack_zone *zone, | |
363 | gfp_t flags) | |
0838aa7f | 364 | { |
30322309 | 365 | struct nf_conn *tmpl, *p; |
0838aa7f | 366 | |
30322309 FW |
367 | if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) { |
368 | tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags); | |
369 | if (!tmpl) | |
370 | return NULL; | |
371 | ||
372 | p = tmpl; | |
373 | tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p); | |
374 | if (tmpl != p) { | |
375 | tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p); | |
376 | tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p; | |
377 | } | |
378 | } else { | |
379 | tmpl = kzalloc(sizeof(*tmpl), flags); | |
380 | if (!tmpl) | |
381 | return NULL; | |
382 | } | |
0838aa7f PNA |
383 | |
384 | tmpl->status = IPS_TEMPLATE; | |
385 | write_pnet(&tmpl->ct_net, net); | |
6c8dee98 | 386 | nf_ct_zone_add(tmpl, zone); |
0838aa7f PNA |
387 | atomic_set(&tmpl->ct_general.use, 0); |
388 | ||
389 | return tmpl; | |
0838aa7f PNA |
390 | } |
391 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc); | |
392 | ||
9cf94eab | 393 | void nf_ct_tmpl_free(struct nf_conn *tmpl) |
0838aa7f PNA |
394 | { |
395 | nf_ct_ext_destroy(tmpl); | |
396 | nf_ct_ext_free(tmpl); | |
30322309 FW |
397 | |
398 | if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) | |
399 | kfree((char *)tmpl - tmpl->proto.tmpl_padto); | |
400 | else | |
401 | kfree(tmpl); | |
0838aa7f | 402 | } |
9cf94eab | 403 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_free); |
0838aa7f | 404 | |
9fb9cbb1 YK |
405 | static void |
406 | destroy_conntrack(struct nf_conntrack *nfct) | |
407 | { | |
408 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
b3480fe0 | 409 | const struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 410 | |
0d53778e | 411 | pr_debug("destroy_conntrack(%p)\n", ct); |
44d6e2f2 | 412 | WARN_ON(atomic_read(&nfct->use) != 0); |
9fb9cbb1 | 413 | |
0838aa7f PNA |
414 | if (unlikely(nf_ct_is_template(ct))) { |
415 | nf_ct_tmpl_free(ct); | |
416 | return; | |
417 | } | |
5e8fbe2a | 418 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
4b4ceb9d | 419 | if (l4proto->destroy) |
605dcad6 | 420 | l4proto->destroy(ct); |
9fb9cbb1 | 421 | |
ca7433df | 422 | local_bh_disable(); |
9fb9cbb1 YK |
423 | /* Expectations will have been removed in clean_from_lists, |
424 | * except TFTP can create an expectation on the first packet, | |
425 | * before connection is in the list, so we need to clean here, | |
ca7433df JDB |
426 | * too. |
427 | */ | |
c1d10adb | 428 | nf_ct_remove_expectations(ct); |
9fb9cbb1 | 429 | |
756cef5e | 430 | #if defined(CONFIG_NETFILTER_XT_MATCH_LAYER7) || defined(CONFIG_NETFILTER_XT_MATCH_LAYER7_MODULE) |
756cef5e AF |
431 | if(ct->layer7.app_data) |
432 | kfree(ct->layer7.app_data); | |
433 | #endif | |
434 | ||
b7779d06 | 435 | nf_ct_del_from_dying_or_unconfirmed_list(ct); |
9fb9cbb1 | 436 | |
ca7433df | 437 | local_bh_enable(); |
9fb9cbb1 YK |
438 | |
439 | if (ct->master) | |
440 | nf_ct_put(ct->master); | |
441 | ||
0d53778e | 442 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
443 | nf_conntrack_free(ct); |
444 | } | |
445 | ||
02982c27 | 446 | static void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 447 | { |
0d55af87 | 448 | struct net *net = nf_ct_net(ct); |
93bb0ceb | 449 | unsigned int hash, reply_hash; |
93bb0ceb | 450 | unsigned int sequence; |
9fb9cbb1 | 451 | |
9858a3ae | 452 | nf_ct_helper_destroy(ct); |
93bb0ceb JDB |
453 | |
454 | local_bh_disable(); | |
455 | do { | |
a3efd812 | 456 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 457 | hash = hash_conntrack(net, |
93bb0ceb | 458 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 459 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
460 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
461 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
462 | ||
9fb9cbb1 | 463 | clean_from_lists(ct); |
93bb0ceb JDB |
464 | nf_conntrack_double_unlock(hash, reply_hash); |
465 | ||
b7779d06 | 466 | nf_ct_add_to_dying_list(ct); |
93bb0ceb | 467 | |
93bb0ceb | 468 | local_bh_enable(); |
dd7669a9 | 469 | } |
dd7669a9 | 470 | |
02982c27 | 471 | bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report) |
dd7669a9 | 472 | { |
a992ca2a PNA |
473 | struct nf_conn_tstamp *tstamp; |
474 | ||
f330a7fd FW |
475 | if (test_and_set_bit(IPS_DYING_BIT, &ct->status)) |
476 | return false; | |
477 | ||
a992ca2a PNA |
478 | tstamp = nf_conn_tstamp_find(ct); |
479 | if (tstamp && tstamp->stop == 0) | |
d2de875c | 480 | tstamp->stop = ktime_get_real_ns(); |
dd7669a9 | 481 | |
9500507c FW |
482 | if (nf_conntrack_event_report(IPCT_DESTROY, ct, |
483 | portid, report) < 0) { | |
f330a7fd FW |
484 | /* destroy event was not delivered. nf_ct_put will |
485 | * be done by event cache worker on redelivery. | |
486 | */ | |
dd7669a9 | 487 | nf_ct_delete_from_lists(ct); |
9500507c | 488 | nf_conntrack_ecache_delayed_work(nf_ct_net(ct)); |
02982c27 | 489 | return false; |
dd7669a9 | 490 | } |
9500507c FW |
491 | |
492 | nf_conntrack_ecache_work(nf_ct_net(ct)); | |
dd7669a9 | 493 | nf_ct_delete_from_lists(ct); |
9fb9cbb1 | 494 | nf_ct_put(ct); |
02982c27 FW |
495 | return true; |
496 | } | |
497 | EXPORT_SYMBOL_GPL(nf_ct_delete); | |
498 | ||
c6825c09 AV |
499 | static inline bool |
500 | nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, | |
308ac914 | 501 | const struct nf_conntrack_tuple *tuple, |
e0c7d472 FW |
502 | const struct nf_conntrack_zone *zone, |
503 | const struct net *net) | |
c6825c09 AV |
504 | { |
505 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
506 | ||
507 | /* A conntrack can be recreated with the equal tuple, | |
508 | * so we need to check that the conntrack is confirmed | |
509 | */ | |
510 | return nf_ct_tuple_equal(tuple, &h->tuple) && | |
deedb590 | 511 | nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) && |
e0c7d472 FW |
512 | nf_ct_is_confirmed(ct) && |
513 | net_eq(net, nf_ct_net(ct)); | |
c6825c09 AV |
514 | } |
515 | ||
f330a7fd FW |
516 | /* caller must hold rcu readlock and none of the nf_conntrack_locks */ |
517 | static void nf_ct_gc_expired(struct nf_conn *ct) | |
518 | { | |
519 | if (!atomic_inc_not_zero(&ct->ct_general.use)) | |
520 | return; | |
521 | ||
522 | if (nf_ct_should_gc(ct)) | |
523 | nf_ct_kill(ct); | |
524 | ||
525 | nf_ct_put(ct); | |
526 | } | |
527 | ||
ea781f19 ED |
528 | /* |
529 | * Warning : | |
530 | * - Caller must take a reference on returned object | |
531 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
ea781f19 | 532 | */ |
99f07e91 | 533 | static struct nf_conntrack_tuple_hash * |
308ac914 | 534 | ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 535 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
536 | { |
537 | struct nf_conntrack_tuple_hash *h; | |
5e3c61f9 | 538 | struct hlist_nulls_head *ct_hash; |
ea781f19 | 539 | struct hlist_nulls_node *n; |
92e47ba8 | 540 | unsigned int bucket, hsize; |
9fb9cbb1 | 541 | |
ea781f19 | 542 | begin: |
92e47ba8 LZ |
543 | nf_conntrack_get_ht(&ct_hash, &hsize); |
544 | bucket = reciprocal_scale(hash, hsize); | |
5e3c61f9 FW |
545 | |
546 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) { | |
f330a7fd FW |
547 | struct nf_conn *ct; |
548 | ||
549 | ct = nf_ct_tuplehash_to_ctrack(h); | |
550 | if (nf_ct_is_expired(ct)) { | |
551 | nf_ct_gc_expired(ct); | |
552 | continue; | |
553 | } | |
554 | ||
555 | if (nf_ct_is_dying(ct)) | |
556 | continue; | |
557 | ||
8e8118f8 | 558 | if (nf_ct_key_equal(h, tuple, zone, net)) |
9fb9cbb1 | 559 | return h; |
9fb9cbb1 | 560 | } |
ea781f19 ED |
561 | /* |
562 | * if the nulls value we got at the end of this lookup is | |
563 | * not the expected one, we must restart lookup. | |
564 | * We probably met an item that was moved to another chain. | |
565 | */ | |
99f07e91 | 566 | if (get_nulls_value(n) != bucket) { |
2cf12348 | 567 | NF_CT_STAT_INC_ATOMIC(net, search_restart); |
ea781f19 | 568 | goto begin; |
af740b2c | 569 | } |
9fb9cbb1 YK |
570 | |
571 | return NULL; | |
572 | } | |
99f07e91 | 573 | |
9fb9cbb1 | 574 | /* Find a connection corresponding to a tuple. */ |
99f07e91 | 575 | static struct nf_conntrack_tuple_hash * |
308ac914 | 576 | __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 577 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
578 | { |
579 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 580 | struct nf_conn *ct; |
9fb9cbb1 | 581 | |
76507f69 | 582 | rcu_read_lock(); |
ea781f19 | 583 | begin: |
99f07e91 | 584 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
585 | if (h) { |
586 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
587 | if (unlikely(nf_ct_is_dying(ct) || |
588 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 589 | h = NULL; |
ea781f19 | 590 | else { |
e0c7d472 | 591 | if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) { |
ea781f19 ED |
592 | nf_ct_put(ct); |
593 | goto begin; | |
594 | } | |
595 | } | |
76507f69 PM |
596 | } |
597 | rcu_read_unlock(); | |
9fb9cbb1 YK |
598 | |
599 | return h; | |
600 | } | |
99f07e91 CG |
601 | |
