]>
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 */ |
e5072053 FW |
88 | #define GC_MAX_BUCKETS_DIV 128u |
89 | /* upper bound of full table scan */ | |
90 | #define GC_MAX_SCAN_JIFFIES (16u * HZ) | |
91 | /* desired ratio of entries found to be expired */ | |
92 | #define GC_EVICT_RATIO 50u | |
b87a2f91 FW |
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; |
141658fb | 183 | static unsigned int nf_conntrack_hash_rnd __read_mostly; |
9fb9cbb1 | 184 | |
1b8c8a9f FW |
185 | static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, |
186 | const struct net *net) | |
9fb9cbb1 | 187 | { |
0794935e | 188 | unsigned int n; |
1b8c8a9f | 189 | u32 seed; |
0794935e | 190 | |
141658fb FW |
191 | get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd)); |
192 | ||
0794935e PM |
193 | /* The direction must be ignored, so we hash everything up to the |
194 | * destination ports (which is a multiple of 4) and treat the last | |
195 | * three bytes manually. | |
196 | */ | |
1b8c8a9f | 197 | seed = nf_conntrack_hash_rnd ^ net_hash_mix(net); |
0794935e | 198 | n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); |
1b8c8a9f | 199 | return jhash2((u32 *)tuple, n, seed ^ |
99f07e91 CG |
200 | (((__force __u16)tuple->dst.u.all << 16) | |
201 | tuple->dst.protonum)); | |
202 | } | |
203 | ||
56d52d48 | 204 | static u32 scale_hash(u32 hash) |
99f07e91 | 205 | { |
56d52d48 | 206 | return reciprocal_scale(hash, nf_conntrack_htable_size); |
99f07e91 | 207 | } |
0794935e | 208 | |
1b8c8a9f FW |
209 | static u32 __hash_conntrack(const struct net *net, |
210 | const struct nf_conntrack_tuple *tuple, | |
211 | unsigned int size) | |
99f07e91 | 212 | { |
1b8c8a9f | 213 | return reciprocal_scale(hash_conntrack_raw(tuple, net), size); |
9fb9cbb1 YK |
214 | } |
215 | ||
1b8c8a9f FW |
216 | static u32 hash_conntrack(const struct net *net, |
217 | const struct nf_conntrack_tuple *tuple) | |
9fb9cbb1 | 218 | { |
56d52d48 | 219 | return scale_hash(hash_conntrack_raw(tuple, net)); |
9fb9cbb1 YK |
220 | } |
221 | ||
5f2b4c90 | 222 | bool |
9fb9cbb1 YK |
223 | nf_ct_get_tuple(const struct sk_buff *skb, |
224 | unsigned int nhoff, | |
225 | unsigned int dataoff, | |
226 | u_int16_t l3num, | |
227 | u_int8_t protonum, | |
a31f1adc | 228 | struct net *net, |
9fb9cbb1 YK |
229 | struct nf_conntrack_tuple *tuple, |
230 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 231 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 232 | { |
443a70d5 | 233 | memset(tuple, 0, sizeof(*tuple)); |
9fb9cbb1 YK |
234 | |
235 | tuple->src.l3num = l3num; | |
236 | if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) | |
5f2b4c90 | 237 | return false; |
9fb9cbb1 YK |
238 | |
239 | tuple->dst.protonum = protonum; | |
240 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; | |
241 | ||
a31f1adc | 242 | return l4proto->pkt_to_tuple(skb, dataoff, net, tuple); |
9fb9cbb1 | 243 | } |
13b18339 | 244 | EXPORT_SYMBOL_GPL(nf_ct_get_tuple); |
9fb9cbb1 | 245 | |
5f2b4c90 | 246 | bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, |
a31f1adc EB |
247 | u_int16_t l3num, |
248 | struct net *net, struct nf_conntrack_tuple *tuple) | |
e2a3123f YK |
249 | { |
250 | struct nf_conntrack_l3proto *l3proto; | |
251 | struct nf_conntrack_l4proto *l4proto; | |
252 | unsigned int protoff; | |
253 | u_int8_t protonum; | |
254 | int ret; | |
255 | ||
256 | rcu_read_lock(); | |
257 | ||
258 | l3proto = __nf_ct_l3proto_find(l3num); | |
259 | ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); | |
260 | if (ret != NF_ACCEPT) { | |
261 | rcu_read_unlock(); | |
5f2b4c90 | 262 | return false; |
e2a3123f YK |
263 | } |
264 | ||
