]>
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; | |
e0df8cae | 79 | long next_gc_run; |
b87a2f91 FW |
80 | }; |
81 | ||
0c5366b3 | 82 | static __read_mostly struct kmem_cache *nf_conntrack_cachep; |
b16c2919 | 83 | static __read_mostly spinlock_t nf_conntrack_locks_all_lock; |
70d72b7e | 84 | static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock); |
b16c2919 SL |
85 | static __read_mostly bool nf_conntrack_locks_all; |
86 | ||
e0df8cae | 87 | /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */ |
b87a2f91 | 88 | #define GC_MAX_BUCKETS_DIV 64u |
e0df8cae FW |
89 | /* upper bound of scan intervals */ |
90 | #define GC_INTERVAL_MAX (2 * HZ) | |
91 | /* maximum conntracks to evict per gc run */ | |
b87a2f91 FW |
92 | #define GC_MAX_EVICTS 256u |
93 | ||
94 | static struct conntrack_gc_work conntrack_gc_work; | |
95 | ||
b16c2919 SL |
96 | void nf_conntrack_lock(spinlock_t *lock) __acquires(lock) |
97 | { | |
98 | spin_lock(lock); | |
99 | while (unlikely(nf_conntrack_locks_all)) { | |
100 | spin_unlock(lock); | |
b316ff78 PZ |
101 | |
102 | /* | |
103 | * Order the 'nf_conntrack_locks_all' load vs. the | |
104 | * spin_unlock_wait() loads below, to ensure | |
105 | * that 'nf_conntrack_locks_all_lock' is indeed held: | |
106 | */ | |
107 | smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */ | |
e39365be | 108 | spin_unlock_wait(&nf_conntrack_locks_all_lock); |
b16c2919 SL |
109 | spin_lock(lock); |
110 | } | |
111 | } | |
112 | EXPORT_SYMBOL_GPL(nf_conntrack_lock); | |
113 | ||
93bb0ceb JDB |
114 | static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2) |
115 | { | |
116 | h1 %= CONNTRACK_LOCKS; | |
117 | h2 %= CONNTRACK_LOCKS; | |
118 | spin_unlock(&nf_conntrack_locks[h1]); | |
119 | if (h1 != h2) | |
120 | spin_unlock(&nf_conntrack_locks[h2]); | |
121 | } | |
122 | ||
123 | /* return true if we need to recompute hashes (in case hash table was resized) */ | |
124 | static bool nf_conntrack_double_lock(struct net *net, unsigned int h1, | |
125 | unsigned int h2, unsigned int sequence) | |
126 | { | |
127 | h1 %= CONNTRACK_LOCKS; | |
128 | h2 %= CONNTRACK_LOCKS; | |
129 | if (h1 <= h2) { | |
b16c2919 | 130 | nf_conntrack_lock(&nf_conntrack_locks[h1]); |
93bb0ceb JDB |
131 | if (h1 != h2) |
132 | spin_lock_nested(&nf_conntrack_locks[h2], | |
133 | SINGLE_DEPTH_NESTING); | |
134 | } else { | |
b16c2919 | 135 | nf_conntrack_lock(&nf_conntrack_locks[h2]); |
93bb0ceb JDB |
136 | spin_lock_nested(&nf_conntrack_locks[h1], |
137 | SINGLE_DEPTH_NESTING); | |
138 | } | |
a3efd812 | 139 | if (read_seqcount_retry(&nf_conntrack_generation, sequence)) { |
93bb0ceb JDB |
140 | nf_conntrack_double_unlock(h1, h2); |
141 | return true; | |
142 | } | |
143 | return false; | |
144 | } | |
145 | ||
146 | static void nf_conntrack_all_lock(void) | |
147 | { | |
148 | int i; | |
149 | ||
b16c2919 SL |
150 | spin_lock(&nf_conntrack_locks_all_lock); |
151 | nf_conntrack_locks_all = true; | |
152 | ||
b316ff78 PZ |
153 | /* |
154 | * Order the above store of 'nf_conntrack_locks_all' against | |
155 | * the spin_unlock_wait() loads below, such that if | |
156 | * nf_conntrack_lock() observes 'nf_conntrack_locks_all' | |
157 | * we must observe nf_conntrack_locks[] held: | |
158 | */ | |
159 | smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */ | |
160 | ||
b16c2919 | 161 | for (i = 0; i < CONNTRACK_LOCKS; i++) { |
e39365be | 162 | spin_unlock_wait(&nf_conntrack_locks[i]); |
b16c2919 | 163 | } |
93bb0ceb JDB |
164 | } |
165 | ||
166 | static void nf_conntrack_all_unlock(void) | |
167 | { | |
b316ff78 PZ |
168 | /* |
169 | * All prior stores must be complete before we clear | |
170 | * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock() | |
171 | * might observe the false value but not the entire | |
172 | * critical section: | |
173 | */ | |
174 | smp_store_release(&nf_conntrack_locks_all, false); | |
b16c2919 | 175 | spin_unlock(&nf_conntrack_locks_all_lock); |
93bb0ceb JDB |
176 | } |
177 | ||
e2b7606c | 178 | unsigned int nf_conntrack_htable_size __read_mostly; |
2567c4ea PNA |
179 | EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); |
180 | ||
e478075c | 181 | unsigned int nf_conntrack_max __read_mostly; |
92e47ba8 | 182 | seqcount_t nf_conntrack_generation __read_mostly; |
13b18339 | 183 | |
b3c5163f ED |
184 | DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked); |
185 | EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked); | |
13b18339 | 186 | |
141658fb | 187 | static unsigned int nf_conntrack_hash_rnd __read_mostly; |
9fb9cbb1 | 188 | |
1b8c8a9f FW |
189 | static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, |
190 | const struct net *net) | |
9fb9cbb1 | 191 | { |
0794935e | 192 | unsigned int n; |
1b8c8a9f | 193 | u32 seed; |
0794935e | 194 | |
141658fb FW |
195 | get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd)); |
196 | ||
0794935e PM |
197 | /* The direction must be ignored, so we hash everything up to the |
198 | * destination ports (which is a multiple of 4) and treat the last | |
199 | * three bytes manually. | |
200 | */ | |
1b8c8a9f | 201 | seed = nf_conntrack_hash_rnd ^ net_hash_mix(net); |
0794935e | 202 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); |
1b8c8a9f | 203 | return jhash2((u32 *)tuple, n, seed ^ |
99f07e91 CG |
204 | (((__force __u16)tuple->dst.u.all << 16) | |
205 | tuple->dst.protonum)); | |
206 | } | |
207 | ||
56d52d48 | 208 | static u32 scale_hash(u32 hash) |
99f07e91 | 209 | { |
56d52d48 | 210 | return reciprocal_scale(hash, nf_conntrack_htable_size); |
99f07e91 | 211 | } |
0794935e | 212 | |
1b8c8a9f FW |
213 | static u32 __hash_conntrack(const struct net *net, |
214 | const struct nf_conntrack_tuple *tuple, | |
215 | unsigned int size) | |
99f07e91 | 216 | { |
1b8c8a9f | 217 | return reciprocal_scale(hash_conntrack_raw(tuple, net), size); |
9fb9cbb1 YK |
218 | } |
219 | ||
1b8c8a9f FW |
220 | static u32 hash_conntrack(const struct net *net, |
221 | const struct nf_conntrack_tuple *tuple) | |
9fb9cbb1 | 222 | { |
56d52d48 | 223 | return scale_hash(hash_conntrack_raw(tuple, net)); |
9fb9cbb1 YK |
224 | } |
225 | ||
5f2b4c90 | 226 | bool |
9fb9cbb1 YK |
227 | nf_ct_get_tuple(const struct sk_buff *skb, |
228 | unsigned int nhoff, | |
229 | unsigned int dataoff, | |
230 | u_int16_t l3num, | |
231 | u_int8_t protonum, | |
a31f1adc | 232 | struct net *net, |
9fb9cbb1 YK |
233 | struct nf_conntrack_tuple *tuple, |
234 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 235 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 236 | { |
443a70d5 | 237 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
238 | |
239 | tuple->src.l3num = l3num; | |
240 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 241 | return false; |
9fb9cbb1 YK |
242 | |
243 | tuple->dst.protonum = protonum; | |
244 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
245 | ||
a31f1adc | 246 | return l4proto->pkt_to_tuple(skb, dataoff, net, tuple); |
9fb9cbb1 | 247 | } |
13b18339 | 248 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 249 | |
5f2b4c90 | 250 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
a31f1adc EB |
251 | u_int16_t l3num, |
252 | struct net *net, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
253 | { |
254 | struct nf_conntrack_l3proto *l3proto; | |
255 | struct nf_conntrack_l4proto *l4proto; | |
256 | unsigned int protoff; | |
257 | u_int8_t protonum; | |
258 | int ret; | |
259 | ||
260 | rcu_read_lock(); | |
261 | ||
262 | l3proto = __nf_ct_l3proto_find(l3num); | |
263 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
264 | if (ret != NF_ACCEPT) { | |
265 | rcu_read_unlock(); | |
5f2b4c90 | 266 | return false; |
e2a3123f YK |
267 | } |
268 | ||
269 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
270 | ||
a31f1adc | 271 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple, |
e2a3123f YK |
272 | l3proto, l4proto); |
273 | ||
274 | rcu_read_unlock(); | |
275 | return ret; | |
276 | } | |
277 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
278 | ||
5f2b4c90 | 279 | bool |
9fb9cbb1 YK |
280 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
281 | const struct nf_conntrack_tuple *orig, | |
282 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 283 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 284 | { |
443a70d5 | 285 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
286 | |
287 | inverse->src.l3num = orig->src.l3num; | |
288 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 289 | return false; |
9fb9cbb1 YK |
290 | |
291 | inverse->dst.dir = !orig->dst.dir; | |
292 | ||
293 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 294 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 295 | } |
13b18339 | 296 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 297 | |
9fb9cbb1 YK |
298 | static void |
299 | clean_from_lists(struct nf_conn *ct) | |
300 | { | |
0d53778e | 301 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
302 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
303 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
304 | |
305 | /* Destroy all pending expectations */ | |
c1d10adb | 306 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
307 | } |
308 | ||
b7779d06 JDB |
309 | /* must be called with local_bh_disable */ |
310 | static void nf_ct_add_to_dying_list(struct nf_conn *ct) | |
311 | { | |
312 | struct ct_pcpu *pcpu; | |
313 | ||
314 | /* add this conntrack to the (per cpu) dying list */ | |
315 | ct->cpu = smp_processor_id(); | |
316 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
317 | ||
318 | spin_lock(&pcpu->lock); | |