602 | struct nf_conntrack_tuple_hash * | |
308ac914 | 603 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 CG |
604 | const struct nf_conntrack_tuple *tuple) |
605 | { | |
606 | return __nf_conntrack_find_get(net, zone, tuple, | |
1b8c8a9f | 607 | hash_conntrack_raw(tuple, net)); |
99f07e91 | 608 | } |
13b18339 | 609 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 610 | |
c1d10adb PNA |
611 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
612 | unsigned int hash, | |
b476b72a | 613 | unsigned int reply_hash) |
c1d10adb | 614 | { |
ea781f19 | 615 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
56d52d48 | 616 | &nf_conntrack_hash[hash]); |
ea781f19 | 617 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
56d52d48 | 618 | &nf_conntrack_hash[reply_hash]); |
c1d10adb PNA |
619 | } |
620 | ||
7d367e06 JK |
621 | int |
622 | nf_conntrack_hash_check_insert(struct nf_conn *ct) | |
c1d10adb | 623 | { |
308ac914 | 624 | const struct nf_conntrack_zone *zone; |
d696c7bd | 625 | struct net *net = nf_ct_net(ct); |
b476b72a | 626 | unsigned int hash, reply_hash; |
7d367e06 JK |
627 | struct nf_conntrack_tuple_hash *h; |
628 | struct hlist_nulls_node *n; | |
93bb0ceb | 629 | unsigned int sequence; |
c1d10adb | 630 | |
5d0aa2cc | 631 | zone = nf_ct_zone(ct); |
7d367e06 | 632 | |
93bb0ceb JDB |
633 | local_bh_disable(); |
634 | do { | |
a3efd812 | 635 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 636 | hash = hash_conntrack(net, |
93bb0ceb | 637 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 638 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
639 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
640 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
7d367e06 JK |
641 | |
642 | /* See if there's one in the list already, including reverse */ | |
56d52d48 | 643 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 644 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 645 | zone, net)) |
7d367e06 | 646 | goto out; |
86804348 | 647 | |
56d52d48 | 648 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 649 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 650 | zone, net)) |
7d367e06 | 651 | goto out; |
c1d10adb | 652 | |
e53376be PNA |
653 | smp_wmb(); |
654 | /* The caller holds a reference to this object */ | |
655 | atomic_set(&ct->ct_general.use, 2); | |
b476b72a | 656 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 657 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 658 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 659 | local_bh_enable(); |
7d367e06 JK |
660 | return 0; |
661 | ||
662 | out: | |
93bb0ceb | 663 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 664 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 665 | local_bh_enable(); |
7d367e06 | 666 | return -EEXIST; |
c1d10adb | 667 | } |
7d367e06 | 668 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); |
c1d10adb | 669 | |
ba76738c PNA |
670 | static inline void nf_ct_acct_update(struct nf_conn *ct, |
671 | enum ip_conntrack_info ctinfo, | |
672 | unsigned int len) | |
673 | { | |
674 | struct nf_conn_acct *acct; | |
675 | ||
676 | acct = nf_conn_acct_find(ct); | |
677 | if (acct) { | |
678 | struct nf_conn_counter *counter = acct->counter; | |
679 | ||
680 | atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); | |
681 | atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes); | |
682 | } | |
683 | } | |
684 | ||
71d8c47f PNA |
685 | static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo, |
686 | const struct nf_conn *loser_ct) | |
687 | { | |
688 | struct nf_conn_acct *acct; | |
689 | ||
690 | acct = nf_conn_acct_find(loser_ct); | |
691 | if (acct) { | |
692 | struct nf_conn_counter *counter = acct->counter; | |
71d8c47f PNA |
693 | unsigned int bytes; |
694 | ||
695 | /* u32 should be fine since we must have seen one packet. */ | |
696 | bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes); | |
697 | nf_ct_acct_update(ct, ctinfo, bytes); | |
698 | } | |
699 | } | |
700 | ||
701 | /* Resolve race on insertion if this protocol allows this. */ | |
702 | static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb, | |
703 | enum ip_conntrack_info ctinfo, | |
704 | struct nf_conntrack_tuple_hash *h) | |
705 | { | |
706 | /* This is the conntrack entry already in hashes that won race. */ | |
707 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
b3480fe0 | 708 | const struct nf_conntrack_l4proto *l4proto; |
71d8c47f PNA |
709 | |
710 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); | |
711 | if (l4proto->allow_clash && | |
6e699867 | 712 | ((ct->status & IPS_NAT_DONE_MASK) == 0) && |
71d8c47f PNA |
713 | !nf_ct_is_dying(ct) && |
714 | atomic_inc_not_zero(&ct->ct_general.use)) { | |
97a6ad13 FW |
715 | enum ip_conntrack_info oldinfo; |
716 | struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo); | |
717 | ||
718 | nf_ct_acct_merge(ct, ctinfo, loser_ct); | |
719 | nf_conntrack_put(&loser_ct->ct_general); | |
c74454fa | 720 | nf_ct_set(skb, ct, oldinfo); |
71d8c47f PNA |
721 | return NF_ACCEPT; |
722 | } | |
723 | NF_CT_STAT_INC(net, drop); | |
724 | return NF_DROP; | |
725 | } | |
726 | ||
9fb9cbb1 YK |
727 | /* Confirm a connection given skb; places it in hash table */ |
728 | int | |
3db05fea | 729 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 | 730 | { |
308ac914 | 731 | const struct nf_conntrack_zone *zone; |
b476b72a | 732 | unsigned int hash, reply_hash; |
df0933dc | 733 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 734 | struct nf_conn *ct; |
df0933dc | 735 | struct nf_conn_help *help; |
a992ca2a | 736 | struct nf_conn_tstamp *tstamp; |
ea781f19 | 737 | struct hlist_nulls_node *n; |
9fb9cbb1 | 738 | enum ip_conntrack_info ctinfo; |
400dad39 | 739 | struct net *net; |
93bb0ceb | 740 | unsigned int sequence; |
71d8c47f | 741 | int ret = NF_DROP; |
9fb9cbb1 | 742 | |
3db05fea | 743 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 744 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
745 | |
746 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
747 | ICMP/TCP RST packets in other direction. Actual packet | |
748 | which created connection will be IP_CT_NEW or for an | |
749 | expected connection, IP_CT_RELATED. */ | |
750 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
751 | return NF_ACCEPT; | |
752 | ||
5d0aa2cc | 753 | zone = nf_ct_zone(ct); |
93bb0ceb JDB |
754 | local_bh_disable(); |
755 | ||
756 | do { | |
a3efd812 | 757 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb JDB |
758 | /* reuse the hash saved before */ |
759 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
56d52d48 | 760 | hash = scale_hash(hash); |
deedb590 | 761 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
762 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
763 | ||
764 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
9fb9cbb1 YK |
765 | |
766 | /* We're not in hash table, and we refuse to set up related | |
93bb0ceb JDB |
767 | * connections for unconfirmed conns. But packet copies and |
768 | * REJECT will give spurious warnings here. | |
769 | */ | |
9fb9cbb1 | 770 | |
25985edc | 771 | /* No external references means no one else could have |
93bb0ceb JDB |
772 | * confirmed us. |
773 | */ | |
44d6e2f2 | 774 | WARN_ON(nf_ct_is_confirmed(ct)); |
0d53778e | 775 | pr_debug("Confirming conntrack %p\n", ct); |
8ca3f5e9 PNA |
776 | /* We have to check the DYING flag after unlink to prevent |
777 | * a race against nf_ct_get_next_corpse() possibly called from | |
778 | * user context, else we insert an already 'dead' hash, blocking | |
779 | * further use of that particular connection -JM. | |
780 | */ | |
781 | nf_ct_del_from_dying_or_unconfirmed_list(ct); | |
782 | ||
71d8c47f PNA |
783 | if (unlikely(nf_ct_is_dying(ct))) { |
784 | nf_ct_add_to_dying_list(ct); | |
785 | goto dying; | |
786 | } | |
fc350777 | 787 | |
9fb9cbb1 YK |
788 | /* See if there's one in the list already, including reverse: |
789 | NAT could have grabbed it without realizing, since we're | |
790 | not in the hash. If there is, we lost race. */ | |
56d52d48 | 791 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 792 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 793 | zone, net)) |
df0933dc | 794 | goto out; |
86804348 | 795 | |
56d52d48 | 796 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 797 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 798 | zone, net)) |
df0933dc | 799 | goto out; |
9fb9cbb1 | 800 | |
df0933dc PM |
801 | /* Timer relative to confirmation time, not original |
802 | setting time, otherwise we'd get timer wrap in | |
803 | weird delay cases. */ | |
f330a7fd | 804 | ct->timeout += nfct_time_stamp; |
df0933dc | 805 | atomic_inc(&ct->ct_general.use); |
45eec341 | 806 | ct->status |= IPS_CONFIRMED; |
5c8ec910 | 807 | |
a992ca2a PNA |
808 | /* set conntrack timestamp, if enabled. */ |
809 | tstamp = nf_conn_tstamp_find(ct); | |
810 | if (tstamp) { | |
2456e855 | 811 | if (skb->tstamp == 0) |
e3192690 | 812 | __net_timestamp(skb); |
a992ca2a PNA |
813 | |