265 | l4proto = __nf_ct_l4proto_find(l3num, protonum); | |
266 | ||
a31f1adc | 267 | ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple, |
e2a3123f YK |
268 | l3proto, l4proto); |
269 | ||
270 | rcu_read_unlock(); | |
271 | return ret; | |
272 | } | |
273 | EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); | |
274 | ||
5f2b4c90 | 275 | bool |
9fb9cbb1 YK |
276 | nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, |
277 | const struct nf_conntrack_tuple *orig, | |
278 | const struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 279 | const struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 280 | { |
443a70d5 | 281 | memset(inverse, 0, sizeof(*inverse)); |
9fb9cbb1 YK |
282 | |
283 | inverse->src.l3num = orig->src.l3num; | |
284 | if (l3proto->invert_tuple(inverse, orig) == 0) | |
5f2b4c90 | 285 | return false; |
9fb9cbb1 YK |
286 | |
287 | inverse->dst.dir = !orig->dst.dir; | |
288 | ||
289 | inverse->dst.protonum = orig->dst.protonum; | |
605dcad6 | 290 | return l4proto->invert_tuple(inverse, orig); |
9fb9cbb1 | 291 | } |
13b18339 | 292 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); |
9fb9cbb1 | 293 | |
9fb9cbb1 YK |
294 | static void |
295 | clean_from_lists(struct nf_conn *ct) | |
296 | { | |
0d53778e | 297 | pr_debug("clean_from_lists(%p)\n", ct); |
ea781f19 ED |
298 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); |
299 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); | |
9fb9cbb1 YK |
300 | |
301 | /* Destroy all pending expectations */ | |
c1d10adb | 302 | nf_ct_remove_expectations(ct); |
9fb9cbb1 YK |
303 | } |
304 | ||
b7779d06 JDB |
305 | /* must be called with local_bh_disable */ |
306 | static void nf_ct_add_to_dying_list(struct nf_conn *ct) | |
307 | { | |
308 | struct ct_pcpu *pcpu; | |
309 | ||
310 | /* add this conntrack to the (per cpu) dying list */ | |
311 | ct->cpu = smp_processor_id(); | |
312 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
313 | ||
314 | spin_lock(&pcpu->lock); | |
315 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
316 | &pcpu->dying); | |
317 | spin_unlock(&pcpu->lock); | |
318 | } | |
319 | ||
320 | /* must be called with local_bh_disable */ | |
321 | static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct) | |
322 | { | |
323 | struct ct_pcpu *pcpu; | |
324 | ||
325 | /* add this conntrack to the (per cpu) unconfirmed list */ | |
326 | ct->cpu = smp_processor_id(); | |
327 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
328 | ||
329 | spin_lock(&pcpu->lock); | |
330 | hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, | |
331 | &pcpu->unconfirmed); | |
332 | spin_unlock(&pcpu->lock); | |
333 | } | |
334 | ||
335 | /* must be called with local_bh_disable */ | |
336 | static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct) | |
337 | { | |
338 | struct ct_pcpu *pcpu; | |
339 | ||
340 | /* We overload first tuple to link into unconfirmed or dying list.*/ | |
341 | pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); | |
342 | ||
343 | spin_lock(&pcpu->lock); | |
344 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); | |
345 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
346 | spin_unlock(&pcpu->lock); | |
347 | } | |
348 | ||
30322309 FW |
349 | #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK) |
350 | ||
0838aa7f | 351 | /* Released via destroy_conntrack() */ |
308ac914 DB |
352 | struct nf_conn *nf_ct_tmpl_alloc(struct net *net, |
353 | const struct nf_conntrack_zone *zone, | |
354 | gfp_t flags) | |
0838aa7f | 355 | { |
30322309 | 356 | struct nf_conn *tmpl, *p; |
0838aa7f | 357 | |
30322309 FW |
358 | if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) { |