319 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
320 | &pcpu->dying); | |
321 | spin_unlock(&pcpu->lock); | |
322 | } | |
323 | ||
324 | /* must be called with local_bh_disable */ | |
325 | static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct) | |
326 | { | |
327 | struct ct_pcpu *pcpu; | |
328 | ||
329 | /* add this conntrack to the (per cpu) unconfirmed list */ | |
330 | ct->cpu = smp_processor_id(); | |
331 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
332 | ||
333 | spin_lock(&pcpu->lock); | |
334 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
335 | &pcpu->unconfirmed); | |
336 | spin_unlock(&pcpu->lock); | |
337 | } | |
338 | ||
339 | /* must be called with local_bh_disable */ | |
340 | static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct) | |
341 | { | |
342 | struct ct_pcpu *pcpu; | |
343 | ||
344 | /* We overload first tuple to link into unconfirmed or dying list.*/ | |
345 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
346 | ||
347 | spin_lock(&pcpu->lock); | |
348 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); | |
349 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
350 | spin_unlock(&pcpu->lock); | |
351 | } | |
352 | ||
0838aa7f | 353 | /* Released via destroy_conntrack() */ |
308ac914 DB |
354 | struct nf_conn *nf_ct_tmpl_alloc(struct net *net, |
355 | const struct nf_conntrack_zone *zone, | |
356 | gfp_t flags) | |
0838aa7f PNA |
357 | { |
358 | struct nf_conn *tmpl; | |
359 | ||
f58e5aa7 | 360 | tmpl = kzalloc(sizeof(*tmpl), flags); |
0838aa7f PNA |
361 | if (tmpl == NULL) |
362 | return NULL; | |
363 | ||
364 | tmpl->status = IPS_TEMPLATE; | |
365 | write_pnet(&tmpl->ct_net, net); | |
6c8dee98 | 366 | nf_ct_zone_add(tmpl, zone); |
0838aa7f PNA |
367 | atomic_set(&tmpl->ct_general.use, 0); |
368 | ||
369 | return tmpl; | |
0838aa7f PNA |
370 | } |
371 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc); | |
372 | ||
9cf94eab | 373 | void nf_ct_tmpl_free(struct nf_conn *tmpl) |
0838aa7f PNA |
374 | { |
375 | nf_ct_ext_destroy(tmpl); | |
376 | nf_ct_ext_free(tmpl); | |
377 | kfree(tmpl); | |
378 | } | |
9cf94eab | 379 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_free); |
0838aa7f | 380 | |
9fb9cbb1 YK |
381 | static void |
382 | destroy_conntrack(struct nf_conntrack *nfct) | |
383 | { | |
384 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 385 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 386 | |
0d53778e | 387 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 | 388 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
9fb9cbb1 | 389 | |
0838aa7f PNA |
390 | if (unlikely(nf_ct_is_template(ct))) { |
391 | nf_ct_tmpl_free(ct); | |
392 | return; | |
393 | } | |
923f4902 | 394 | rcu_read_lock(); |
5e8fbe2a | 395 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
4b4ceb9d | 396 | if (l4proto->destroy) |
605dcad6 | 397 | l4proto->destroy(ct); |
9fb9cbb1 | 398 | |
982d9a9c | 399 | rcu_read_unlock(); |
9fb9cbb1 | 400 | |
ca7433df | 401 | local_bh_disable(); |
9fb9cbb1 YK |
402 | /* Expectations will have been removed in clean_from_lists, |
403 | * except TFTP can create an expectation on the first packet, | |
404 | * before connection is in the list, so we need to clean here, | |
ca7433df JDB |
405 | * too. |
406 | */ | |
c1d10adb | 407 | nf_ct_remove_expectations(ct); |
9fb9cbb1 | 408 | |
b7779d06 | 409 | nf_ct_del_from_dying_or_unconfirmed_list(ct); |
9fb9cbb1 | 410 | |
ca7433df | 411 | local_bh_enable(); |
9fb9cbb1 YK |
412 | |
413 | if (ct->master) | |
414 | nf_ct_put(ct->master); | |
415 | ||
0d53778e | 416 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
417 | nf_conntrack_free(ct); |
418 | } | |
419 | ||
02982c27 | 420 | static void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 421 | { |
0d55af87 | 422 | struct net *net = nf_ct_net(ct); |
93bb0ceb | 423 | unsigned int hash, reply_hash; |
93bb0ceb | 424 | unsigned int sequence; |
9fb9cbb1 | 425 | |
9858a3ae | 426 | nf_ct_helper_destroy(ct); |
93bb0ceb JDB |
427 | |
428 | local_bh_disable(); | |
429 | do { | |
a3efd812 | 430 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 431 | hash = hash_conntrack(net, |
93bb0ceb | 432 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 433 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
434 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
435 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
436 | ||
9fb9cbb1 | 437 | clean_from_lists(ct); |
93bb0ceb JDB |
438 | nf_conntrack_double_unlock(hash, reply_hash); |
439 | ||
b7779d06 | 440 | nf_ct_add_to_dying_list(ct); |
93bb0ceb | 441 | |
93bb0ceb | 442 | local_bh_enable(); |
dd7669a9 | 443 | } |
dd7669a9 | 444 | |
02982c27 | 445 | bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report) |
dd7669a9 | 446 | { |
a992ca2a PNA |
447 | struct nf_conn_tstamp *tstamp; |
448 | ||
f330a7fd FW |
449 | if (test_and_set_bit(IPS_DYING_BIT, &ct->status)) |
450 | return false; | |
451 | ||
a992ca2a PNA |
452 | tstamp = nf_conn_tstamp_find(ct); |
453 | if (tstamp && tstamp->stop == 0) | |
d2de875c | 454 | tstamp->stop = ktime_get_real_ns(); |
dd7669a9 | 455 | |
9500507c FW |
456 | if (nf_conntrack_event_report(IPCT_DESTROY, ct, |
457 | portid, report) < 0) { | |
f330a7fd FW |
458 | /* destroy event was not delivered. nf_ct_put will |
459 | * be done by event cache worker on redelivery. | |
460 | */ | |
dd7669a9 | 461 | nf_ct_delete_from_lists(ct); |
9500507c | 462 | nf_conntrack_ecache_delayed_work(nf_ct_net(ct)); |
02982c27 | 463 | return false; |
dd7669a9 | 464 | } |
9500507c FW |
465 | |
466 | nf_conntrack_ecache_work(nf_ct_net(ct)); | |
dd7669a9 | 467 | nf_ct_delete_from_lists(ct); |
9fb9cbb1 | 468 | nf_ct_put(ct); |
02982c27 FW |
469 | return true; |
470 | } | |
471 | EXPORT_SYMBOL_GPL(nf_ct_delete); | |
472 | ||
c6825c09 AV |
473 | static inline bool |
474 | nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, | |
308ac914 | 475 | const struct nf_conntrack_tuple *tuple, |
e0c7d472 FW |
476 | const struct nf_conntrack_zone *zone, |
477 | const struct net *net) | |
c6825c09 AV |
478 | { |
479 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
480 | ||
481 | /* A conntrack can be recreated with the equal tuple, | |
482 | * so we need to check that the conntrack is confirmed | |
483 | */ | |
484 | return nf_ct_tuple_equal(tuple, &h->tuple) && | |
deedb590 | 485 | nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) && |
e0c7d472 FW |
486 | nf_ct_is_confirmed(ct) && |
487 | net_eq(net, nf_ct_net(ct)); | |
c6825c09 AV |
488 | } |
489 | ||
f330a7fd FW |
490 | /* caller must hold rcu readlock and none of the nf_conntrack_locks */ |
491 | static void nf_ct_gc_expired(struct nf_conn *ct) | |
492 | { | |
493 | if (!atomic_inc_not_zero(&ct->ct_general.use)) | |
494 | return; | |
495 | ||
496 | if (nf_ct_should_gc(ct)) | |
497 | nf_ct_kill(ct); | |
498 | ||
499 | nf_ct_put(ct); | |
500 | } | |
501 | ||
ea781f19 ED |
502 | /* |
503 | * Warning : | |
504 | * - Caller must take a reference on returned object | |
505 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
ea781f19 | 506 | */ |
99f07e91 | 507 | static struct nf_conntrack_tuple_hash * |
308ac914 | 508 | ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 509 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
510 | { |
511 | struct nf_conntrack_tuple_hash *h; | |
5e3c61f9 | 512 | struct hlist_nulls_head *ct_hash; |
ea781f19 | 513 | struct hlist_nulls_node *n; |
92e47ba8 | 514 | unsigned int bucket, hsize; |
9fb9cbb1 | 515 | |
ea781f19 | 516 | begin: |
92e47ba8 LZ |
517 | nf_conntrack_get_ht(&ct_hash, &hsize); |
518 | bucket = reciprocal_scale(hash, hsize); | |
5e3c61f9 FW |
519 | |
520 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) { | |
f330a7fd FW |
521 | struct nf_conn *ct; |
522 | ||
523 | ct = nf_ct_tuplehash_to_ctrack(h); | |
524 | if (nf_ct_is_expired(ct)) { | |
525 | nf_ct_gc_expired(ct); | |
526 | continue; | |
527 | } | |
528 | ||
529 | if (nf_ct_is_dying(ct)) | |
530 | continue; | |
531 | ||
8e8118f8 | 532 | if (nf_ct_key_equal(h, tuple, zone, net)) |
9fb9cbb1 | 533 | return h; |
9fb9cbb1 | 534 | } |
ea781f19 ED |
535 | /* |
536 | * if the nulls value we got at the end of this lookup is | |
537 | * not the expected one, we must restart lookup. | |
538 | * We probably met an item that was moved to another chain. | |
539 | */ | |
99f07e91 | 540 | if (get_nulls_value(n) != bucket) { |
2cf12348 | 541 | NF_CT_STAT_INC_ATOMIC(net, search_restart); |
ea781f19 | 542 | goto begin; |
af740b2c | 543 | } |
9fb9cbb1 YK |
544 | |
545 | return NULL; | |
546 | } | |
99f07e91 | 547 | |
9fb9cbb1 | 548 | /* Find a connection corresponding to a tuple. */ |
99f07e91 | 549 | static struct nf_conntrack_tuple_hash * |
308ac914 | 550 | __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 551 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
552 | { |
553 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 554 | struct nf_conn *ct; |
9fb9cbb1 | 555 | |
76507f69 | 556 | rcu_read_lock(); |
ea781f19 | 557 | begin: |
99f07e91 | 558 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
559 | if (h) { |
560 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
561 | if (unlikely(nf_ct_is_dying(ct) || |
562 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 563 | h = NULL; |
ea781f19 | 564 | else { |
e0c7d472 | 565 | if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) { |
ea781f19 ED |
566 | nf_ct_put(ct); |
567 | goto begin; | |
568 | } | |
569 | } | |
76507f69 PM |
570 | } |
571 | rcu_read_unlock(); | |
9fb9cbb1 YK |
572 | |
573 | return h; | |
574 | } | |
99f07e91 CG |
575 | |
576 | struct nf_conntrack_tuple_hash * | |