814 | tstamp->start = ktime_to_ns(skb->tstamp); | |
815 | } | |
5c8ec910 PM |
816 | /* Since the lookup is lockless, hash insertion must be done after |
817 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
818 | * guarantee that no other CPU can find the conntrack before the above | |
819 | * stores are visible. | |
820 | */ | |
b476b72a | 821 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 822 | nf_conntrack_double_unlock(hash, reply_hash); |
93bb0ceb | 823 | local_bh_enable(); |
5c8ec910 | 824 | |
df0933dc PM |
825 | help = nfct_help(ct); |
826 | if (help && help->helper) | |
a71996fc | 827 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 828 | |
df0933dc | 829 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 830 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 831 | return NF_ACCEPT; |
9fb9cbb1 | 832 | |
df0933dc | 833 | out: |
8ca3f5e9 | 834 | nf_ct_add_to_dying_list(ct); |
71d8c47f PNA |
835 | ret = nf_ct_resolve_clash(net, skb, ctinfo, h); |
836 | dying: | |
93bb0ceb | 837 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 838 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 839 | local_bh_enable(); |
71d8c47f | 840 | return ret; |
9fb9cbb1 | 841 | } |
13b18339 | 842 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
843 | |
844 | /* Returns true if a connection correspondings to the tuple (required | |
845 | for NAT). */ | |
846 | int | |
847 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
848 | const struct nf_conn *ignored_conntrack) | |
849 | { | |
400dad39 | 850 | struct net *net = nf_ct_net(ignored_conntrack); |
308ac914 | 851 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 | 852 | struct nf_conntrack_tuple_hash *h; |
5e3c61f9 | 853 | struct hlist_nulls_head *ct_hash; |
92e47ba8 | 854 | unsigned int hash, hsize; |
ea781f19 | 855 | struct hlist_nulls_node *n; |
5d0aa2cc | 856 | struct nf_conn *ct; |
308ac914 DB |
857 | |
858 | zone = nf_ct_zone(ignored_conntrack); | |
9fb9cbb1 | 859 | |
2cf12348 | 860 | rcu_read_lock(); |
95a8d19f | 861 | begin: |
92e47ba8 LZ |
862 | nf_conntrack_get_ht(&ct_hash, &hsize); |
863 | hash = __hash_conntrack(net, tuple, hsize); | |
5e3c61f9 FW |
864 | |
865 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) { | |
5d0aa2cc | 866 | ct = nf_ct_tuplehash_to_ctrack(h); |
f330a7fd FW |
867 | |
868 | if (ct == ignored_conntrack) | |
869 | continue; | |
870 | ||
871 | if (nf_ct_is_expired(ct)) { | |
872 | nf_ct_gc_expired(ct); | |
873 | continue; | |
874 | } | |
875 | ||
876 | if (nf_ct_key_equal(h, tuple, zone, net)) { | |
2cf12348 FW |
877 | NF_CT_STAT_INC_ATOMIC(net, found); |
878 | rcu_read_unlock(); | |
ba419aff PM |
879 | return 1; |
880 | } | |
ba419aff | 881 | } |
95a8d19f FW |
882 | |
883 | if (get_nulls_value(n) != hash) { | |
884 | NF_CT_STAT_INC_ATOMIC(net, search_restart); | |
885 | goto begin; | |
886 | } | |
887 | ||
2cf12348 | 888 | rcu_read_unlock(); |
9fb9cbb1 | 889 | |
ba419aff | 890 | return 0; |
9fb9cbb1 | 891 | } |
13b18339 | 892 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 893 | |
7ae7730f PM |
894 | #define NF_CT_EVICTION_RANGE 8 |
895 | ||
9fb9cbb1 YK |
896 | /* There's a small race here where we may free a just-assured |
897 | connection. Too bad: we're in trouble anyway. */ | |
242922a0 FW |
898 | static unsigned int early_drop_list(struct net *net, |
899 | struct hlist_nulls_head *head) | |
9fb9cbb1 | 900 | { |
9fb9cbb1 | 901 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 902 | struct hlist_nulls_node *n; |
242922a0 FW |
903 | unsigned int drops = 0; |
904 | struct nf_conn *tmp; | |
3e86638e | 905 | |
242922a0 FW |
906 | hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) { |
907 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
9fb9cbb1 | 908 | |
f330a7fd FW |
909 | if (nf_ct_is_expired(tmp)) { |
910 | nf_ct_gc_expired(tmp); | |
911 | continue; | |
912 | } | |
913 | ||
242922a0 FW |
914 | if (test_bit(IPS_ASSURED_BIT, &tmp->status) || |
915 | !net_eq(nf_ct_net(tmp), net) || | |
916 | nf_ct_is_dying(tmp)) | |
917 | continue; | |
76507f69 | 918 | |
242922a0 FW |
919 | if (!atomic_inc_not_zero(&tmp->ct_general.use)) |
920 | continue; | |
76507f69 | 921 | |
242922a0 | 922 | /* kill only if still in same netns -- might have moved due to |
5f0d5a3a | 923 | * SLAB_TYPESAFE_BY_RCU rules. |
242922a0 FW |
924 | * |
925 | * We steal the timer reference. If that fails timer has | |
926 | * already fired or someone else deleted it. Just drop ref | |
927 | * and move to next entry. | |
928 | */ | |
929 | if (net_eq(nf_ct_net(tmp), net) && | |
930 | nf_ct_is_confirmed(tmp) && | |
242922a0 FW |
931 | nf_ct_delete(tmp, 0, 0)) |
932 | drops++; | |
933 | ||
934 | nf_ct_put(tmp); | |
9fb9cbb1 | 935 | } |
3e86638e | 936 | |
242922a0 FW |
937 | return drops; |
938 | } | |
9fb9cbb1 | 939 | |
242922a0 FW |
940 | static noinline int early_drop(struct net *net, unsigned int _hash) |
941 | { | |
942 | unsigned int i; | |
9fb9cbb1 | 943 | |
242922a0 FW |
944 | for (i = 0; i < NF_CT_EVICTION_RANGE; i++) { |
945 | struct hlist_nulls_head *ct_hash; | |
92e47ba8 | 946 | unsigned int hash, hsize, drops; |
242922a0 | 947 | |
3101e0fc | 948 | rcu_read_lock(); |
92e47ba8 LZ |
949 | nf_conntrack_get_ht(&ct_hash, &hsize); |
950 | hash = reciprocal_scale(_hash++, hsize); | |
242922a0 FW |
951 | |
952 | drops = early_drop_list(net, &ct_hash[hash]); | |
3101e0fc LZ |
953 | rcu_read_unlock(); |
954 | ||
242922a0 FW |
955 | if (drops) { |
956 | NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops); | |
957 | return true; | |
74138511 | 958 | } |
9fb9cbb1 | 959 | } |
3e86638e | 960 | |
242922a0 | 961 | return false; |
9fb9cbb1 YK |
962 | } |
963 | ||
c6dd940b FW |
964 | static bool gc_worker_skip_ct(const struct nf_conn *ct) |
965 | { | |
966 | return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct); | |
967 | } | |
968 | ||
969 | static bool gc_worker_can_early_drop(const struct nf_conn *ct) | |
970 | { | |
971 | const struct nf_conntrack_l4proto *l4proto; | |
972 | ||
973 | if (!test_bit(IPS_ASSURED_BIT, &ct->status)) | |
974 | return true; | |
975 | ||
976 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); | |
977 | if (l4proto->can_early_drop && l4proto->can_early_drop(ct)) | |
978 | return true; | |
979 | ||
980 | return false; | |
981 | } | |
982 | ||
b87a2f91 FW |
983 | static void gc_worker(struct work_struct *work) |
984 | { | |
e5072053 | 985 | unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u); |
b87a2f91 | 986 | unsigned int i, goal, buckets = 0, expired_count = 0; |
c6dd940b | 987 | unsigned int nf_conntrack_max95 = 0; |
b87a2f91 | 988 | struct conntrack_gc_work *gc_work; |
e0df8cae FW |
989 | unsigned int ratio, scanned = 0; |
990 | unsigned long next_run; | |
b87a2f91 FW |
991 | |
992 | gc_work = container_of(work, struct conntrack_gc_work, dwork.work); | |
993 | ||
e0df8cae | 994 | goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV; |
b87a2f91 | 995 | i = gc_work->last_bucket; |
c6dd940b FW |
996 | if (gc_work->early_drop) |
997 | nf_conntrack_max95 = nf_conntrack_max / 100u * 95u; | |
b87a2f91 FW |
998 | |
999 | do { | |
1000 | struct nf_conntrack_tuple_hash *h; | |
1001 | struct hlist_nulls_head *ct_hash; | |
1002 | struct hlist_nulls_node *n; | |
1003 | unsigned int hashsz; | |
1004 | struct nf_conn *tmp; | |
1005 | ||
1006 | i++; | |
1007 | rcu_read_lock(); | |
1008 | ||
1009 | nf_conntrack_get_ht(&ct_hash, &hashsz); | |
1010 | if (i >= hashsz) | |
1011 | i = 0; | |
1012 | ||
1013 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) { | |
c6dd940b FW |
1014 | struct net *net; |
1015 | ||
b87a2f91 FW |
1016 | tmp = nf_ct_tuplehash_to_ctrack(h); |
1017 | ||
c023c0e4 | 1018 | scanned++; |
b87a2f91 FW |
1019 | if (nf_ct_is_expired(tmp)) { |
1020 | nf_ct_gc_expired(tmp); | |
1021 | expired_count++; | |
1022 | continue; | |
1023 | } | |
c6dd940b FW |
1024 | |
1025 | if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp)) | |
1026 | continue; | |
1027 | ||
1028 | net = nf_ct_net(tmp); | |
1029 | if (atomic_read(&net->ct.count) < nf_conntrack_max95) | |
1030 | continue; | |
1031 | ||
1032 | /* need to take reference to avoid possible races */ | |
1033 | if (!atomic_inc_not_zero(&tmp->ct_general.use)) | |
1034 | continue; | |
1035 | ||
1036 | if (gc_worker_skip_ct(tmp)) { | |
1037 | nf_ct_put(tmp); | |
1038 | continue; | |
1039 | } | |
1040 | ||
1041 | if (gc_worker_can_early_drop(tmp)) | |
1042 | nf_ct_kill(tmp); | |
1043 | ||
1044 | nf_ct_put(tmp); | |
b87a2f91 FW |
1045 | } |
1046 | ||
1047 | /* could check get_nulls_value() here and restart if ct | |
1048 | * was moved to another chain. But given gc is best-effort | |
1049 | * we will just continue with next hash slot. | |
1050 | */ | |
1051 | rcu_read_unlock(); | |
1052 | cond_resched_rcu_qs(); | |
524b698d | 1053 | } while (++buckets < goal); |
b87a2f91 FW |
1054 | |
1055 | if (gc_work->exiting) | |
1056 | return; | |
1057 | ||
e0df8cae FW |
1058 | /* |
1059 | * Eviction will normally happen from the packet path, and not | |
1060 | * from this gc worker. | |
1061 | * | |