359 | tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags); | |
360 | if (!tmpl) | |
361 | return NULL; | |
362 | ||
363 | p = tmpl; | |
364 | tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p); | |
365 | if (tmpl != p) { | |
366 | tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p); | |
367 | tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p; | |
368 | } | |
369 | } else { | |
370 | tmpl = kzalloc(sizeof(*tmpl), flags); | |
371 | if (!tmpl) | |
372 | return NULL; | |
373 | } | |
0838aa7f PNA |
374 | |
375 | tmpl->status = IPS_TEMPLATE; | |
376 | write_pnet(&tmpl->ct_net, net); | |
6c8dee98 | 377 | nf_ct_zone_add(tmpl, zone); |
0838aa7f PNA |
378 | atomic_set(&tmpl->ct_general.use, 0); |
379 | ||
380 | return tmpl; | |
0838aa7f PNA |
381 | } |
382 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc); | |
383 | ||
9cf94eab | 384 | void nf_ct_tmpl_free(struct nf_conn *tmpl) |
0838aa7f PNA |
385 | { |
386 | nf_ct_ext_destroy(tmpl); | |
387 | nf_ct_ext_free(tmpl); | |
30322309 FW |
388 | |
389 | if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) | |
390 | kfree((char *)tmpl - tmpl->proto.tmpl_padto); | |
391 | else | |
392 | kfree(tmpl); | |
0838aa7f | 393 | } |
9cf94eab | 394 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_free); |
0838aa7f | 395 | |
9fb9cbb1 YK |
396 | static void |
397 | destroy_conntrack(struct nf_conntrack *nfct) | |
398 | { | |
399 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 400 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 401 | |
0d53778e | 402 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 | 403 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
9fb9cbb1 | 404 | |
0838aa7f PNA |
405 | if (unlikely(nf_ct_is_template(ct))) { |
406 | nf_ct_tmpl_free(ct); | |
407 | return; | |
408 | } | |
923f4902 | 409 | rcu_read_lock(); |
5e8fbe2a | 410 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
4b4ceb9d | 411 | if (l4proto->destroy) |
605dcad6 | 412 | l4proto->destroy(ct); |
9fb9cbb1 | 413 | |
982d9a9c | 414 | rcu_read_unlock(); |
9fb9cbb1 | 415 | |
ca7433df | 416 | local_bh_disable(); |
9fb9cbb1 YK |
417 | /* Expectations will have been removed in clean_from_lists, |
418 | * except TFTP can create an expectation on the first packet, | |
419 | * before connection is in the list, so we need to clean here, | |
ca7433df JDB |
420 | * too. |
421 | */ | |
c1d10adb | 422 | nf_ct_remove_expectations(ct); |
9fb9cbb1 | 423 | |
b7779d06 | 424 | nf_ct_del_from_dying_or_unconfirmed_list(ct); |
9fb9cbb1 | 425 | |
ca7433df | 426 | local_bh_enable(); |
9fb9cbb1 YK |
427 | |
428 | if (ct->master) | |
429 | nf_ct_put(ct->master); | |
430 | ||
0d53778e | 431 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
432 | nf_conntrack_free(ct); |
433 | } | |
434 | ||
02982c27 | 435 | static void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 436 | { |
0d55af87 | 437 | struct net *net = nf_ct_net(ct); |
93bb0ceb | 438 | unsigned int hash, reply_hash; |
93bb0ceb | 439 | unsigned int sequence; |
9fb9cbb1 | 440 | |
9858a3ae | 441 | nf_ct_helper_destroy(ct); |
93bb0ceb JDB |
442 | |
443 | local_bh_disable(); | |
444 | do { | |
a3efd812 | 445 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 446 | hash = hash_conntrack(net, |
93bb0ceb | 447 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 448 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
449 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
450 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
451 | ||
9fb9cbb1 | 452 | clean_from_lists(ct); |
93bb0ceb JDB |
453 | nf_conntrack_double_unlock(hash, reply_hash); |
454 | ||