308ac914 | 577 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 CG |
578 | const struct nf_conntrack_tuple *tuple) |
579 | { | |
580 | return __nf_conntrack_find_get(net, zone, tuple, | |
1b8c8a9f | 581 | hash_conntrack_raw(tuple, net)); |
99f07e91 | 582 | } |
13b18339 | 583 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 584 | |
c1d10adb PNA |
585 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
586 | unsigned int hash, | |
b476b72a | 587 | unsigned int reply_hash) |
c1d10adb | 588 | { |
ea781f19 | 589 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
56d52d48 | 590 | &nf_conntrack_hash[hash]); |
ea781f19 | 591 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
56d52d48 | 592 | &nf_conntrack_hash[reply_hash]); |
c1d10adb PNA |
593 | } |
594 | ||
7d367e06 JK |
595 | int |
596 | nf_conntrack_hash_check_insert(struct nf_conn *ct) | |
c1d10adb | 597 | { |
308ac914 | 598 | const struct nf_conntrack_zone *zone; |
d696c7bd | 599 | struct net *net = nf_ct_net(ct); |
b476b72a | 600 | unsigned int hash, reply_hash; |
7d367e06 JK |
601 | struct nf_conntrack_tuple_hash *h; |
602 | struct hlist_nulls_node *n; | |
93bb0ceb | 603 | unsigned int sequence; |
c1d10adb | 604 | |
5d0aa2cc | 605 | zone = nf_ct_zone(ct); |
7d367e06 | 606 | |
93bb0ceb JDB |
607 | local_bh_disable(); |
608 | do { | |
a3efd812 | 609 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 610 | hash = hash_conntrack(net, |
93bb0ceb | 611 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 612 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
613 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
614 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
7d367e06 JK |
615 | |
616 | /* See if there's one in the list already, including reverse */ | |
56d52d48 | 617 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 618 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 619 | zone, net)) |
7d367e06 | 620 | goto out; |
86804348 | 621 | |
56d52d48 | 622 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 623 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 624 | zone, net)) |
7d367e06 | 625 | goto out; |
c1d10adb | 626 | |
e53376be PNA |
627 | smp_wmb(); |
628 | /* The caller holds a reference to this object */ | |
629 | atomic_set(&ct->ct_general.use, 2); | |
b476b72a | 630 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 631 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 632 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 633 | local_bh_enable(); |
7d367e06 JK |
634 | return 0; |
635 | ||
636 | out: | |
93bb0ceb | 637 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 638 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 639 | local_bh_enable(); |
7d367e06 | 640 | return -EEXIST; |
c1d10adb | 641 | } |
7d367e06 | 642 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); |
c1d10adb | 643 | |
ba76738c PNA |
644 | static inline void nf_ct_acct_update(struct nf_conn *ct, |
645 | enum ip_conntrack_info ctinfo, | |
646 | unsigned int len) | |
647 | { | |
648 | struct nf_conn_acct *acct; | |
649 | ||
650 | acct = nf_conn_acct_find(ct); | |
651 | if (acct) { | |
652 | struct nf_conn_counter *counter = acct->counter; | |
653 | ||
654 | atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); | |
655 | atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes); | |
656 | } | |
657 | } | |
658 | ||
71d8c47f PNA |
659 | static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo, |
660 | const struct nf_conn *loser_ct) | |
661 | { | |
662 | struct nf_conn_acct *acct; | |
663 | ||
664 | acct = nf_conn_acct_find(loser_ct); | |
665 | if (acct) { | |
666 | struct nf_conn_counter *counter = acct->counter; | |
71d8c47f PNA |
667 | unsigned int bytes; |
668 | ||
669 | /* u32 should be fine since we must have seen one packet. */ | |
670 | bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes); | |
671 | nf_ct_acct_update(ct, ctinfo, bytes); | |
672 | } | |
673 | } | |
674 | ||
675 | /* Resolve race on insertion if this protocol allows this. */ | |
676 | static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb, | |
677 | enum ip_conntrack_info ctinfo, | |
678 | struct nf_conntrack_tuple_hash *h) | |
679 | { | |
680 | /* This is the conntrack entry already in hashes that won race. */ | |
681 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
682 | struct nf_conntrack_l4proto *l4proto; | |
683 | ||
684 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); | |
685 | if (l4proto->allow_clash && | |
590b52e1 | 686 | !nfct_nat(ct) && |
71d8c47f PNA |
687 | !nf_ct_is_dying(ct) && |
688 | atomic_inc_not_zero(&ct->ct_general.use)) { | |
689 | nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct); | |
690 | nf_conntrack_put(skb->nfct); | |
691 | /* Assign conntrack already in hashes to this skbuff. Don't | |
692 | * modify skb->nfctinfo to ensure consistent stateful filtering. | |
693 | */ | |
694 | skb->nfct = &ct->ct_general; | |
695 | return NF_ACCEPT; | |
696 | } | |
697 | NF_CT_STAT_INC(net, drop); | |
698 | return NF_DROP; | |
699 | } | |
700 | ||
9fb9cbb1 YK |
701 | /* Confirm a connection given skb; places it in hash table */ |
702 | int | |
3db05fea | 703 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 | 704 | { |
308ac914 | 705 | const struct nf_conntrack_zone *zone; |
b476b72a | 706 | unsigned int hash, reply_hash; |
df0933dc | 707 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 708 | struct nf_conn *ct; |
df0933dc | 709 | struct nf_conn_help *help; |
a992ca2a | 710 | struct nf_conn_tstamp *tstamp; |
ea781f19 | 711 | struct hlist_nulls_node *n; |
9fb9cbb1 | 712 | enum ip_conntrack_info ctinfo; |
400dad39 | 713 | struct net *net; |
93bb0ceb | 714 | unsigned int sequence; |
71d8c47f | 715 | int ret = NF_DROP; |
9fb9cbb1 | 716 | |
3db05fea | 717 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 718 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
719 | |
720 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
721 | ICMP/TCP RST packets in other direction. Actual packet | |
722 | which created connection will be IP_CT_NEW or for an | |
723 | expected connection, IP_CT_RELATED. */ | |
724 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
725 | return NF_ACCEPT; | |
726 | ||
5d0aa2cc | 727 | zone = nf_ct_zone(ct); |
93bb0ceb JDB |
728 | local_bh_disable(); |
729 | ||
730 | do { | |
a3efd812 | 731 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb JDB |
732 | /* reuse the hash saved before */ |
733 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
56d52d48 | 734 | hash = scale_hash(hash); |
deedb590 | 735 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
736 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
737 | ||
738 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
9fb9cbb1 YK |
739 | |
740 | /* We're not in hash table, and we refuse to set up related | |
93bb0ceb JDB |
741 | * connections for unconfirmed conns. But packet copies and |
742 | * REJECT will give spurious warnings here. | |
743 | */ | |
9fb9cbb1 YK |
744 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ |
745 | ||
25985edc | 746 | /* No external references means no one else could have |
93bb0ceb JDB |
747 | * confirmed us. |
748 | */ | |
9fb9cbb1 | 749 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); |
0d53778e | 750 | pr_debug("Confirming conntrack %p\n", ct); |
8ca3f5e9 PNA |
751 | /* We have to check the DYING flag after unlink to prevent |
752 | * a race against nf_ct_get_next_corpse() possibly called from | |
753 | * user context, else we insert an already 'dead' hash, blocking | |
754 | * further use of that particular connection -JM. | |
755 | */ | |
756 | nf_ct_del_from_dying_or_unconfirmed_list(ct); | |
757 | ||
71d8c47f PNA |
758 | if (unlikely(nf_ct_is_dying(ct))) { |
759 | nf_ct_add_to_dying_list(ct); | |
760 | goto dying; | |
761 | } | |
fc350777 | 762 | |
9fb9cbb1 YK |
763 | /* See if there's one in the list already, including reverse: |
764 | NAT could have grabbed it without realizing, since we're | |
765 | not in the hash. If there is, we lost race. */ | |
56d52d48 | 766 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 767 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 768 | zone, net)) |
df0933dc | 769 | goto out; |
86804348 | 770 | |
56d52d48 | 771 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 772 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 773 | zone, net)) |
df0933dc | 774 | goto out; |
9fb9cbb1 | 775 | |
df0933dc PM |
776 | /* Timer relative to confirmation time, not original |
777 | setting time, otherwise we'd get timer wrap in | |
778 | weird delay cases. */ | |
f330a7fd | 779 | ct->timeout += nfct_time_stamp; |
df0933dc | 780 | atomic_inc(&ct->ct_general.use); |
45eec341 | 781 | ct->status |= IPS_CONFIRMED; |
5c8ec910 | 782 | |
a992ca2a PNA |
783 | /* set conntrack timestamp, if enabled. */ |
784 | tstamp = nf_conn_tstamp_find(ct); | |
785 | if (tstamp) { | |
786 | if (skb->tstamp.tv64 == 0) | |
e3192690 | 787 | __net_timestamp(skb); |
a992ca2a PNA |
788 | |
789 | tstamp->start = ktime_to_ns(skb->tstamp); | |
790 | } | |
5c8ec910 PM |
791 | /* Since the lookup is lockless, hash insertion must be done after |
792 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
793 | * guarantee that no other CPU can find the conntrack before the above | |
794 | * stores are visible. | |
795 | */ | |
b476b72a | 796 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 797 | nf_conntrack_double_unlock(hash, reply_hash); |
93bb0ceb | 798 | local_bh_enable(); |
5c8ec910 | 799 | |
df0933dc PM |
800 | help = nfct_help(ct); |