1062 | * This worker is only here to reap expired entries when system went | |
1063 | * idle after a busy period. | |
1064 | * | |
1065 | * The heuristics below are supposed to balance conflicting goals: | |
1066 | * | |
1067 | * 1. Minimize time until we notice a stale entry | |
1068 | * 2. Maximize scan intervals to not waste cycles | |
1069 | * | |
e5072053 | 1070 | * Normally, expire ratio will be close to 0. |
e0df8cae | 1071 | * |
e5072053 FW |
1072 | * As soon as a sizeable fraction of the entries have expired |
1073 | * increase scan frequency. | |
e0df8cae | 1074 | */ |
c023c0e4 | 1075 | ratio = scanned ? expired_count * 100 / scanned : 0; |
e5072053 FW |
1076 | if (ratio > GC_EVICT_RATIO) { |
1077 | gc_work->next_gc_run = min_interval; | |
e0df8cae | 1078 | } else { |
e5072053 | 1079 | unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV; |
e0df8cae | 1080 | |
e5072053 FW |
1081 | BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0); |
1082 | ||
1083 | gc_work->next_gc_run += min_interval; | |
1084 | if (gc_work->next_gc_run > max) | |
1085 | gc_work->next_gc_run = max; | |
e0df8cae | 1086 | } |
c023c0e4 | 1087 | |
e5072053 | 1088 | next_run = gc_work->next_gc_run; |
b87a2f91 | 1089 | gc_work->last_bucket = i; |
c6dd940b | 1090 | gc_work->early_drop = false; |
e0df8cae | 1091 | queue_delayed_work(system_long_wq, &gc_work->dwork, next_run); |
b87a2f91 FW |
1092 | } |
1093 | ||
1094 | static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work) | |
1095 | { | |
a232cd0e | 1096 | INIT_DEFERRABLE_WORK(&gc_work->dwork, gc_worker); |
e5072053 | 1097 | gc_work->next_gc_run = HZ; |
b87a2f91 FW |
1098 | gc_work->exiting = false; |
1099 | } | |
1100 | ||
99f07e91 | 1101 | static struct nf_conn * |
308ac914 DB |
1102 | __nf_conntrack_alloc(struct net *net, |
1103 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1104 | const struct nf_conntrack_tuple *orig, |
1105 | const struct nf_conntrack_tuple *repl, | |
1106 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 1107 | { |
cd7fcbf1 | 1108 | struct nf_conn *ct; |
9fb9cbb1 | 1109 | |
5251e2d2 | 1110 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 1111 | atomic_inc(&net->ct.count); |
5251e2d2 | 1112 | |
76eb9460 | 1113 | if (nf_conntrack_max && |
49ac8713 | 1114 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
93bb0ceb | 1115 | if (!early_drop(net, hash)) { |
c6dd940b FW |
1116 | if (!conntrack_gc_work.early_drop) |
1117 | conntrack_gc_work.early_drop = true; | |
49ac8713 | 1118 | atomic_dec(&net->ct.count); |
e87cc472 | 1119 | net_warn_ratelimited("nf_conntrack: table full, dropping packet\n"); |
9fb9cbb1 YK |
1120 | return ERR_PTR(-ENOMEM); |
1121 | } | |
1122 | } | |
1123 | ||
941297f4 ED |
1124 | /* |
1125 | * Do not use kmem_cache_zalloc(), as this cache uses | |
5f0d5a3a | 1126 | * SLAB_TYPESAFE_BY_RCU. |
941297f4 | 1127 | */ |
0c5366b3 | 1128 | ct = kmem_cache_alloc(nf_conntrack_cachep, gfp); |
5e8018fc DB |
1129 | if (ct == NULL) |
1130 | goto out; | |
1131 | ||
440f0d58 | 1132 | spin_lock_init(&ct->lock); |
c88130bc | 1133 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 1134 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 1135 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
1136 | /* save hash for reusing when confirming */ |
1137 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
c41884ce | 1138 | ct->status = 0; |
c2d9ba9b | 1139 | write_pnet(&ct->ct_net, net); |
c41884ce FW |
1140 | memset(&ct->__nfct_init_offset[0], 0, |
1141 | offsetof(struct nf_conn, proto) - | |
1142 | offsetof(struct nf_conn, __nfct_init_offset[0])); | |
5e8018fc | 1143 | |
6c8dee98 | 1144 | nf_ct_zone_add(ct, zone); |
5e8018fc | 1145 | |
e53376be PNA |
1146 | /* Because we use RCU lookups, we set ct_general.use to zero before |
1147 | * this is inserted in any list. | |
941297f4 | 1148 | */ |
e53376be | 1149 | atomic_set(&ct->ct_general.use, 0); |
c88130bc | 1150 | return ct; |
5e8018fc DB |
1151 | out: |
1152 | atomic_dec(&net->ct.count); | |
5d0aa2cc | 1153 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 1154 | } |
99f07e91 | 1155 | |
308ac914 DB |
1156 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
1157 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1158 | const struct nf_conntrack_tuple *orig, |
1159 | const struct nf_conntrack_tuple *repl, | |
1160 | gfp_t gfp) | |
1161 | { | |
1162 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
1163 | } | |
13b18339 | 1164 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 1165 | |
c88130bc | 1166 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 1167 | { |
1d45209d ED |
1168 | struct net *net = nf_ct_net(ct); |
1169 | ||
e53376be | 1170 | /* A freed object has refcnt == 0, that's |
5f0d5a3a | 1171 | * the golden rule for SLAB_TYPESAFE_BY_RCU |
e53376be | 1172 | */ |
44d6e2f2 | 1173 | WARN_ON(atomic_read(&ct->ct_general.use) != 0); |
e53376be | 1174 | |
ceeff754 | 1175 | nf_ct_ext_destroy(ct); |
ea781f19 | 1176 | nf_ct_ext_free(ct); |
0c5366b3 | 1177 | kmem_cache_free(nf_conntrack_cachep, ct); |
4e857c58 | 1178 | smp_mb__before_atomic(); |
0c3c6c00 | 1179 | atomic_dec(&net->ct.count); |
76507f69 | 1180 | } |
13b18339 | 1181 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 | 1182 | |
c539f017 | 1183 | |
9fb9cbb1 YK |
1184 | /* Allocate a new conntrack: we return -ENOMEM if classification |
1185 | failed due to stress. Otherwise it really is unclassifiable. */ | |
fc09e4a7 | 1186 | static noinline struct nf_conntrack_tuple_hash * |
b2a15a60 | 1187 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 1188 | const struct nf_conntrack_tuple *tuple, |
2a04aabf JL |
1189 | const struct nf_conntrack_l3proto *l3proto, |
1190 | const struct nf_conntrack_l4proto *l4proto, | |
9fb9cbb1 | 1191 | struct sk_buff *skb, |
60b5f8f7 | 1192 | unsigned int dataoff, u32 hash) |
9fb9cbb1 | 1193 | { |
c88130bc | 1194 | struct nf_conn *ct; |
3c158f7f | 1195 | struct nf_conn_help *help; |
9fb9cbb1 | 1196 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 1197 | struct nf_conntrack_ecache *ecache; |
ca7433df | 1198 | struct nf_conntrack_expect *exp = NULL; |
308ac914 | 1199 | const struct nf_conntrack_zone *zone; |
60b5f8f7 | 1200 | struct nf_conn_timeout *timeout_ext; |
5e8018fc | 1201 | struct nf_conntrack_zone tmp; |
60b5f8f7 | 1202 | unsigned int *timeouts; |
9fb9cbb1 | 1203 | |
605dcad6 | 1204 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 1205 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
1206 | return NULL; |
1207 | } | |
1208 | ||
5e8018fc | 1209 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
99f07e91 CG |
1210 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
1211 | hash); | |
0a9ee813 | 1212 | if (IS_ERR(ct)) |
c88130bc | 1213 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 | 1214 | |
4440a2ab GF |
1215 | if (!nf_ct_add_synproxy(ct, tmpl)) { |
1216 | nf_conntrack_free(ct); | |
1217 | return ERR_PTR(-ENOMEM); | |
48b1de4c PM |
1218 | } |
1219 | ||
60b5f8f7 | 1220 | timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; |
ae2d708e PNA |
1221 | if (timeout_ext) { |
1222 | timeouts = nf_ct_timeout_data(timeout_ext); | |
1223 | if (unlikely(!timeouts)) | |
1224 | timeouts = l4proto->get_timeouts(net); | |
1225 | } else { | |
60b5f8f7 | 1226 | timeouts = l4proto->get_timeouts(net); |
ae2d708e | 1227 | } |
60b5f8f7 | 1228 | |
2c8503f5 | 1229 | if (!l4proto->new(ct, skb, dataoff, timeouts)) { |
c88130bc | 1230 | nf_conntrack_free(ct); |
ccd63c20 | 1231 | pr_debug("can't track with proto module\n"); |
9fb9cbb1 YK |
1232 | return NULL; |
1233 | } | |
1234 | ||
60b5f8f7 | 1235 | if (timeout_ext) |
ae2d708e PNA |
1236 | nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout), |
1237 | GFP_ATOMIC); | |
60b5f8f7 | 1238 | |
58401572 | 1239 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a992ca2a | 1240 | nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); |
c539f017 | 1241 | nf_ct_labels_ext_add(ct); |
b2a15a60 PM |
1242 | |
1243 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
1244 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
1245 | ecache ? ecache->expmask : 0, | |
1246 | GFP_ATOMIC); | |
58401572 | 1247 | |
ca7433df JDB |
1248 | local_bh_disable(); |
1249 | if (net->ct.expect_count) { | |
1250 | spin_lock(&nf_conntrack_expect_lock); | |
1251 | exp = nf_ct_find_expectation(net, zone, tuple); | |
1252 | if (exp) { | |
ccd63c20 | 1253 | pr_debug("expectation arrives ct=%p exp=%p\n", |
ca7433df JDB |
1254 | ct, exp); |
1255 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
1256 | __set_bit(IPS_EXPECTED_BIT, &ct->status); | |
1257 | /* exp->master safe, refcnt bumped in nf_ct_find_expectation */ | |
1258 | ct->master = exp->master; | |
1259 | if (exp->helper) { | |
1260 | help = nf_ct_helper_ext_add(ct, exp->helper, | |
1261 | GFP_ATOMIC); | |
1262 | if (help) | |
1263 | rcu_assign_pointer(help->helper, exp->helper); | |
1264 | } | |
ceceae1b | 1265 | |
9fb9cbb1 | 1266 | #ifdef CONFIG_NF_CONNTRACK_MARK |
ca7433df | 1267 | ct->mark = exp->master->mark; |
7c9728c3 JM |
1268 | #endif |