801 | if (help && help->helper) | |
a71996fc | 802 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 803 | |
df0933dc | 804 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 805 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 806 | return NF_ACCEPT; |
9fb9cbb1 | 807 | |
df0933dc | 808 | out: |
8ca3f5e9 | 809 | nf_ct_add_to_dying_list(ct); |
71d8c47f PNA |
810 | ret = nf_ct_resolve_clash(net, skb, ctinfo, h); |
811 | dying: | |
93bb0ceb | 812 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 813 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 814 | local_bh_enable(); |
71d8c47f | 815 | return ret; |
9fb9cbb1 | 816 | } |
13b18339 | 817 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
818 | |
819 | /* Returns true if a connection correspondings to the tuple (required | |
820 | for NAT). */ | |
821 | int | |
822 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
823 | const struct nf_conn *ignored_conntrack) | |
824 | { | |
400dad39 | 825 | struct net *net = nf_ct_net(ignored_conntrack); |
308ac914 | 826 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 | 827 | struct nf_conntrack_tuple_hash *h; |
5e3c61f9 | 828 | struct hlist_nulls_head *ct_hash; |
92e47ba8 | 829 | unsigned int hash, hsize; |
ea781f19 | 830 | struct hlist_nulls_node *n; |
5d0aa2cc | 831 | struct nf_conn *ct; |
308ac914 DB |
832 | |
833 | zone = nf_ct_zone(ignored_conntrack); | |
9fb9cbb1 | 834 | |
2cf12348 | 835 | rcu_read_lock(); |
95a8d19f | 836 | begin: |
92e47ba8 LZ |
837 | nf_conntrack_get_ht(&ct_hash, &hsize); |
838 | hash = __hash_conntrack(net, tuple, hsize); | |
5e3c61f9 FW |
839 | |
840 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) { | |
5d0aa2cc | 841 | ct = nf_ct_tuplehash_to_ctrack(h); |
f330a7fd FW |
842 | |
843 | if (ct == ignored_conntrack) | |
844 | continue; | |
845 | ||
846 | if (nf_ct_is_expired(ct)) { | |
847 | nf_ct_gc_expired(ct); | |
848 | continue; | |
849 | } | |
850 | ||
851 | if (nf_ct_key_equal(h, tuple, zone, net)) { | |
2cf12348 FW |
852 | NF_CT_STAT_INC_ATOMIC(net, found); |
853 | rcu_read_unlock(); | |
ba419aff PM |
854 | return 1; |
855 | } | |
ba419aff | 856 | } |
95a8d19f FW |
857 | |
858 | if (get_nulls_value(n) != hash) { | |
859 | NF_CT_STAT_INC_ATOMIC(net, search_restart); | |
860 | goto begin; | |
861 | } | |
862 | ||
2cf12348 | 863 | rcu_read_unlock(); |
9fb9cbb1 | 864 | |
ba419aff | 865 | return 0; |
9fb9cbb1 | 866 | } |
13b18339 | 867 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 868 | |
7ae7730f PM |
869 | #define NF_CT_EVICTION_RANGE 8 |
870 | ||
9fb9cbb1 YK |
871 | /* There's a small race here where we may free a just-assured |
872 | connection. Too bad: we're in trouble anyway. */ | |
242922a0 FW |
873 | static unsigned int early_drop_list(struct net *net, |
874 | struct hlist_nulls_head *head) | |
9fb9cbb1 | 875 | { |
9fb9cbb1 | 876 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 877 | struct hlist_nulls_node *n; |
242922a0 FW |
878 | unsigned int drops = 0; |
879 | struct nf_conn *tmp; | |
3e86638e | 880 | |
242922a0 FW |
881 | hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) { |
882 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
9fb9cbb1 | 883 | |
f330a7fd FW |
884 | if (nf_ct_is_expired(tmp)) { |
885 | nf_ct_gc_expired(tmp); | |
886 | continue; | |
887 | } | |
888 | ||
242922a0 FW |
889 | if (test_bit(IPS_ASSURED_BIT, &tmp->status) || |
890 | !net_eq(nf_ct_net(tmp), net) || | |
891 | nf_ct_is_dying(tmp)) | |
892 | continue; | |
76507f69 | 893 | |
242922a0 FW |
894 | if (!atomic_inc_not_zero(&tmp->ct_general.use)) |
895 | continue; | |
76507f69 | 896 | |
242922a0 FW |
897 | /* kill only if still in same netns -- might have moved due to |
898 | * SLAB_DESTROY_BY_RCU rules. | |
899 | * | |
900 | * We steal the timer reference. If that fails timer has | |
901 | * already fired or someone else deleted it. Just drop ref | |
902 | * and move to next entry. | |
903 | */ | |
904 | if (net_eq(nf_ct_net(tmp), net) && | |
905 | nf_ct_is_confirmed(tmp) && | |
242922a0 FW |
906 | nf_ct_delete(tmp, 0, 0)) |
907 | drops++; | |
908 | ||
909 | nf_ct_put(tmp); | |
9fb9cbb1 | 910 | } |
3e86638e | 911 | |
242922a0 FW |
912 | return drops; |
913 | } | |
9fb9cbb1 | 914 | |
242922a0 FW |
915 | static noinline int early_drop(struct net *net, unsigned int _hash) |
916 | { | |
917 | unsigned int i; | |
9fb9cbb1 | 918 | |
242922a0 FW |
919 | for (i = 0; i < NF_CT_EVICTION_RANGE; i++) { |
920 | struct hlist_nulls_head *ct_hash; | |
92e47ba8 | 921 | unsigned int hash, hsize, drops; |
242922a0 | 922 | |
3101e0fc | 923 | rcu_read_lock(); |
92e47ba8 LZ |
924 | nf_conntrack_get_ht(&ct_hash, &hsize); |
925 | hash = reciprocal_scale(_hash++, hsize); | |
242922a0 FW |
926 | |
927 | drops = early_drop_list(net, &ct_hash[hash]); | |
3101e0fc LZ |
928 | rcu_read_unlock(); |
929 | ||
242922a0 FW |
930 | if (drops) { |
931 | NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops); | |
932 | return true; | |
74138511 | 933 | } |
9fb9cbb1 | 934 | } |
3e86638e | 935 | |
242922a0 | 936 | return false; |
9fb9cbb1 YK |
937 | } |
938 | ||
b87a2f91 FW |
939 | static void gc_worker(struct work_struct *work) |
940 | { | |
941 | unsigned int i, goal, buckets = 0, expired_count = 0; | |
b87a2f91 | 942 | struct conntrack_gc_work *gc_work; |
e0df8cae FW |
943 | unsigned int ratio, scanned = 0; |
944 | unsigned long next_run; | |
b87a2f91 FW |
945 | |
946 | gc_work = container_of(work, struct conntrack_gc_work, dwork.work); | |
947 | ||
e0df8cae | 948 | goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV; |
b87a2f91 FW |
949 | i = gc_work->last_bucket; |
950 | ||
951 | do { | |
952 | struct nf_conntrack_tuple_hash *h; | |
953 | struct hlist_nulls_head *ct_hash; | |
954 | struct hlist_nulls_node *n; | |
955 | unsigned int hashsz; | |
956 | struct nf_conn *tmp; | |
957 | ||
958 | i++; | |
959 | rcu_read_lock(); | |
960 | ||
961 | nf_conntrack_get_ht(&ct_hash, &hashsz); | |
962 | if (i >= hashsz) | |
963 | i = 0; | |
964 | ||
965 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) { | |
966 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
967 | ||
c023c0e4 | 968 | scanned++; |
b87a2f91 FW |
969 | if (nf_ct_is_expired(tmp)) { |
970 | nf_ct_gc_expired(tmp); | |
971 | expired_count++; | |
972 | continue; | |
973 | } | |
974 | } | |
975 | ||
976 | /* could check get_nulls_value() here and restart if ct | |
977 | * was moved to another chain. But given gc is best-effort | |
978 | * we will just continue with next hash slot. | |
979 | */ | |
980 | rcu_read_unlock(); | |
981 | cond_resched_rcu_qs(); | |
982 | } while (++buckets < goal && | |
983 | expired_count < GC_MAX_EVICTS); | |
984 | ||
985 | if (gc_work->exiting) | |
986 | return; | |
987 | ||
e0df8cae FW |
988 | /* |
989 | * Eviction will normally happen from the packet path, and not | |
990 | * from this gc worker. | |
991 | * | |
992 | * This worker is only here to reap expired entries when system went | |
993 | * idle after a busy period. | |
994 | * | |
995 | * The heuristics below are supposed to balance conflicting goals: | |
996 | * | |
997 | * 1. Minimize time until we notice a stale entry | |
998 | * 2. Maximize scan intervals to not waste cycles | |
999 | * | |
1000 | * Normally, expired_count will be 0, this increases the next_run time | |
1001 | * to priorize 2) above. | |
1002 | * | |
1003 | * As soon as a timed-out entry is found, move towards 1) and increase | |
1004 | * the scan frequency. | |
1005 | * In case we have lots of evictions next scan is done immediately. | |
1006 | */ | |
c023c0e4 | 1007 | ratio = scanned ? expired_count * 100 / scanned : 0; |
e0df8cae FW |
1008 | if (ratio >= 90 || expired_count == GC_MAX_EVICTS) { |
1009 | gc_work->next_gc_run = 0; | |
c023c0e4 | 1010 | next_run = 0; |
e0df8cae FW |
1011 | } else if (expired_count) { |
1012 | gc_work->next_gc_run /= 2U; | |
1013 | next_run = msecs_to_jiffies(1); | |
1014 | } else { | |
1015 | if (gc_work->next_gc_run < GC_INTERVAL_MAX) | |
1016 | gc_work->next_gc_run += msecs_to_jiffies(1); | |
1017 | ||
1018 | next_run = gc_work->next_gc_run; | |
1019 | } | |
c023c0e4 | 1020 | |
b87a2f91 | 1021 | gc_work->last_bucket = i; |
e0df8cae | 1022 | queue_delayed_work(system_long_wq, &gc_work->dwork, next_run); |
b87a2f91 FW |
1023 | } |
1024 | ||
1025 | static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work) | |
1026 | { | |
1027 | INIT_DELAYED_WORK(&gc_work->dwork, gc_worker); | |
e0df8cae | 1028 | gc_work->next_gc_run = GC_INTERVAL_MAX; |
b87a2f91 FW |
1029 | gc_work->exiting = false; |
1030 | } | |
1031 | ||
99f07e91 | 1032 | static struct nf_conn * |
308ac914 DB |
1033 | __nf_conntrack_alloc(struct net *net, |
1034 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1035 | const struct nf_conntrack_tuple *orig, |
1036 | const struct nf_conntrack_tuple *repl, | |
1037 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 1038 | { |
cd7fcbf1 | 1039 | struct nf_conn *ct; |
9fb9cbb1 | 1040 | |
5251e2d2 | 1041 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 1042 | atomic_inc(&net->ct.count); |
5251e2d2 | 1043 | |
76eb9460 | 1044 | if (nf_conntrack_max && |
49ac8713 | 1045 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
93bb0ceb | 1046 | if (!early_drop(net, hash)) { |
49ac8713 | 1047 | atomic_dec(&net->ct.count); |
e87cc472 | 1048 | net_warn_ratelimited("nf_conntrack: table full, dropping packet\n"); |
9fb9cbb1 YK |
1049 | return ERR_PTR(-ENOMEM); |
1050 | } | |
1051 | } | |
1052 | ||
941297f4 ED |
1053 | /* |
1054 | * Do not use kmem_cache_zalloc(), as this cache uses | |
1055 | * SLAB_DESTROY_BY_RCU. | |
1056 | */ | |