1269 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
ca7433df | 1270 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 1271 | #endif |
ca7433df JDB |
1272 | NF_CT_STAT_INC(net, expect_new); |
1273 | } | |
1274 | spin_unlock(&nf_conntrack_expect_lock); | |
1275 | } | |
8e8118f8 | 1276 | if (!exp) |
b2a15a60 | 1277 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
9fb9cbb1 | 1278 | |
e53376be PNA |
1279 | /* Now it is inserted into the unconfirmed list, bump refcount */ |
1280 | nf_conntrack_get(&ct->ct_general); | |
b7779d06 | 1281 | nf_ct_add_to_unconfirmed_list(ct); |
9fb9cbb1 | 1282 | |
ca7433df | 1283 | local_bh_enable(); |
9fb9cbb1 YK |
1284 | |
1285 | if (exp) { | |
1286 | if (exp->expectfn) | |
c88130bc | 1287 | exp->expectfn(ct, exp); |
6823645d | 1288 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
1289 | } |
1290 | ||
c88130bc | 1291 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
1292 | } |
1293 | ||
fc09e4a7 FW |
1294 | /* On success, returns 0, sets skb->_nfct | ctinfo */ |
1295 | static int | |
b2a15a60 | 1296 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 1297 | struct sk_buff *skb, |
9fb9cbb1 YK |
1298 | unsigned int dataoff, |
1299 | u_int16_t l3num, | |
1300 | u_int8_t protonum, | |
2a04aabf JL |
1301 | const struct nf_conntrack_l3proto *l3proto, |
1302 | const struct nf_conntrack_l4proto *l4proto) | |
9fb9cbb1 | 1303 | { |
308ac914 | 1304 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 YK |
1305 | struct nf_conntrack_tuple tuple; |
1306 | struct nf_conntrack_tuple_hash *h; | |
fc09e4a7 | 1307 | enum ip_conntrack_info ctinfo; |
5e8018fc | 1308 | struct nf_conntrack_zone tmp; |
9fb9cbb1 | 1309 | struct nf_conn *ct; |
99f07e91 | 1310 | u32 hash; |
9fb9cbb1 | 1311 | |
bbe735e4 | 1312 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
a31f1adc | 1313 | dataoff, l3num, protonum, net, &tuple, l3proto, |
605dcad6 | 1314 | l4proto)) { |
ccd63c20 | 1315 | pr_debug("Can't get tuple\n"); |
fc09e4a7 | 1316 | return 0; |
9fb9cbb1 YK |
1317 | } |
1318 | ||
1319 | /* look for tuple match */ | |
5e8018fc | 1320 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
1b8c8a9f | 1321 | hash = hash_conntrack_raw(&tuple, net); |
99f07e91 | 1322 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); |
9fb9cbb1 | 1323 | if (!h) { |
b2a15a60 | 1324 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
60b5f8f7 | 1325 | skb, dataoff, hash); |
9fb9cbb1 | 1326 | if (!h) |
fc09e4a7 | 1327 | return 0; |
9fb9cbb1 | 1328 | if (IS_ERR(h)) |
fc09e4a7 | 1329 | return PTR_ERR(h); |
9fb9cbb1 YK |
1330 | } |
1331 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1332 | ||
1333 | /* It exists; we have (non-exclusive) reference. */ | |
1334 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
fc09e4a7 | 1335 | ctinfo = IP_CT_ESTABLISHED_REPLY; |
9fb9cbb1 YK |
1336 | } else { |
1337 | /* Once we've had two way comms, always ESTABLISHED. */ | |
1338 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
ccd63c20 | 1339 | pr_debug("normal packet for %p\n", ct); |
fc09e4a7 | 1340 | ctinfo = IP_CT_ESTABLISHED; |
9fb9cbb1 | 1341 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { |
ccd63c20 | 1342 | pr_debug("related packet for %p\n", ct); |
fc09e4a7 | 1343 | ctinfo = IP_CT_RELATED; |
9fb9cbb1 | 1344 | } else { |
ccd63c20 | 1345 | pr_debug("new packet for %p\n", ct); |
fc09e4a7 | 1346 | ctinfo = IP_CT_NEW; |
9fb9cbb1 | 1347 | } |
9fb9cbb1 | 1348 | } |
fc09e4a7 FW |
1349 | nf_ct_set(skb, ct, ctinfo); |
1350 | return 0; | |
9fb9cbb1 YK |
1351 | } |
1352 | ||
1353 | unsigned int | |
a702a65f AD |
1354 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
1355 | struct sk_buff *skb) | |
9fb9cbb1 | 1356 | { |
b3480fe0 FW |
1357 | const struct nf_conntrack_l3proto *l3proto; |
1358 | const struct nf_conntrack_l4proto *l4proto; | |
97a6ad13 | 1359 | struct nf_conn *ct, *tmpl; |
9fb9cbb1 | 1360 | enum ip_conntrack_info ctinfo; |
2c8503f5 | 1361 | unsigned int *timeouts; |
9fb9cbb1 YK |
1362 | unsigned int dataoff; |
1363 | u_int8_t protonum; | |
9fb9cbb1 YK |
1364 | int ret; |
1365 | ||
97a6ad13 | 1366 | tmpl = nf_ct_get(skb, &ctinfo); |
cc41c84b | 1367 | if (tmpl || ctinfo == IP_CT_UNTRACKED) { |
b2a15a60 | 1368 | /* Previously seen (loopback or untracked)? Ignore. */ |
cc41c84b FW |
1369 | if ((tmpl && !nf_ct_is_template(tmpl)) || |
1370 | ctinfo == IP_CT_UNTRACKED) { | |
b2a15a60 PM |
1371 | NF_CT_STAT_INC_ATOMIC(net, ignore); |
1372 | return NF_ACCEPT; | |
1373 | } | |
a9e419dc | 1374 | skb->_nfct = 0; |
9fb9cbb1 YK |
1375 | } |
1376 | ||
e2361cb9 | 1377 | /* rcu_read_lock()ed by nf_hook_thresh */ |
76108cea | 1378 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 1379 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
1380 | &dataoff, &protonum); |
1381 | if (ret <= 0) { | |
25985edc | 1382 | pr_debug("not prepared to track yet or error occurred\n"); |
0d55af87 AD |
1383 | NF_CT_STAT_INC_ATOMIC(net, error); |
1384 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1385 | ret = -ret; |
1386 | goto out; | |
9fb9cbb1 YK |
1387 | } |
1388 | ||
76108cea | 1389 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
1390 | |
1391 | /* It may be an special packet, error, unclean... | |
1392 | * inverse of the return code tells to the netfilter | |
1393 | * core what to do with the packet. */ | |
74c51a14 | 1394 | if (l4proto->error != NULL) { |
11df4b76 | 1395 | ret = l4proto->error(net, tmpl, skb, dataoff, pf, hooknum); |
74c51a14 | 1396 | if (ret <= 0) { |
0d55af87 AD |
1397 | NF_CT_STAT_INC_ATOMIC(net, error); |
1398 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1399 | ret = -ret; |
1400 | goto out; | |
74c51a14 | 1401 | } |
88ed01d1 | 1402 | /* ICMP[v6] protocol trackers may assign one conntrack. */ |
a9e419dc | 1403 | if (skb->_nfct) |
88ed01d1 | 1404 | goto out; |
9fb9cbb1 | 1405 | } |
08733a0c | 1406 | repeat: |
fc09e4a7 FW |
1407 | ret = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
1408 | l3proto, l4proto); | |
1409 | if (ret < 0) { | |
9fb9cbb1 | 1410 | /* Too stressed to deal. */ |
0d55af87 | 1411 | NF_CT_STAT_INC_ATOMIC(net, drop); |
b2a15a60 PM |
1412 | ret = NF_DROP; |
1413 | goto out; | |
9fb9cbb1 YK |
1414 | } |
1415 | ||
fc09e4a7 FW |
1416 | ct = nf_ct_get(skb, &ctinfo); |
1417 | if (!ct) { | |
1418 | /* Not valid part of a connection */ | |
1419 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
1420 | ret = NF_ACCEPT; | |
1421 | goto out; | |
1422 | } | |
9fb9cbb1 | 1423 | |
60b5f8f7 | 1424 | /* Decide what timeout policy we want to apply to this flow. */ |
84b5ee93 | 1425 | timeouts = nf_ct_timeout_lookup(net, ct, l4proto); |
60b5f8f7 | 1426 | |
d1c1e39d | 1427 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, timeouts); |
ec8d5409 | 1428 | if (ret <= 0) { |
9fb9cbb1 YK |
1429 | /* Invalid: inverse of the return code tells |
1430 | * the netfilter core what to do */ | |
0d53778e | 1431 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
97a6ad13 | 1432 | nf_conntrack_put(&ct->ct_general); |
a9e419dc | 1433 | skb->_nfct = 0; |
0d55af87 | 1434 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
1435 | if (ret == -NF_DROP) |
1436 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
56a62e22 AB |
1437 | /* Special case: TCP tracker reports an attempt to reopen a |
1438 | * closed/aborted connection. We have to go back and create a | |
1439 | * fresh conntrack. | |
1440 | */ | |
1441 | if (ret == -NF_REPEAT) | |
1442 | goto repeat; | |
b2a15a60 PM |
1443 | ret = -ret; |
1444 | goto out; | |
9fb9cbb1 YK |
1445 | } |
1446 | ||
fc09e4a7 FW |
1447 | if (ctinfo == IP_CT_ESTABLISHED_REPLY && |
1448 | !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 1449 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 | 1450 | out: |
56a62e22 AB |
1451 | if (tmpl) |
1452 | nf_ct_put(tmpl); | |
9fb9cbb1 YK |
1453 | |
1454 | return ret; | |
1455 | } | |
13b18339 | 1456 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 1457 | |
5f2b4c90 JE |
1458 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
1459 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 1460 | { |
5f2b4c90 | 1461 | bool ret; |
923f4902 PM |
1462 | |
1463 | rcu_read_lock(); | |
1464 | ret = nf_ct_invert_tuple(inverse, orig, | |
1465 | __nf_ct_l3proto_find(orig->src.l3num), | |
1466 | __nf_ct_l4proto_find(orig->src.l3num, | |
1467 | orig->dst.protonum)); | |
1468 | rcu_read_unlock(); | |
1469 | return ret; | |
9fb9cbb1 | 1470 | } |
13b18339 | 1471 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 1472 | |
5b1158e9 JK |
1473 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
1474 | implicitly racy: see __nf_conntrack_confirm */ | |
1475 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
1476 | const struct nf_conntrack_tuple *newreply) | |
1477 | { | |
1478 | struct nf_conn_help *help = nfct_help(ct); | |
1479 | ||
5b1158e9 | 1480 | /* Should be unconfirmed, so not in hash table yet */ |
44d6e2f2 | 1481 | WARN_ON(nf_ct_is_confirmed(ct)); |
5b1158e9 | 1482 | |
0d53778e | 1483 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 1484 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