0c5366b3 | 1057 | ct = kmem_cache_alloc(nf_conntrack_cachep, gfp); |
5e8018fc DB |
1058 | if (ct == NULL) |
1059 | goto out; | |
1060 | ||
440f0d58 | 1061 | spin_lock_init(&ct->lock); |
c88130bc | 1062 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 1063 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 1064 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
1065 | /* save hash for reusing when confirming */ |
1066 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
c41884ce | 1067 | ct->status = 0; |
c2d9ba9b | 1068 | write_pnet(&ct->ct_net, net); |
c41884ce FW |
1069 | memset(&ct->__nfct_init_offset[0], 0, |
1070 | offsetof(struct nf_conn, proto) - | |
1071 | offsetof(struct nf_conn, __nfct_init_offset[0])); | |
5e8018fc | 1072 | |
6c8dee98 | 1073 | nf_ct_zone_add(ct, zone); |
5e8018fc | 1074 | |
e53376be PNA |
1075 | /* Because we use RCU lookups, we set ct_general.use to zero before |
1076 | * this is inserted in any list. | |
941297f4 | 1077 | */ |
e53376be | 1078 | atomic_set(&ct->ct_general.use, 0); |
c88130bc | 1079 | return ct; |
5e8018fc DB |
1080 | out: |
1081 | atomic_dec(&net->ct.count); | |
5d0aa2cc | 1082 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 1083 | } |
99f07e91 | 1084 | |
308ac914 DB |
1085 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
1086 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1087 | const struct nf_conntrack_tuple *orig, |
1088 | const struct nf_conntrack_tuple *repl, | |
1089 | gfp_t gfp) | |
1090 | { | |
1091 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
1092 | } | |
13b18339 | 1093 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 1094 | |
c88130bc | 1095 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 1096 | { |
1d45209d ED |
1097 | struct net *net = nf_ct_net(ct); |
1098 | ||
e53376be PNA |
1099 | /* A freed object has refcnt == 0, that's |
1100 | * the golden rule for SLAB_DESTROY_BY_RCU | |
1101 | */ | |
1102 | NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0); | |
1103 | ||
ceeff754 | 1104 | nf_ct_ext_destroy(ct); |
ea781f19 | 1105 | nf_ct_ext_free(ct); |
0c5366b3 | 1106 | kmem_cache_free(nf_conntrack_cachep, ct); |
4e857c58 | 1107 | smp_mb__before_atomic(); |
0c3c6c00 | 1108 | atomic_dec(&net->ct.count); |
76507f69 | 1109 | } |
13b18339 | 1110 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 | 1111 | |
c539f017 | 1112 | |
9fb9cbb1 YK |
1113 | /* Allocate a new conntrack: we return -ENOMEM if classification |
1114 | failed due to stress. Otherwise it really is unclassifiable. */ | |
1115 | static struct nf_conntrack_tuple_hash * | |
b2a15a60 | 1116 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 1117 | const struct nf_conntrack_tuple *tuple, |
9fb9cbb1 | 1118 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 1119 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 1120 | struct sk_buff *skb, |
60b5f8f7 | 1121 | unsigned int dataoff, u32 hash) |
9fb9cbb1 | 1122 | { |
c88130bc | 1123 | struct nf_conn *ct; |
3c158f7f | 1124 | struct nf_conn_help *help; |
9fb9cbb1 | 1125 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 1126 | struct nf_conntrack_ecache *ecache; |
ca7433df | 1127 | struct nf_conntrack_expect *exp = NULL; |
308ac914 | 1128 | const struct nf_conntrack_zone *zone; |
60b5f8f7 | 1129 | struct nf_conn_timeout *timeout_ext; |
5e8018fc | 1130 | struct nf_conntrack_zone tmp; |
60b5f8f7 | 1131 | unsigned int *timeouts; |
9fb9cbb1 | 1132 | |
605dcad6 | 1133 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 1134 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
1135 | return NULL; |
1136 | } | |
1137 | ||
5e8018fc | 1138 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
99f07e91 CG |
1139 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
1140 | hash); | |
0a9ee813 | 1141 | if (IS_ERR(ct)) |
c88130bc | 1142 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 | 1143 | |
4440a2ab GF |
1144 | if (!nf_ct_add_synproxy(ct, tmpl)) { |
1145 | nf_conntrack_free(ct); | |
1146 | return ERR_PTR(-ENOMEM); | |
48b1de4c PM |
1147 | } |
1148 | ||
60b5f8f7 | 1149 | timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; |
ae2d708e PNA |
1150 | if (timeout_ext) { |
1151 | timeouts = nf_ct_timeout_data(timeout_ext); | |
1152 | if (unlikely(!timeouts)) | |
1153 | timeouts = l4proto->get_timeouts(net); | |
1154 | } else { | |
60b5f8f7 | 1155 | timeouts = l4proto->get_timeouts(net); |
ae2d708e | 1156 | } |
60b5f8f7 | 1157 | |
2c8503f5 | 1158 | if (!l4proto->new(ct, skb, dataoff, timeouts)) { |
c88130bc | 1159 | nf_conntrack_free(ct); |
ccd63c20 | 1160 | pr_debug("can't track with proto module\n"); |
9fb9cbb1 YK |
1161 | return NULL; |
1162 | } | |
1163 | ||
60b5f8f7 | 1164 | if (timeout_ext) |
ae2d708e PNA |
1165 | nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout), |
1166 | GFP_ATOMIC); | |
60b5f8f7 | 1167 | |
58401572 | 1168 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a992ca2a | 1169 | nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); |
c539f017 | 1170 | nf_ct_labels_ext_add(ct); |
b2a15a60 PM |
1171 | |
1172 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
1173 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
1174 | ecache ? ecache->expmask : 0, | |
1175 | GFP_ATOMIC); | |
58401572 | 1176 | |
ca7433df JDB |
1177 | local_bh_disable(); |
1178 | if (net->ct.expect_count) { | |
1179 | spin_lock(&nf_conntrack_expect_lock); | |
1180 | exp = nf_ct_find_expectation(net, zone, tuple); | |
1181 | if (exp) { | |
ccd63c20 | 1182 | pr_debug("expectation arrives ct=%p exp=%p\n", |
ca7433df JDB |
1183 | ct, exp); |
1184 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
1185 | __set_bit(IPS_EXPECTED_BIT, &ct->status); | |
1186 | /* exp->master safe, refcnt bumped in nf_ct_find_expectation */ | |
1187 | ct->master = exp->master; | |
1188 | if (exp->helper) { | |
1189 | help = nf_ct_helper_ext_add(ct, exp->helper, | |
1190 | GFP_ATOMIC); | |
1191 | if (help) | |
1192 | rcu_assign_pointer(help->helper, exp->helper); | |
1193 | } | |
ceceae1b | 1194 | |
9fb9cbb1 | 1195 | #ifdef CONFIG_NF_CONNTRACK_MARK |
ca7433df | 1196 | ct->mark = exp->master->mark; |
7c9728c3 JM |
1197 | #endif |
1198 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
ca7433df | 1199 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 1200 | #endif |
ca7433df JDB |
1201 | NF_CT_STAT_INC(net, expect_new); |
1202 | } | |
1203 | spin_unlock(&nf_conntrack_expect_lock); | |
1204 | } | |
8e8118f8 | 1205 | if (!exp) |
b2a15a60 | 1206 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
9fb9cbb1 | 1207 | |
e53376be PNA |
1208 | /* Now it is inserted into the unconfirmed list, bump refcount */ |
1209 | nf_conntrack_get(&ct->ct_general); | |
b7779d06 | 1210 | nf_ct_add_to_unconfirmed_list(ct); |
9fb9cbb1 | 1211 | |
ca7433df | 1212 | local_bh_enable(); |
9fb9cbb1 YK |
1213 | |
1214 | if (exp) { | |
1215 | if (exp->expectfn) | |
c88130bc | 1216 | exp->expectfn(ct, exp); |
6823645d | 1217 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
1218 | } |
1219 | ||
c88130bc | 1220 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
1221 | } |
1222 | ||
1223 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ | |
1224 | static inline struct nf_conn * | |
b2a15a60 | 1225 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 1226 | struct sk_buff *skb, |
9fb9cbb1 YK |
1227 | unsigned int dataoff, |
1228 | u_int16_t l3num, | |
1229 | u_int8_t protonum, | |
1230 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 1231 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 1232 | int *set_reply, |
60b5f8f7 | 1233 | enum ip_conntrack_info *ctinfo) |
9fb9cbb1 | 1234 | { |
308ac914 | 1235 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 YK |
1236 | struct nf_conntrack_tuple tuple; |
1237 | struct nf_conntrack_tuple_hash *h; | |
5e8018fc | 1238 | struct nf_conntrack_zone tmp; |
9fb9cbb1 | 1239 | struct nf_conn *ct; |
99f07e91 | 1240 | u32 hash; |
9fb9cbb1 | 1241 | |
bbe735e4 | 1242 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
a31f1adc | 1243 | dataoff, l3num, protonum, net, &tuple, l3proto, |
605dcad6 | 1244 | l4proto)) { |
ccd63c20 | 1245 | pr_debug("Can't get tuple\n"); |
9fb9cbb1 YK |
1246 | return NULL; |
1247 | } | |
1248 | ||
1249 | /* look for tuple match */ | |
5e8018fc | 1250 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
1b8c8a9f | 1251 | hash = hash_conntrack_raw(&tuple, net); |
99f07e91 | 1252 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); |
9fb9cbb1 | 1253 | if (!h) { |
b2a15a60 | 1254 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
60b5f8f7 | 1255 | skb, dataoff, hash); |
9fb9cbb1 YK |
1256 | if (!h) |
1257 | return NULL; | |
1258 | if (IS_ERR(h)) | |
1259 | return (void *)h; | |
1260 | } | |
1261 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1262 | ||
1263 | /* It exists; we have (non-exclusive) reference. */ | |
1264 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
fb048833 | 1265 | *ctinfo = IP_CT_ESTABLISHED_REPLY; |
9fb9cbb1 YK |
1266 | /* Please set reply bit if this packet OK */ |
1267 | *set_reply = 1; | |
1268 | } else { | |
1269 | /* Once we've had two way comms, always ESTABLISHED. */ | |
1270 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
ccd63c20 | 1271 | pr_debug("normal packet for %p\n", ct); |
9fb9cbb1 YK |
1272 | *ctinfo = IP_CT_ESTABLISHED; |
1273 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
ccd63c20 | 1274 | pr_debug("related packet for %p\n", ct); |
9fb9cbb1 YK |
1275 | *ctinfo = IP_CT_RELATED; |
1276 | } else { | |
ccd63c20 | 1277 | pr_debug("new packet for %p\n", ct); |
9fb9cbb1 YK |
1278 | *ctinfo = IP_CT_NEW; |
1279 | } | |
1280 | *set_reply = 0; | |