1485 | |
1486 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 1487 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 1488 | return; |
ceceae1b | 1489 | |
c52fbb41 | 1490 | rcu_read_lock(); |
b2a15a60 | 1491 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 1492 | rcu_read_unlock(); |
5b1158e9 | 1493 | } |
13b18339 | 1494 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 1495 | |
9fb9cbb1 YK |
1496 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
1497 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
1498 | enum ip_conntrack_info ctinfo, | |
1499 | const struct sk_buff *skb, | |
1500 | unsigned long extra_jiffies, | |
1501 | int do_acct) | |
1502 | { | |
44d6e2f2 | 1503 | WARN_ON(!skb); |
9fb9cbb1 | 1504 | |
997ae831 | 1505 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1506 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1507 | goto acct; | |
997ae831 | 1508 | |
9fb9cbb1 | 1509 | /* If not in hash table, timer will not be active yet */ |
f330a7fd FW |
1510 | if (nf_ct_is_confirmed(ct)) |
1511 | extra_jiffies += nfct_time_stamp; | |
9fb9cbb1 | 1512 | |
f330a7fd | 1513 | ct->timeout = extra_jiffies; |
47d95045 | 1514 | acct: |
ba76738c PNA |
1515 | if (do_acct) |
1516 | nf_ct_acct_update(ct, ctinfo, skb->len); | |
9fb9cbb1 | 1517 | } |
13b18339 | 1518 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1519 | |
ad66713f FW |
1520 | bool nf_ct_kill_acct(struct nf_conn *ct, |
1521 | enum ip_conntrack_info ctinfo, | |
1522 | const struct sk_buff *skb) | |
51091764 | 1523 | { |
ad66713f | 1524 | nf_ct_acct_update(ct, ctinfo, skb->len); |
58401572 | 1525 | |
f330a7fd | 1526 | return nf_ct_delete(ct, 0, 0); |
51091764 | 1527 | } |
ad66713f | 1528 | EXPORT_SYMBOL_GPL(nf_ct_kill_acct); |
51091764 | 1529 | |
c0cd1156 | 1530 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
c1d10adb PNA |
1531 | |
1532 | #include <linux/netfilter/nfnetlink.h> | |
1533 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1534 | #include <linux/mutex.h> |
1535 | ||
c1d10adb PNA |
1536 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1537 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1538 | * or depend on ctnetlink */ | |
fdf70832 | 1539 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1540 | const struct nf_conntrack_tuple *tuple) |
1541 | { | |
bae65be8 DM |
1542 | if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) || |
1543 | nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port)) | |
1544 | goto nla_put_failure; | |
c1d10adb PNA |
1545 | return 0; |
1546 | ||
df6fb868 | 1547 | nla_put_failure: |
c1d10adb PNA |
1548 | return -1; |
1549 | } | |
fdf70832 | 1550 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1551 | |
f73e924c PM |
1552 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1553 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1554 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1555 | }; |
f73e924c | 1556 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1557 | |
fdf70832 | 1558 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1559 | struct nf_conntrack_tuple *t) |
1560 | { | |
df6fb868 | 1561 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1562 | return -EINVAL; |
1563 | ||
77236b6e PM |
1564 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1565 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1566 | |
1567 | return 0; | |
1568 | } | |
fdf70832 | 1569 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1570 | |
1571 | int nf_ct_port_nlattr_tuple_size(void) | |
1572 | { | |
1573 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1574 | } | |
1575 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1576 | #endif |
1577 | ||
9fb9cbb1 | 1578 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
312a0c16 | 1579 | static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb) |
9fb9cbb1 YK |
1580 | { |
1581 | struct nf_conn *ct; | |
1582 | enum ip_conntrack_info ctinfo; | |
1583 | ||
1584 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1585 | ct = nf_ct_get(skb, &ctinfo); | |
1586 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
fb048833 | 1587 | ctinfo = IP_CT_RELATED_REPLY; |
9fb9cbb1 YK |
1588 | else |
1589 | ctinfo = IP_CT_RELATED; | |
1590 | ||
1591 | /* Attach to new skbuff, and increment count */ | |
c74454fa | 1592 | nf_ct_set(nskb, ct, ctinfo); |
cb9c6836 | 1593 | nf_conntrack_get(skb_nfct(nskb)); |
9fb9cbb1 YK |
1594 | } |
1595 | ||
9fb9cbb1 | 1596 | /* Bring out ya dead! */ |
df0933dc | 1597 | static struct nf_conn * |
2843fb69 | 1598 | get_next_corpse(int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1599 | void *data, unsigned int *bucket) |
1600 | { | |
df0933dc PM |
1601 | struct nf_conntrack_tuple_hash *h; |
1602 | struct nf_conn *ct; | |
ea781f19 | 1603 | struct hlist_nulls_node *n; |
93bb0ceb | 1604 | spinlock_t *lockp; |
9fb9cbb1 | 1605 | |
56d52d48 | 1606 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
93bb0ceb JDB |
1607 | lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS]; |
1608 | local_bh_disable(); | |
b16c2919 | 1609 | nf_conntrack_lock(lockp); |
56d52d48 FW |
1610 | if (*bucket < nf_conntrack_htable_size) { |
1611 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) { | |
93bb0ceb JDB |
1612 | if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL) |
1613 | continue; | |
1614 | ct = nf_ct_tuplehash_to_ctrack(h); | |
2843fb69 | 1615 | if (iter(ct, data)) |
93bb0ceb JDB |
1616 | goto found; |
1617 | } | |
df0933dc | 1618 | } |
93bb0ceb JDB |
1619 | spin_unlock(lockp); |
1620 | local_bh_enable(); | |
d93c6258 | 1621 | cond_resched(); |
601e68e1 | 1622 | } |
b7779d06 | 1623 | |
b0feacaa FW |
1624 | return NULL; |
1625 | found: | |
1626 | atomic_inc(&ct->ct_general.use); | |
1627 | spin_unlock(lockp); | |
1628 | local_bh_enable(); | |
1629 | return ct; | |
1630 | } | |
1631 | ||
2843fb69 FW |
1632 | static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), |
1633 | void *data, u32 portid, int report) | |
1634 | { | |
0d02d564 | 1635 | unsigned int bucket = 0, sequence; |
2843fb69 | 1636 | struct nf_conn *ct; |
2843fb69 FW |
1637 | |
1638 | might_sleep(); | |
1639 | ||
0d02d564 FW |
1640 | for (;;) { |
1641 | sequence = read_seqcount_begin(&nf_conntrack_generation); | |
2843fb69 | 1642 | |
0d02d564 FW |
1643 | while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { |
1644 | /* Time to push up daises... */ | |
1645 | ||
1646 | nf_ct_delete(ct, portid, report); | |
1647 | nf_ct_put(ct); | |
1648 | cond_resched(); | |
1649 | } | |
1650 | ||
1651 | if (!read_seqcount_retry(&nf_conntrack_generation, sequence)) | |
1652 | break; | |
1653 | bucket = 0; | |
2843fb69 FW |
1654 | } |
1655 | } | |
1656 | ||
1657 | struct iter_data { | |
1658 | int (*iter)(struct nf_conn *i, void *data); | |
1659 | void *data; | |
1660 | struct net *net; | |
1661 | }; | |
1662 | ||
1663 | static int iter_net_only(struct nf_conn *i, void *data) | |
1664 | { | |
1665 | struct iter_data *d = data; | |
1666 | ||
1667 | if (!net_eq(d->net, nf_ct_net(i))) | |
1668 | return 0; | |
1669 | ||
1670 | return d->iter(i, d->data); | |
1671 | } | |
1672 | ||
b0feacaa FW |
1673 | static void |
1674 | __nf_ct_unconfirmed_destroy(struct net *net) | |
1675 | { | |
1676 | int cpu; | |
1677 | ||
b7779d06 | 1678 | for_each_possible_cpu(cpu) { |
b0feacaa FW |
1679 | struct nf_conntrack_tuple_hash *h; |
1680 | struct hlist_nulls_node *n; | |
1681 | struct ct_pcpu *pcpu; | |
1682 | ||
1683 | pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
b7779d06 JDB |
1684 | |
1685 | spin_lock_bh(&pcpu->lock); | |
1686 | hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) { | |
b0feacaa FW |
1687 | struct nf_conn *ct; |
1688 | ||
b7779d06 | 1689 | ct = nf_ct_tuplehash_to_ctrack(h); |
b0feacaa FW |
1690 | |
1691 | /* we cannot call iter() on unconfirmed list, the | |
1692 | * owning cpu can reallocate ct->ext at any time. | |
1693 | */ | |
1694 | set_bit(IPS_DYING_BIT, &ct->status); | |
b7779d06 JDB |
1695 | } |
1696 | spin_unlock_bh(&pcpu->lock); | |
d93c6258 | 1697 | cond_resched(); |
b7779d06 | 1698 | } |
9fb9cbb1 YK |
1699 | } |
1700 | ||
84657984 FW |
1701 | void nf_ct_unconfirmed_destroy(struct net *net) |
1702 | { | |
1703 | might_sleep(); | |
1704 | ||
1705 | if (atomic_read(&net->ct.count) > 0) { | |
1706 | __nf_ct_unconfirmed_destroy(net); | |
e2a75007 | 1707 | nf_queue_nf_hook_drop(net); |
84657984 FW |
1708 | synchronize_net(); |
1709 | } | |
1710 | } | |
1711 | EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy); | |
1712 | ||
9fd6452d FW |
1713 | void nf_ct_iterate_cleanup_net(struct net *net, |
1714 | int (*iter)(struct nf_conn *i, void *data), | |
1715 | void *data, u32 portid, int report) | |
9fb9cbb1 | 1716 | { |
2843fb69 | 1717 | struct iter_data d; |
9fb9cbb1 | 1718 | |
d93c6258 FW |
1719 | might_sleep(); |
1720 | ||
88b68bc5 FW |
1721 | if (atomic_read(&net->ct.count) == 0) |
1722 | return; | |
1723 | ||
2843fb69 FW |
1724 | d.iter = iter; |
1725 | d.data = data; | |
1726 | d.net = net; | |
1727 | ||
2843fb69 FW |
1728 | nf_ct_iterate_cleanup(iter_net_only, &d, portid, report); |
1729 | } | |
1730 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net); | |
9fb9cbb1 | 1731 | |
2843fb69 FW |
1732 | /** |
1733 | * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table | |
1734 | * @iter: callback to invoke for each conntrack | |
1735 | * @data: data to pass to @iter | |
1736 | * | |
1737 | * Like nf_ct_iterate_cleanup, but first marks conntracks on the | |
1738 | * unconfirmed list as dying (so they will not be inserted into | |
1739 | * main table). | |
7866cc57 FW |
1740 | * |
1741 | * Can only be called in module exit path. | |
2843fb69 FW |
1742 | */ |
1743 | void | |
1744 | nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data) | |
1745 | { | |
1746 | struct net *net; | |
1747 | ||
1748 | rtnl_lock(); | |
1749 | for_each_net(net) { | |
1750 | if (atomic_read(&net->ct.count) == 0) | |
1751 | continue; | |
1752 | __nf_ct_unconfirmed_destroy(net); | |
e2a75007 | 1753 | nf_queue_nf_hook_drop(net); |
9fb9cbb1 | 1754 | } |
2843fb69 FW |
1755 | rtnl_unlock(); |
1756 | ||
7866cc57 FW |
1757 | /* Need to wait for netns cleanup worker to finish, if its |
1758 | * running -- it might have deleted a net namespace from | |
1759 | * the global list, so our __nf_ct_unconfirmed_destroy() might | |
1760 | * not have affected all namespaces. | |
1761 | */ | |
1762 | net_ns_barrier(); | |
1763 | ||
2843fb69 FW |
1764 | /* a conntrack could have been unlinked from unconfirmed list |
1765 | * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy(). | |
1766 | * This makes sure its inserted into conntrack table. | |
1767 | */ | |
1768 | synchronize_net(); | |
1769 | ||
1770 | nf_ct_iterate_cleanup(iter, data, 0, 0); | |
9fb9cbb1 | 1771 | } |
2843fb69 | 1772 | EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy); |
9fb9cbb1 | 1773 | |
274d383b PNA |
1774 | static int kill_all(struct nf_conn *i, void *data) |
1775 | { | |
2843fb69 | 1776 | return net_eq(nf_ct_net(i), data); |
274d383b PNA |
1777 | } |
1778 | ||
d862a662 | 1779 | void nf_ct_free_hashtable(void *hash, unsigned int size) |
9fb9cbb1 | 1780 | { |
d862a662 | 1781 | if (is_vmalloc_addr(hash)) |
9fb9cbb1 YK |
1782 | vfree(hash); |
1783 | else | |
601e68e1 | 1784 | free_pages((unsigned long)hash, |
f205c5e0 | 1785 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1786 | } |
ac565e5f | 1787 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1788 | |
f94161c1 | 1789 | void nf_conntrack_cleanup_start(void) |
9fb9cbb1 | 1790 | { |
b87a2f91 | 1791 | conntrack_gc_work.exiting = true; |
f94161c1 G |
1792 | RCU_INIT_POINTER(ip_ct_attach, NULL); |
1793 | } | |
1794 | ||
1795 | void nf_conntrack_cleanup_end(void) | |
1796 | { | |
1797 | RCU_INIT_POINTER(nf_ct_destroy, NULL); | |
9edd7ca0 | 1798 | |
b87a2f91 | 1799 | cancel_delayed_work_sync(&conntrack_gc_work.dwork); |
56d52d48 FW |
1800 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
1801 | ||
04d87001 | 1802 | nf_conntrack_proto_fini(); |
41d73ec0 | 1803 | nf_conntrack_seqadj_fini(); |
5f69b8f5 | 1804 | nf_conntrack_labels_fini(); |
5e615b22 | 1805 | nf_conntrack_helper_fini(); |
8684094c | 1806 | nf_conntrack_timeout_fini(); |
3fe0f943 | 1807 | nf_conntrack_ecache_fini(); |
73f4001a | 1808 | nf_conntrack_tstamp_fini(); |
b7ff3a1f | 1809 | nf_conntrack_acct_fini(); |
83b4dbe1 | 1810 | nf_conntrack_expect_fini(); |
77571149 FW |
1811 | |
1812 | kmem_cache_destroy(nf_conntrack_cachep); | |
08f6547d | 1813 | } |
9fb9cbb1 | 1814 | |
f94161c1 G |
1815 | /* |
1816 | * Mishearing the voices in his head, our hero wonders how he's | |
1817 | * supposed to kill the mall. | |
1818 | */ | |
1819 | void nf_conntrack_cleanup_net(struct net *net) | |
08f6547d | 1820 | { |
dece40e8 VD |
1821 | LIST_HEAD(single); |
1822 | ||
1823 | list_add(&net->exit_list, &single); | |
1824 | nf_conntrack_cleanup_net_list(&single); | |
1825 | } | |
1826 | ||
1827 | void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list) | |
1828 | { | |
1829 | int busy; | |
1830 | struct net *net; | |
1831 | ||
f94161c1 G |
1832 | /* |
1833 | * This makes sure all current packets have passed through | |
1834 | * netfilter framework. Roll on, two-stage module | |
1835 | * delete... | |
1836 | */ | |
1837 | synchronize_net(); | |
dece40e8 VD |
1838 | i_see_dead_people: |
1839 | busy = 0; | |
1840 | list_for_each_entry(net, net_exit_list, exit_list) { | |
2843fb69 | 1841 | nf_ct_iterate_cleanup(kill_all, net, 0, 0); |
dece40e8 VD |
1842 | if (atomic_read(&net->ct.count) != 0) |
1843 | busy = 1; | |
1844 | } | |
1845 | if (busy) { | |
9fb9cbb1 YK |
1846 | schedule(); |
1847 | goto i_see_dead_people; | |
1848 | } | |
1849 | ||
dece40e8 | 1850 | list_for_each_entry(net, net_exit_list, exit_list) { |
dece40e8 VD |
1851 | nf_conntrack_proto_pernet_fini(net); |
1852 | nf_conntrack_helper_pernet_fini(net); | |
1853 | nf_conntrack_ecache_pernet_fini(net); | |
1854 | nf_conntrack_tstamp_pernet_fini(net); | |
1855 | nf_conntrack_acct_pernet_fini(net); | |
1856 | nf_conntrack_expect_pernet_fini(net); | |
dece40e8 | 1857 | free_percpu(net->ct.stat); |
b7779d06 | 1858 | free_percpu(net->ct.pcpu_lists); |
dece40e8 | 1859 | } |
08f6547d AD |
1860 | } |
1861 | ||
d862a662 | 1862 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) |
9fb9cbb1 | 1863 | { |
ea781f19 ED |
1864 | struct hlist_nulls_head *hash; |
1865 | unsigned int nr_slots, i; | |
1866 | size_t sz; | |
9fb9cbb1 | 1867 | |
9cc1c73a FW |
1868 | if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head))) |
1869 | return NULL; | |
1870 | ||
ea781f19 ED |
1871 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1872 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
9cc1c73a FW |
1873 | |
1874 | if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head))) | |
1875 | return NULL; | |
1876 | ||
ea781f19 ED |
1877 | sz = nr_slots * sizeof(struct hlist_nulls_head); |
1878 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1879 | get_order(sz)); | |
f0ad4621 | 1880 | if (!hash) |
966567b7 | 1881 | hash = vzalloc(sz); |
9fb9cbb1 | 1882 | |
ea781f19 ED |
1883 | if (hash && nulls) |
1884 | for (i = 0; i < nr_slots; i++) | |
1885 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1886 | |
1887 | return hash; | |
1888 | } | |
ac565e5f | 1889 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1890 | |
3183ab89 | 1891 | int nf_conntrack_hash_resize(unsigned int hashsize) |
9fb9cbb1 | 1892 | { |
3183ab89 FW |
1893 | int i, bucket; |
1894 | unsigned int old_size; | |
ea781f19 | 1895 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1896 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1897 | struct nf_conn *ct; |
9fb9cbb1 | 1898 | |
9fb9cbb1 YK |
1899 | if (!hashsize) |
1900 | return -EINVAL; | |
1901 | ||
d862a662 | 1902 | hash = nf_ct_alloc_hashtable(&hashsize, 1); |
9fb9cbb1 YK |
1903 | if (!hash) |
1904 | return -ENOMEM; | |
1905 | ||
3183ab89 FW |
1906 | old_size = nf_conntrack_htable_size; |
1907 | if (old_size == hashsize) { | |
1908 | nf_ct_free_hashtable(hash, hashsize); | |
1909 | return 0; | |
1910 | } | |
1911 | ||
93bb0ceb JDB |
1912 | local_bh_disable(); |
1913 | nf_conntrack_all_lock(); | |
a3efd812 | 1914 | write_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb | 1915 | |
76507f69 PM |
1916 | /* Lookups in the old hash might happen in parallel, which means we |
1917 | * might get false negatives during connection lookup. New connections | |
1918 | * created because of a false negative won't make it into the hash | |
93bb0ceb | 1919 | * though since that required taking the locks. |
76507f69 | 1920 | */ |
93bb0ceb | 1921 | |
56d52d48 FW |
1922 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
1923 | while (!hlist_nulls_empty(&nf_conntrack_hash[i])) { | |
1924 | h = hlist_nulls_entry(nf_conntrack_hash[i].first, | |
1925 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1926 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1927 | hlist_nulls_del_rcu(&h->hnnode); |
1b8c8a9f FW |
1928 | bucket = __hash_conntrack(nf_ct_net(ct), |
1929 | &h->tuple, hashsize); | |
ea781f19 | 1930 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1931 | } |
1932 | } | |
56d52d48 FW |
1933 | old_size = nf_conntrack_htable_size; |
1934 | old_hash = nf_conntrack_hash; | |
9fb9cbb1 | 1935 | |
56d52d48 FW |
1936 | nf_conntrack_hash = hash; |
1937 | nf_conntrack_htable_size = hashsize; | |
93bb0ceb | 1938 | |
a3efd812 | 1939 | write_seqcount_end(&nf_conntrack_generation); |
93bb0ceb JDB |
1940 | nf_conntrack_all_unlock(); |
1941 | local_bh_enable(); | |
9fb9cbb1 | 1942 | |
5e3c61f9 | 1943 | synchronize_net(); |
d862a662 | 1944 | nf_ct_free_hashtable(old_hash, old_size); |
9fb9cbb1 YK |
1945 | return 0; |
1946 | } | |
3183ab89 FW |
1947 | |
1948 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) | |
1949 | { | |
1950 | unsigned int hashsize; | |
1951 | int rc; | |
1952 | ||
1953 | if (current->nsproxy->net_ns != &init_net) | |
1954 | return -EOPNOTSUPP; | |
1955 | ||
1956 | /* On boot, we can set this without any fancy locking. */ | |
1957 | if (!nf_conntrack_htable_size) | |