1281 | } | |
1282 | skb->nfct = &ct->ct_general; | |
1283 | skb->nfctinfo = *ctinfo; | |
1284 | return ct; | |
1285 | } | |
1286 | ||
1287 | unsigned int | |
a702a65f AD |
1288 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
1289 | struct sk_buff *skb) | |
9fb9cbb1 | 1290 | { |
b2a15a60 | 1291 | struct nf_conn *ct, *tmpl = NULL; |
9fb9cbb1 YK |
1292 | enum ip_conntrack_info ctinfo; |
1293 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 1294 | struct nf_conntrack_l4proto *l4proto; |
2c8503f5 | 1295 | unsigned int *timeouts; |
9fb9cbb1 YK |
1296 | unsigned int dataoff; |
1297 | u_int8_t protonum; | |
1298 | int set_reply = 0; | |
1299 | int ret; | |
1300 | ||
3db05fea | 1301 | if (skb->nfct) { |
b2a15a60 PM |
1302 | /* Previously seen (loopback or untracked)? Ignore. */ |
1303 | tmpl = (struct nf_conn *)skb->nfct; | |
1304 | if (!nf_ct_is_template(tmpl)) { | |
1305 | NF_CT_STAT_INC_ATOMIC(net, ignore); | |
1306 | return NF_ACCEPT; | |
1307 | } | |
1308 | skb->nfct = NULL; | |
9fb9cbb1 YK |
1309 | } |
1310 | ||
e2361cb9 | 1311 | /* rcu_read_lock()ed by nf_hook_thresh */ |
76108cea | 1312 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 1313 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
1314 | &dataoff, &protonum); |
1315 | if (ret <= 0) { | |
25985edc | 1316 | pr_debug("not prepared to track yet or error occurred\n"); |
0d55af87 AD |
1317 | NF_CT_STAT_INC_ATOMIC(net, error); |
1318 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1319 | ret = -ret; |
1320 | goto out; | |
9fb9cbb1 YK |
1321 | } |
1322 | ||
76108cea | 1323 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
1324 | |
1325 | /* It may be an special packet, error, unclean... | |
1326 | * inverse of the return code tells to the netfilter | |
1327 | * core what to do with the packet. */ | |
74c51a14 | 1328 | if (l4proto->error != NULL) { |
8fea97ec PM |
1329 | ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo, |
1330 | pf, hooknum); | |
74c51a14 | 1331 | if (ret <= 0) { |
0d55af87 AD |
1332 | NF_CT_STAT_INC_ATOMIC(net, error); |
1333 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1334 | ret = -ret; |
1335 | goto out; | |
74c51a14 | 1336 | } |
88ed01d1 PNA |
1337 | /* ICMP[v6] protocol trackers may assign one conntrack. */ |
1338 | if (skb->nfct) | |
1339 | goto out; | |
9fb9cbb1 | 1340 | } |
08733a0c | 1341 | repeat: |
b2a15a60 | 1342 | ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
60b5f8f7 | 1343 | l3proto, l4proto, &set_reply, &ctinfo); |
9fb9cbb1 YK |
1344 | if (!ct) { |
1345 | /* Not valid part of a connection */ | |
0d55af87 | 1346 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
b2a15a60 PM |
1347 | ret = NF_ACCEPT; |
1348 | goto out; | |
9fb9cbb1 YK |
1349 | } |
1350 | ||
1351 | if (IS_ERR(ct)) { | |
1352 | /* Too stressed to deal. */ | |
0d55af87 | 1353 | NF_CT_STAT_INC_ATOMIC(net, drop); |
b2a15a60 PM |
1354 | ret = NF_DROP; |
1355 | goto out; | |
9fb9cbb1 YK |
1356 | } |
1357 | ||
3db05fea | 1358 | NF_CT_ASSERT(skb->nfct); |
9fb9cbb1 | 1359 | |
60b5f8f7 | 1360 | /* Decide what timeout policy we want to apply to this flow. */ |
84b5ee93 | 1361 | timeouts = nf_ct_timeout_lookup(net, ct, l4proto); |
60b5f8f7 | 1362 | |
2c8503f5 | 1363 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); |
ec8d5409 | 1364 | if (ret <= 0) { |
9fb9cbb1 YK |
1365 | /* Invalid: inverse of the return code tells |
1366 | * the netfilter core what to do */ | |
0d53778e | 1367 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
3db05fea HX |
1368 | nf_conntrack_put(skb->nfct); |
1369 | skb->nfct = NULL; | |
0d55af87 | 1370 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
1371 | if (ret == -NF_DROP) |
1372 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
56a62e22 AB |
1373 | /* Special case: TCP tracker reports an attempt to reopen a |
1374 | * closed/aborted connection. We have to go back and create a | |
1375 | * fresh conntrack. | |
1376 | */ | |
1377 | if (ret == -NF_REPEAT) | |
1378 | goto repeat; | |
b2a15a60 PM |
1379 | ret = -ret; |
1380 | goto out; | |
9fb9cbb1 YK |
1381 | } |
1382 | ||
1383 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 1384 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 | 1385 | out: |
56a62e22 AB |
1386 | if (tmpl) |
1387 | nf_ct_put(tmpl); | |
9fb9cbb1 YK |
1388 | |
1389 | return ret; | |
1390 | } | |
13b18339 | 1391 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 1392 | |
5f2b4c90 JE |
1393 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
1394 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 1395 | { |
5f2b4c90 | 1396 | bool ret; |
923f4902 PM |
1397 | |
1398 | rcu_read_lock(); | |
1399 | ret = nf_ct_invert_tuple(inverse, orig, | |
1400 | __nf_ct_l3proto_find(orig->src.l3num), | |
1401 | __nf_ct_l4proto_find(orig->src.l3num, | |
1402 | orig->dst.protonum)); | |
1403 | rcu_read_unlock(); | |
1404 | return ret; | |
9fb9cbb1 | 1405 | } |
13b18339 | 1406 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 1407 | |
5b1158e9 JK |
1408 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
1409 | implicitly racy: see __nf_conntrack_confirm */ | |
1410 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
1411 | const struct nf_conntrack_tuple *newreply) | |
1412 | { | |
1413 | struct nf_conn_help *help = nfct_help(ct); | |
1414 | ||
5b1158e9 JK |
1415 | /* Should be unconfirmed, so not in hash table yet */ |
1416 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
1417 | ||
0d53778e | 1418 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 1419 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
1420 | |
1421 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 1422 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 1423 | return; |
ceceae1b | 1424 | |
c52fbb41 | 1425 | rcu_read_lock(); |
b2a15a60 | 1426 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 1427 | rcu_read_unlock(); |
5b1158e9 | 1428 | } |
13b18339 | 1429 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 1430 | |
9fb9cbb1 YK |
1431 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
1432 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
1433 | enum ip_conntrack_info ctinfo, | |
1434 | const struct sk_buff *skb, | |
1435 | unsigned long extra_jiffies, | |
1436 | int do_acct) | |
1437 | { | |
9fb9cbb1 YK |
1438 | NF_CT_ASSERT(skb); |
1439 | ||
997ae831 | 1440 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1441 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1442 | goto acct; | |
997ae831 | 1443 | |
9fb9cbb1 | 1444 | /* If not in hash table, timer will not be active yet */ |
f330a7fd FW |
1445 | if (nf_ct_is_confirmed(ct)) |
1446 | extra_jiffies += nfct_time_stamp; | |
9fb9cbb1 | 1447 | |
f330a7fd | 1448 | ct->timeout = extra_jiffies; |
47d95045 | 1449 | acct: |
ba76738c PNA |
1450 | if (do_acct) |
1451 | nf_ct_acct_update(ct, ctinfo, skb->len); | |
9fb9cbb1 | 1452 | } |
13b18339 | 1453 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1454 | |
ad66713f FW |
1455 | bool nf_ct_kill_acct(struct nf_conn *ct, |
1456 | enum ip_conntrack_info ctinfo, | |
1457 | const struct sk_buff *skb) | |
51091764 | 1458 | { |
ad66713f | 1459 | nf_ct_acct_update(ct, ctinfo, skb->len); |
58401572 | 1460 | |
f330a7fd | 1461 | return nf_ct_delete(ct, 0, 0); |
51091764 | 1462 | } |
ad66713f | 1463 | EXPORT_SYMBOL_GPL(nf_ct_kill_acct); |
51091764 | 1464 | |
c0cd1156 | 1465 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
c1d10adb PNA |
1466 | |
1467 | #include <linux/netfilter/nfnetlink.h> | |
1468 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1469 | #include <linux/mutex.h> |
1470 | ||
c1d10adb PNA |
1471 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1472 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1473 | * or depend on ctnetlink */ | |
fdf70832 | 1474 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1475 | const struct nf_conntrack_tuple *tuple) |
1476 | { | |
bae65be8 DM |
1477 | if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) || |
1478 | nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port)) | |
1479 | goto nla_put_failure; | |
c1d10adb PNA |
1480 | return 0; |
1481 | ||
df6fb868 | 1482 | nla_put_failure: |
c1d10adb PNA |
1483 | return -1; |
1484 | } | |
fdf70832 | 1485 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1486 | |
f73e924c PM |
1487 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1488 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1489 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1490 | }; |
f73e924c | 1491 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1492 | |
fdf70832 | 1493 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1494 | struct nf_conntrack_tuple *t) |
1495 | { | |
df6fb868 | 1496 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1497 | return -EINVAL; |
1498 | ||
77236b6e PM |
1499 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1500 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1501 | |
1502 | return 0; | |
1503 | } | |
fdf70832 | 1504 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1505 | |
1506 | int nf_ct_port_nlattr_tuple_size(void) | |
1507 | { | |
1508 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1509 | } | |
1510 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1511 | #endif |
1512 | ||
9fb9cbb1 | 1513 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
312a0c16 | 1514 | static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb) |
9fb9cbb1 YK |
1515 | { |
1516 | struct nf_conn *ct; | |
1517 | enum ip_conntrack_info ctinfo; | |
1518 | ||
1519 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1520 | ct = nf_ct_get(skb, &ctinfo); | |