1958 | return param_set_uint(val, kp); | |
1959 | ||
1960 | rc = kstrtouint(val, 0, &hashsize); | |
1961 | if (rc) | |
1962 | return rc; | |
1963 | ||
1964 | return nf_conntrack_hash_resize(hashsize); | |
1965 | } | |
fae718dd | 1966 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1967 | |
fae718dd | 1968 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1969 | &nf_conntrack_htable_size, 0600); |
1970 | ||
ab71632c | 1971 | static __always_inline unsigned int total_extension_size(void) |
b3a5db10 FW |
1972 | { |
1973 | /* remember to add new extensions below */ | |
1974 | BUILD_BUG_ON(NF_CT_EXT_NUM > 9); | |
1975 | ||
1976 | return sizeof(struct nf_ct_ext) + | |
1977 | sizeof(struct nf_conn_help) | |
1978 | #if IS_ENABLED(CONFIG_NF_NAT) | |
1979 | + sizeof(struct nf_conn_nat) | |
1980 | #endif | |
1981 | + sizeof(struct nf_conn_seqadj) | |
1982 | + sizeof(struct nf_conn_acct) | |
1983 | #ifdef CONFIG_NF_CONNTRACK_EVENTS | |
1984 | + sizeof(struct nf_conntrack_ecache) | |
1985 | #endif | |
1986 | #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP | |
1987 | + sizeof(struct nf_conn_tstamp) | |
1988 | #endif | |
1989 | #ifdef CONFIG_NF_CONNTRACK_TIMEOUT | |
1990 | + sizeof(struct nf_conn_timeout) | |
1991 | #endif | |
1992 | #ifdef CONFIG_NF_CONNTRACK_LABELS | |
1993 | + sizeof(struct nf_conn_labels) | |
1994 | #endif | |
1995 | #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY) | |
1996 | + sizeof(struct nf_conn_synproxy) | |
1997 | #endif | |
1998 | ; | |
1999 | }; | |
2000 | ||
f94161c1 | 2001 | int nf_conntrack_init_start(void) |
9fb9cbb1 | 2002 | { |
f205c5e0 | 2003 | int max_factor = 8; |
0c5366b3 | 2004 | int ret = -ENOMEM; |
cc41c84b | 2005 | int i; |
93bb0ceb | 2006 | |
b3a5db10 FW |
2007 | /* struct nf_ct_ext uses u8 to store offsets/size */ |
2008 | BUILD_BUG_ON(total_extension_size() > 255u); | |
2009 | ||
a3efd812 FW |
2010 | seqcount_init(&nf_conntrack_generation); |
2011 | ||
d5d20912 | 2012 | for (i = 0; i < CONNTRACK_LOCKS; i++) |
93bb0ceb | 2013 | spin_lock_init(&nf_conntrack_locks[i]); |
9fb9cbb1 | 2014 | |
9fb9cbb1 | 2015 | if (!nf_conntrack_htable_size) { |
88eab472 ML |
2016 | /* Idea from tcp.c: use 1/16384 of memory. |
2017 | * On i386: 32MB machine has 512 buckets. | |
2018 | * >= 1GB machines have 16384 buckets. | |
2019 | * >= 4GB machines have 65536 buckets. | |
2020 | */ | |
9fb9cbb1 | 2021 | nf_conntrack_htable_size |
4481374c | 2022 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 2023 | / sizeof(struct hlist_head)); |
88eab472 ML |
2024 | if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE))) |
2025 | nf_conntrack_htable_size = 65536; | |
2026 | else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
f205c5e0 PM |
2027 | nf_conntrack_htable_size = 16384; |
2028 | if (nf_conntrack_htable_size < 32) | |
2029 | nf_conntrack_htable_size = 32; | |
2030 | ||
2031 | /* Use a max. factor of four by default to get the same max as | |
2032 | * with the old struct list_heads. When a table size is given | |
2033 | * we use the old value of 8 to avoid reducing the max. | |
2034 | * entries. */ | |
2035 | max_factor = 4; | |
9fb9cbb1 | 2036 | } |
56d52d48 FW |
2037 | |
2038 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1); | |
2039 | if (!nf_conntrack_hash) | |
2040 | return -ENOMEM; | |
2041 | ||
f205c5e0 | 2042 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 2043 | |
0c5366b3 | 2044 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
a9e419dc FW |
2045 | sizeof(struct nf_conn), |
2046 | NFCT_INFOMASK + 1, | |
5f0d5a3a | 2047 | SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL); |
0c5366b3 FW |
2048 | if (!nf_conntrack_cachep) |
2049 | goto err_cachep; | |
2050 | ||
654d0fbd | 2051 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
2052 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
2053 | nf_conntrack_max); | |
83b4dbe1 G |
2054 | |
2055 | ret = nf_conntrack_expect_init(); | |
2056 | if (ret < 0) | |
2057 | goto err_expect; | |
2058 | ||
b7ff3a1f G |
2059 | ret = nf_conntrack_acct_init(); |
2060 | if (ret < 0) | |
2061 | goto err_acct; | |
2062 | ||
73f4001a G |
2063 | ret = nf_conntrack_tstamp_init(); |
2064 | if (ret < 0) | |
2065 | goto err_tstamp; | |
2066 | ||
3fe0f943 G |
2067 | ret = nf_conntrack_ecache_init(); |
2068 | if (ret < 0) | |
2069 | goto err_ecache; | |
2070 | ||
8684094c G |
2071 | ret = nf_conntrack_timeout_init(); |
2072 | if (ret < 0) | |
2073 | goto err_timeout; | |
2074 | ||
5e615b22 G |
2075 | ret = nf_conntrack_helper_init(); |
2076 | if (ret < 0) | |
2077 | goto err_helper; | |
2078 | ||
5f69b8f5 G |
2079 | ret = nf_conntrack_labels_init(); |
2080 | if (ret < 0) | |
2081 | goto err_labels; | |
2082 | ||
41d73ec0 PM |
2083 | ret = nf_conntrack_seqadj_init(); |
2084 | if (ret < 0) | |
2085 | goto err_seqadj; | |
2086 | ||
04d87001 G |
2087 | ret = nf_conntrack_proto_init(); |
2088 | if (ret < 0) | |
2089 | goto err_proto; | |
2090 | ||
b87a2f91 | 2091 | conntrack_gc_work_init(&conntrack_gc_work); |
e5072053 | 2092 | queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, HZ); |
b87a2f91 | 2093 | |
08f6547d AD |
2094 | return 0; |
2095 | ||
04d87001 | 2096 | err_proto: |
41d73ec0 PM |
2097 | nf_conntrack_seqadj_fini(); |
2098 | err_seqadj: | |
04d87001 | 2099 | nf_conntrack_labels_fini(); |
5f69b8f5 G |
2100 | err_labels: |
2101 | nf_conntrack_helper_fini(); | |
5e615b22 G |
2102 | err_helper: |
2103 | nf_conntrack_timeout_fini(); | |
8684094c G |
2104 | err_timeout: |
2105 | nf_conntrack_ecache_fini(); | |
3fe0f943 G |
2106 | err_ecache: |
2107 | nf_conntrack_tstamp_fini(); | |
73f4001a G |
2108 | err_tstamp: |
2109 | nf_conntrack_acct_fini(); | |
b7ff3a1f G |
2110 | err_acct: |
2111 | nf_conntrack_expect_fini(); | |
83b4dbe1 | 2112 | err_expect: |
0c5366b3 FW |
2113 | kmem_cache_destroy(nf_conntrack_cachep); |
2114 | err_cachep: | |
56d52d48 | 2115 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
08f6547d AD |
2116 | return ret; |
2117 | } | |
2118 | ||
f94161c1 G |
2119 | void nf_conntrack_init_end(void) |
2120 | { | |
2121 | /* For use by REJECT target */ | |
2122 | RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); | |
2123 | RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); | |
f94161c1 G |
2124 | } |
2125 | ||
8cc20198 ED |
2126 | /* |
2127 | * We need to use special "null" values, not used in hash table | |
2128 | */ | |
2129 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
2130 | #define DYING_NULLS_VAL ((1<<30)+1) | |
252b3e8c | 2131 | #define TEMPLATE_NULLS_VAL ((1<<30)+2) |
8cc20198 | 2132 | |
f94161c1 | 2133 | int nf_conntrack_init_net(struct net *net) |
08f6547d | 2134 | { |
b7779d06 JDB |
2135 | int ret = -ENOMEM; |
2136 | int cpu; | |
ceceae1b | 2137 | |
cc41c84b | 2138 | BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER); |
08f6547d | 2139 | atomic_set(&net->ct.count, 0); |
b7779d06 JDB |
2140 | |
2141 | net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu); | |
2142 | if (!net->ct.pcpu_lists) | |
08f6547d | 2143 | goto err_stat; |
b7779d06 JDB |
2144 | |
2145 | for_each_possible_cpu(cpu) { | |
2146 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
2147 | ||
2148 | spin_lock_init(&pcpu->lock); | |
2149 | INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL); | |
2150 | INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL); | |
08f6547d | 2151 | } |
5b3501fa | 2152 | |
b7779d06 JDB |
2153 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
2154 | if (!net->ct.stat) | |
2155 | goto err_pcpu_lists; | |
2156 | ||
83b4dbe1 | 2157 | ret = nf_conntrack_expect_pernet_init(net); |
08f6547d AD |
2158 | if (ret < 0) |
2159 | goto err_expect; | |
b7ff3a1f | 2160 | ret = nf_conntrack_acct_pernet_init(net); |
58401572 | 2161 | if (ret < 0) |
08f6547d | 2162 | goto err_acct; |
73f4001a | 2163 | ret = nf_conntrack_tstamp_pernet_init(net); |
a992ca2a PNA |
2164 | if (ret < 0) |
2165 | goto err_tstamp; | |
3fe0f943 | 2166 | ret = nf_conntrack_ecache_pernet_init(net); |
a0891aa6 PNA |
2167 | if (ret < 0) |
2168 | goto err_ecache; | |
5e615b22 | 2169 | ret = nf_conntrack_helper_pernet_init(net); |
a9006892 EL |
2170 | if (ret < 0) |
2171 | goto err_helper; | |
04d87001 | 2172 | ret = nf_conntrack_proto_pernet_init(net); |
f94161c1 G |
2173 | if (ret < 0) |
2174 | goto err_proto; | |
08f6547d | 2175 | return 0; |
c539f017 | 2176 | |
f94161c1 | 2177 | err_proto: |
5e615b22 | 2178 | nf_conntrack_helper_pernet_fini(net); |
a9006892 | 2179 | err_helper: |
3fe0f943 | 2180 | nf_conntrack_ecache_pernet_fini(net); |
a0891aa6 | 2181 | err_ecache: |
73f4001a | 2182 | nf_conntrack_tstamp_pernet_fini(net); |
a992ca2a | 2183 | err_tstamp: |
b7ff3a1f | 2184 | nf_conntrack_acct_pernet_fini(net); |
08f6547d | 2185 | err_acct: |
83b4dbe1 | 2186 | nf_conntrack_expect_pernet_fini(net); |
08f6547d | 2187 | err_expect: |
0d55af87 | 2188 | free_percpu(net->ct.stat); |
b7779d06 JDB |
2189 | err_pcpu_lists: |
2190 | free_percpu(net->ct.pcpu_lists); | |
0d55af87 | 2191 | err_stat: |
08f6547d AD |
2192 | return ret; |
2193 | } |