1521 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
fb048833 | 1522 | ctinfo = IP_CT_RELATED_REPLY; |
9fb9cbb1 YK |
1523 | else |
1524 | ctinfo = IP_CT_RELATED; | |
1525 | ||
1526 | /* Attach to new skbuff, and increment count */ | |
1527 | nskb->nfct = &ct->ct_general; | |
1528 | nskb->nfctinfo = ctinfo; | |
1529 | nf_conntrack_get(nskb->nfct); | |
1530 | } | |
1531 | ||
9fb9cbb1 | 1532 | /* Bring out ya dead! */ |
df0933dc | 1533 | static struct nf_conn * |
400dad39 | 1534 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1535 | void *data, unsigned int *bucket) |
1536 | { | |
df0933dc PM |
1537 | struct nf_conntrack_tuple_hash *h; |
1538 | struct nf_conn *ct; | |
ea781f19 | 1539 | struct hlist_nulls_node *n; |
b7779d06 | 1540 | int cpu; |
93bb0ceb | 1541 | spinlock_t *lockp; |
9fb9cbb1 | 1542 | |
56d52d48 | 1543 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
93bb0ceb JDB |
1544 | lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS]; |
1545 | local_bh_disable(); | |
b16c2919 | 1546 | nf_conntrack_lock(lockp); |
56d52d48 FW |
1547 | if (*bucket < nf_conntrack_htable_size) { |
1548 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) { | |
93bb0ceb JDB |
1549 | if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL) |
1550 | continue; | |
1551 | ct = nf_ct_tuplehash_to_ctrack(h); | |
e0c7d472 FW |
1552 | if (net_eq(nf_ct_net(ct), net) && |
1553 | iter(ct, data)) | |
93bb0ceb JDB |
1554 | goto found; |
1555 | } | |
df0933dc | 1556 | } |
93bb0ceb JDB |
1557 | spin_unlock(lockp); |
1558 | local_bh_enable(); | |
d93c6258 | 1559 | cond_resched(); |
601e68e1 | 1560 | } |
b7779d06 JDB |
1561 | |
1562 | for_each_possible_cpu(cpu) { | |
1563 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1564 | ||
1565 | spin_lock_bh(&pcpu->lock); | |
1566 | hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) { | |
1567 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1568 | if (iter(ct, data)) | |
1569 | set_bit(IPS_DYING_BIT, &ct->status); | |
1570 | } | |
1571 | spin_unlock_bh(&pcpu->lock); | |
d93c6258 | 1572 | cond_resched(); |
b7779d06 | 1573 | } |
df0933dc PM |
1574 | return NULL; |
1575 | found: | |
c073e3fa | 1576 | atomic_inc(&ct->ct_general.use); |
93bb0ceb JDB |
1577 | spin_unlock(lockp); |
1578 | local_bh_enable(); | |
df0933dc | 1579 | return ct; |
9fb9cbb1 YK |
1580 | } |
1581 | ||
400dad39 AD |
1582 | void nf_ct_iterate_cleanup(struct net *net, |
1583 | int (*iter)(struct nf_conn *i, void *data), | |
c655bc68 | 1584 | void *data, u32 portid, int report) |
9fb9cbb1 | 1585 | { |
df0933dc | 1586 | struct nf_conn *ct; |
9fb9cbb1 YK |
1587 | unsigned int bucket = 0; |
1588 | ||
d93c6258 FW |
1589 | might_sleep(); |
1590 | ||
88b68bc5 FW |
1591 | if (atomic_read(&net->ct.count) == 0) |
1592 | return; | |
1593 | ||
400dad39 | 1594 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 | 1595 | /* Time to push up daises... */ |
9fb9cbb1 | 1596 | |
f330a7fd | 1597 | nf_ct_delete(ct, portid, report); |
9fb9cbb1 | 1598 | nf_ct_put(ct); |
d93c6258 | 1599 | cond_resched(); |
9fb9cbb1 YK |
1600 | } |
1601 | } | |
13b18339 | 1602 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1603 | |
274d383b PNA |
1604 | static int kill_all(struct nf_conn *i, void *data) |
1605 | { | |
1606 | return 1; | |
1607 | } | |
1608 | ||
d862a662 | 1609 | void nf_ct_free_hashtable(void *hash, unsigned int size) |
9fb9cbb1 | 1610 | { |
d862a662 | 1611 | if (is_vmalloc_addr(hash)) |
9fb9cbb1 YK |
1612 | vfree(hash); |
1613 | else | |
601e68e1 | 1614 | free_pages((unsigned long)hash, |
f205c5e0 | 1615 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1616 | } |
ac565e5f | 1617 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1618 | |
b3c5163f ED |
1619 | static int untrack_refs(void) |
1620 | { | |
1621 | int cnt = 0, cpu; | |
1622 | ||
1623 | for_each_possible_cpu(cpu) { | |
1624 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
1625 | ||
1626 | cnt += atomic_read(&ct->ct_general.use) - 1; | |
1627 | } | |
1628 | return cnt; | |
1629 | } | |
1630 | ||
f94161c1 | 1631 | void nf_conntrack_cleanup_start(void) |
9fb9cbb1 | 1632 | { |
b87a2f91 | 1633 | conntrack_gc_work.exiting = true; |
f94161c1 G |
1634 | RCU_INIT_POINTER(ip_ct_attach, NULL); |
1635 | } | |
1636 | ||
1637 | void nf_conntrack_cleanup_end(void) | |
1638 | { | |
1639 | RCU_INIT_POINTER(nf_ct_destroy, NULL); | |
b3c5163f | 1640 | while (untrack_refs() > 0) |
9edd7ca0 PM |
1641 | schedule(); |
1642 | ||
b87a2f91 | 1643 | cancel_delayed_work_sync(&conntrack_gc_work.dwork); |
56d52d48 FW |
1644 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
1645 | ||
04d87001 | 1646 | nf_conntrack_proto_fini(); |
41d73ec0 | 1647 | nf_conntrack_seqadj_fini(); |
5f69b8f5 | 1648 | nf_conntrack_labels_fini(); |
5e615b22 | 1649 | nf_conntrack_helper_fini(); |
8684094c | 1650 | nf_conntrack_timeout_fini(); |
3fe0f943 | 1651 | nf_conntrack_ecache_fini(); |
73f4001a | 1652 | nf_conntrack_tstamp_fini(); |
b7ff3a1f | 1653 | nf_conntrack_acct_fini(); |
83b4dbe1 | 1654 | nf_conntrack_expect_fini(); |
77571149 FW |
1655 | |
1656 | kmem_cache_destroy(nf_conntrack_cachep); | |
08f6547d | 1657 | } |
9fb9cbb1 | 1658 | |
f94161c1 G |
1659 | /* |
1660 | * Mishearing the voices in his head, our hero wonders how he's | |
1661 | * supposed to kill the mall. | |
1662 | */ | |
1663 | void nf_conntrack_cleanup_net(struct net *net) | |
08f6547d | 1664 | { |
dece40e8 VD |
1665 | LIST_HEAD(single); |
1666 | ||
1667 | list_add(&net->exit_list, &single); | |
1668 | nf_conntrack_cleanup_net_list(&single); | |
1669 | } | |
1670 | ||
1671 | void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list) | |
1672 | { | |
1673 | int busy; | |
1674 | struct net *net; | |
1675 | ||
f94161c1 G |
1676 | /* |
1677 | * This makes sure all current packets have passed through | |
1678 | * netfilter framework. Roll on, two-stage module | |
1679 | * delete... | |
1680 | */ | |
1681 | synchronize_net(); | |
dece40e8 VD |
1682 | i_see_dead_people: |
1683 | busy = 0; | |
1684 | list_for_each_entry(net, net_exit_list, exit_list) { | |
c655bc68 | 1685 | nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0); |
dece40e8 VD |
1686 | if (atomic_read(&net->ct.count) != 0) |
1687 | busy = 1; | |
1688 | } | |
1689 | if (busy) { | |
9fb9cbb1 YK |
1690 | schedule(); |
1691 | goto i_see_dead_people; | |
1692 | } | |
1693 | ||
dece40e8 | 1694 | list_for_each_entry(net, net_exit_list, exit_list) { |
dece40e8 VD |
1695 | nf_conntrack_proto_pernet_fini(net); |
1696 | nf_conntrack_helper_pernet_fini(net); | |
1697 | nf_conntrack_ecache_pernet_fini(net); | |
1698 | nf_conntrack_tstamp_pernet_fini(net); | |
1699 | nf_conntrack_acct_pernet_fini(net); | |
1700 | nf_conntrack_expect_pernet_fini(net); | |
dece40e8 | 1701 | free_percpu(net->ct.stat); |
b7779d06 | 1702 | free_percpu(net->ct.pcpu_lists); |
dece40e8 | 1703 | } |
08f6547d AD |
1704 | } |
1705 | ||
d862a662 | 1706 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) |
9fb9cbb1 | 1707 | { |
ea781f19 ED |
1708 | struct hlist_nulls_head *hash; |
1709 | unsigned int nr_slots, i; | |
1710 | size_t sz; | |
9fb9cbb1 | 1711 | |
9cc1c73a FW |
1712 | if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head))) |
1713 | return NULL; | |
1714 | ||
ea781f19 ED |
1715 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1716 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
9cc1c73a FW |
1717 | |
1718 | if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head))) | |
1719 | return NULL; | |
1720 | ||
ea781f19 ED |
1721 | sz = nr_slots * sizeof(struct hlist_nulls_head); |
1722 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1723 | get_order(sz)); | |
f0ad4621 | 1724 | if (!hash) |
966567b7 | 1725 | hash = vzalloc(sz); |
9fb9cbb1 | 1726 | |
ea781f19 ED |
1727 | if (hash && nulls) |
1728 | for (i = 0; i < nr_slots; i++) | |
1729 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1730 | |
1731 | return hash; | |
1732 | } | |
ac565e5f | 1733 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1734 | |
3183ab89 | 1735 | int nf_conntrack_hash_resize(unsigned int hashsize) |
9fb9cbb1 | 1736 | { |
3183ab89 FW |
1737 | int i, bucket; |
1738 | unsigned int old_size; | |
ea781f19 | 1739 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1740 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1741 | struct nf_conn *ct; |
9fb9cbb1 | 1742 | |
9fb9cbb1 YK |
1743 | if (!hashsize) |
1744 | return -EINVAL; | |
1745 | ||
d862a662 | 1746 | hash = nf_ct_alloc_hashtable(&hashsize, 1); |
9fb9cbb1 YK |
1747 | if (!hash) |
1748 | return -ENOMEM; | |
1749 | ||
3183ab89 FW |
1750 | old_size = nf_conntrack_htable_size; |
1751 | if (old_size == hashsize) { | |
1752 | nf_ct_free_hashtable(hash, hashsize); | |
1753 | return 0; | |
1754 | } | |
1755 | ||
93bb0ceb JDB |
1756 | local_bh_disable(); |
1757 | nf_conntrack_all_lock(); | |
a3efd812 | 1758 | write_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb | 1759 | |
76507f69 PM |
1760 | /* Lookups in the old hash might happen in parallel, which means we |
1761 | * might get false negatives during connection lookup. New connections | |
1762 | * created because of a false negative won't make it into the hash | |
93bb0ceb | 1763 | * though since that required taking the locks. |
76507f69 | 1764 | */ |
93bb0ceb | 1765 | |
56d52d48 FW |
1766 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
1767 | while (!hlist_nulls_empty(&nf_conntrack_hash[i])) { | |
1768 | h = hlist_nulls_entry(nf_conntrack_hash[i].first, | |
1769 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1770 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1771 | hlist_nulls_del_rcu(&h->hnnode); |
1b8c8a9f FW |
1772 | bucket = __hash_conntrack(nf_ct_net(ct), |
1773 | &h->tuple, hashsize); | |
ea781f19 | 1774 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1775 | } |
1776 | } | |
56d52d48 FW |
1777 | old_size = nf_conntrack_htable_size; |
1778 | old_hash = nf_conntrack_hash; | |
9fb9cbb1 | 1779 | |
56d52d48 FW |
1780 | nf_conntrack_hash = hash; |
1781 | nf_conntrack_htable_size = hashsize; | |
93bb0ceb | 1782 | |
a3efd812 | 1783 | write_seqcount_end(&nf_conntrack_generation); |
93bb0ceb JDB |
1784 | nf_conntrack_all_unlock(); |
1785 | local_bh_enable(); | |
9fb9cbb1 | 1786 | |
5e3c61f9 | 1787 | synchronize_net(); |
d862a662 | 1788 | nf_ct_free_hashtable(old_hash, old_size); |
9fb9cbb1 YK |
1789 | return 0; |
1790 | } | |
3183ab89 FW |
1791 | |
1792 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) | |
1793 | { | |
1794 | unsigned int hashsize; | |
1795 | int rc; | |
1796 | ||
1797 | if (current->nsproxy->net_ns != &init_net) | |
1798 | return -EOPNOTSUPP; | |
1799 | ||
1800 | /* On boot, we can set this without any fancy locking. */ | |
1801 | if (!nf_conntrack_htable_size) | |
1802 | return param_set_uint(val, kp); | |
1803 | ||
1804 | rc = kstrtouint(val, 0, &hashsize); | |
1805 | if (rc) | |
1806 | return rc; | |
1807 | ||
1808 | return nf_conntrack_hash_resize(hashsize); | |
1809 | } | |
fae718dd | 1810 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1811 | |
fae718dd | 1812 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1813 | &nf_conntrack_htable_size, 0600); |
1814 | ||
5bfddbd4 ED |
1815 | void nf_ct_untracked_status_or(unsigned long bits) |
1816 | { | |
b3c5163f ED |
1817 | int cpu; |
1818 | ||
1819 | for_each_possible_cpu(cpu) | |
1820 | per_cpu(nf_conntrack_untracked, cpu).status |= bits; | |
5bfddbd4 ED |
1821 | } |
1822 | EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); | |
1823 | ||
f94161c1 | 1824 | int nf_conntrack_init_start(void) |
9fb9cbb1 | 1825 | { |
f205c5e0 | 1826 | int max_factor = 8; |
0c5366b3 FW |
1827 | int ret = -ENOMEM; |
1828 | int i, cpu; | |
93bb0ceb | 1829 | |
a3efd812 FW |
1830 | seqcount_init(&nf_conntrack_generation); |
1831 | ||
d5d20912 | 1832 | for (i = 0; i < CONNTRACK_LOCKS; i++) |
93bb0ceb | 1833 | spin_lock_init(&nf_conntrack_locks[i]); |
9fb9cbb1 | 1834 | |
9fb9cbb1 | 1835 | if (!nf_conntrack_htable_size) { |
88eab472 ML |
1836 | /* Idea from tcp.c: use 1/16384 of memory. |
1837 | * On i386: 32MB machine has 512 buckets. | |
1838 | * >= 1GB machines have 16384 buckets. | |
1839 | * >= 4GB machines have 65536 buckets. | |
1840 | */ | |
9fb9cbb1 | 1841 | nf_conntrack_htable_size |
4481374c | 1842 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1843 | / sizeof(struct hlist_head)); |
88eab472 ML |
1844 | if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE))) |
1845 | nf_conntrack_htable_size = 65536; | |
1846 | else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
f205c5e0 PM |
1847 | nf_conntrack_htable_size = 16384; |
1848 | if (nf_conntrack_htable_size < 32) | |
1849 | nf_conntrack_htable_size = 32; | |
1850 | ||
1851 | /* Use a max. factor of four by default to get the same max as | |
1852 | * with the old struct list_heads. When a table size is given | |
1853 | * we use the old value of 8 to avoid reducing the max. | |
1854 | * entries. */ | |
1855 | max_factor = 4; | |
9fb9cbb1 | 1856 | } |
56d52d48 FW |
1857 | |
1858 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1); | |
1859 | if (!nf_conntrack_hash) | |
1860 | return -ENOMEM; | |
1861 | ||
f205c5e0 | 1862 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 1863 | |
0c5366b3 FW |
1864 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
1865 | sizeof(struct nf_conn), 0, | |
5a75cdeb | 1866 | SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL); |
0c5366b3 FW |
1867 | if (!nf_conntrack_cachep) |
1868 | goto err_cachep; | |
1869 | ||
654d0fbd | 1870 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
1871 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
1872 | nf_conntrack_max); | |
83b4dbe1 G |
1873 | |
1874 | ret = nf_conntrack_expect_init(); | |
1875 | if (ret < 0) | |
1876 | goto err_expect; | |
1877 | ||
b7ff3a1f G |
1878 | ret = nf_conntrack_acct_init(); |
1879 | if (ret < 0) | |
1880 | goto err_acct; | |
1881 | ||
73f4001a G |
1882 | ret = nf_conntrack_tstamp_init(); |
1883 | if (ret < 0) | |
1884 | goto err_tstamp; | |
1885 | ||
3fe0f943 G |
1886 | ret = nf_conntrack_ecache_init(); |
1887 | if (ret < 0) | |
1888 | goto err_ecache; | |
1889 | ||
8684094c G |
1890 | ret = nf_conntrack_timeout_init(); |
1891 | if (ret < 0) | |
1892 | goto err_timeout; | |
1893 | ||
5e615b22 G |
1894 | ret = nf_conntrack_helper_init(); |
1895 | if (ret < 0) | |
1896 | goto err_helper; | |
1897 | ||
5f69b8f5 G |
1898 | ret = nf_conntrack_labels_init(); |
1899 | if (ret < 0) | |
1900 | goto err_labels; | |
1901 | ||
41d73ec0 PM |
1902 | ret = nf_conntrack_seqadj_init(); |
1903 | if (ret < 0) | |
1904 | goto err_seqadj; | |
1905 | ||
04d87001 G |
1906 | ret = nf_conntrack_proto_init(); |
1907 | if (ret < 0) | |
1908 | goto err_proto; | |
1909 | ||
9edd7ca0 | 1910 | /* Set up fake conntrack: to never be deleted, not in any hashes */ |
b3c5163f ED |
1911 | for_each_possible_cpu(cpu) { |
1912 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
b3c5163f ED |
1913 | write_pnet(&ct->ct_net, &init_net); |
1914 | atomic_set(&ct->ct_general.use, 1); | |
1915 | } | |
9edd7ca0 | 1916 | /* - and look it like as a confirmed connection */ |
5bfddbd4 | 1917 | nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); |
b87a2f91 FW |
1918 | |
1919 | conntrack_gc_work_init(&conntrack_gc_work); | |
e0df8cae | 1920 | queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, GC_INTERVAL_MAX); |
b87a2f91 | 1921 | |
08f6547d AD |
1922 | return 0; |
1923 | ||
04d87001 | 1924 | err_proto: |
41d73ec0 PM |
1925 | nf_conntrack_seqadj_fini(); |
1926 | err_seqadj: | |
04d87001 | 1927 | nf_conntrack_labels_fini(); |
5f69b8f5 G |
1928 | err_labels: |
1929 | nf_conntrack_helper_fini(); | |
5e615b22 G |
1930 | err_helper: |
1931 | nf_conntrack_timeout_fini(); | |
8684094c G |
1932 | err_timeout: |
1933 | nf_conntrack_ecache_fini(); | |
3fe0f943 G |
1934 | err_ecache: |
1935 | nf_conntrack_tstamp_fini(); | |
73f4001a G |
1936 | err_tstamp: |
1937 | nf_conntrack_acct_fini(); | |
b7ff3a1f G |
1938 | err_acct: |
1939 | nf_conntrack_expect_fini(); | |
83b4dbe1 | 1940 | err_expect: |
0c5366b3 FW |
1941 | kmem_cache_destroy(nf_conntrack_cachep); |
1942 | err_cachep: | |
56d52d48 | 1943 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
08f6547d AD |
1944 | return ret; |
1945 | } | |
1946 | ||
f94161c1 G |
1947 | void nf_conntrack_init_end(void) |
1948 | { | |
1949 | /* For use by REJECT target */ | |
1950 | RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); | |
1951 | RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); | |
f94161c1 G |
1952 | } |
1953 | ||
8cc20198 ED |
1954 | /* |
1955 | * We need to use special "null" values, not used in hash table | |
1956 | */ | |
1957 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1958 | #define DYING_NULLS_VAL ((1<<30)+1) | |
252b3e8c | 1959 | #define TEMPLATE_NULLS_VAL ((1<<30)+2) |
8cc20198 | 1960 | |
f94161c1 | 1961 | int nf_conntrack_init_net(struct net *net) |
08f6547d | 1962 | { |
b7779d06 JDB |
1963 | int ret = -ENOMEM; |
1964 | int cpu; | |
ceceae1b | 1965 | |
08f6547d | 1966 | atomic_set(&net->ct.count, 0); |
b7779d06 JDB |
1967 | |
1968 | net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu); | |
1969 | if (!net->ct.pcpu_lists) | |
08f6547d | 1970 | goto err_stat; |
b7779d06 JDB |
1971 | |
1972 | for_each_possible_cpu(cpu) { | |
1973 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1974 | ||
1975 | spin_lock_init(&pcpu->lock); | |
1976 | INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL); | |
1977 | INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL); | |
08f6547d | 1978 | } |
5b3501fa | 1979 | |
b7779d06 JDB |
1980 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1981 | if (!net->ct.stat) | |
1982 | goto err_pcpu_lists; | |
1983 | ||
83b4dbe1 | 1984 | ret = nf_conntrack_expect_pernet_init(net); |
08f6547d AD |
1985 | if (ret < 0) |
1986 | goto err_expect; | |
b7ff3a1f | 1987 | ret = nf_conntrack_acct_pernet_init(net); |
58401572 | 1988 | if (ret < 0) |
08f6547d | 1989 | goto err_acct; |
73f4001a | 1990 | ret = nf_conntrack_tstamp_pernet_init(net); |
a992ca2a PNA |
1991 | if (ret < 0) |
1992 | goto err_tstamp; | |
3fe0f943 | 1993 | ret = nf_conntrack_ecache_pernet_init(net); |
a0891aa6 PNA |
1994 | if (ret < 0) |
1995 | goto err_ecache; | |
5e615b22 | 1996 | ret = nf_conntrack_helper_pernet_init(net); |
a9006892 EL |
1997 | if (ret < 0) |
1998 | goto err_helper; | |
04d87001 | 1999 | ret = nf_conntrack_proto_pernet_init(net); |
f94161c1 G |
2000 | if (ret < 0) |
2001 | goto err_proto; | |
08f6547d | 2002 | return 0; |
c539f017 | 2003 | |
f94161c1 | 2004 | err_proto: |
5e615b22 | 2005 | nf_conntrack_helper_pernet_fini(net); |
a9006892 | 2006 | err_helper: |
3fe0f943 | 2007 | nf_conntrack_ecache_pernet_fini(net); |
a0891aa6 | 2008 | err_ecache: |
73f4001a | 2009 | nf_conntrack_tstamp_pernet_fini(net); |
a992ca2a | 2010 | err_tstamp: |
b7ff3a1f | 2011 | nf_conntrack_acct_pernet_fini(net); |
08f6547d | 2012 | err_acct: |
83b4dbe1 | 2013 | nf_conntrack_expect_pernet_fini(net); |
08f6547d | 2014 | err_expect: |
0d55af87 | 2015 | free_percpu(net->ct.stat); |
b7779d06 JDB |
2016 | err_pcpu_lists: |
2017 | free_percpu(net->ct.pcpu_lists); | |
0d55af87 | 2018 | err_stat: |
08f6547d AD |
2019 | return ret; |
2020 | } |