b7779d06 | 455 | nf_ct_add_to_dying_list(ct); |
93bb0ceb | 456 | |
93bb0ceb | 457 | local_bh_enable(); |
dd7669a9 | 458 | } |
dd7669a9 | 459 | |
02982c27 | 460 | bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report) |
dd7669a9 | 461 | { |
a992ca2a PNA |
462 | struct nf_conn_tstamp *tstamp; |
463 | ||
f330a7fd FW |
464 | if (test_and_set_bit(IPS_DYING_BIT, &ct->status)) |
465 | return false; | |
466 | ||
a992ca2a PNA |
467 | tstamp = nf_conn_tstamp_find(ct); |
468 | if (tstamp && tstamp->stop == 0) | |
d2de875c | 469 | tstamp->stop = ktime_get_real_ns(); |
dd7669a9 | 470 | |
9500507c FW |
471 | if (nf_conntrack_event_report(IPCT_DESTROY, ct, |
472 | portid, report) < 0) { | |
f330a7fd FW |
473 | /* destroy event was not delivered. nf_ct_put will |
474 | * be done by event cache worker on redelivery. | |
475 | */ | |
dd7669a9 | 476 | nf_ct_delete_from_lists(ct); |
9500507c | 477 | nf_conntrack_ecache_delayed_work(nf_ct_net(ct)); |
02982c27 | 478 | return false; |
dd7669a9 | 479 | } |
9500507c FW |
480 | |
481 | nf_conntrack_ecache_work(nf_ct_net(ct)); | |
dd7669a9 | 482 | nf_ct_delete_from_lists(ct); |
9fb9cbb1 | 483 | nf_ct_put(ct); |
02982c27 FW |
484 | return true; |
485 | } | |
486 | EXPORT_SYMBOL_GPL(nf_ct_delete); | |
487 | ||
c6825c09 AV |
488 | static inline bool |
489 | nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, | |
308ac914 | 490 | const struct nf_conntrack_tuple *tuple, |
e0c7d472 FW |
491 | const struct nf_conntrack_zone *zone, |
492 | const struct net *net) | |
c6825c09 AV |
493 | { |
494 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
495 | ||
496 | /* A conntrack can be recreated with the equal tuple, | |
497 | * so we need to check that the conntrack is confirmed | |
498 | */ | |
499 | return nf_ct_tuple_equal(tuple, &h->tuple) && | |
deedb590 | 500 | nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) && |
e0c7d472 FW |
501 | nf_ct_is_confirmed(ct) && |
502 | net_eq(net, nf_ct_net(ct)); | |
c6825c09 AV |
503 | } |
504 | ||
f330a7fd FW |
505 | /* caller must hold rcu readlock and none of the nf_conntrack_locks */ |
506 | static void nf_ct_gc_expired(struct nf_conn *ct) | |
507 | { | |
508 | if (!atomic_inc_not_zero(&ct->ct_general.use)) | |
509 | return; | |
510 | ||
511 | if (nf_ct_should_gc(ct)) | |
512 | nf_ct_kill(ct); | |
513 | ||
514 | nf_ct_put(ct); | |
515 | } | |
516 | ||
ea781f19 ED |
517 | /* |
518 | * Warning : | |
519 | * - Caller must take a reference on returned object | |
520 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
ea781f19 | 521 | */ |
99f07e91 | 522 | static struct nf_conntrack_tuple_hash * |
308ac914 | 523 | ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 524 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
525 | { |
526 | struct nf_conntrack_tuple_hash *h; | |
5e3c61f9 | 527 | struct hlist_nulls_head *ct_hash; |
ea781f19 | 528 | struct hlist_nulls_node *n; |
92e47ba8 | 529 | unsigned int bucket, hsize; |
9fb9cbb1 | 530 | |
ea781f19 | 531 | begin: |
92e47ba8 LZ |
532 | nf_conntrack_get_ht(&ct_hash, &hsize); |
533 | bucket = reciprocal_scale(hash, hsize); | |
5e3c61f9 FW |
534 | |
535 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) { | |
f330a7fd FW |
536 | struct nf_conn *ct; |
537 | ||
538 | ct = nf_ct_tuplehash_to_ctrack(h); | |
539 | if (nf_ct_is_expired(ct)) { | |
540 | nf_ct_gc_expired(ct); | |
541 | continue; | |
542 | } | |
543 | ||
544 | if (nf_ct_is_dying(ct)) | |
545 | continue; | |
546 | ||
8e8118f8 | 547 | if (nf_ct_key_equal(h, tuple, zone, net)) |
9fb9cbb1 | 548 | return h; |
9fb9cbb1 | 549 | } |
ea781f19 ED |
550 | /* |
551 | * if the nulls value we got at the end of this lookup is | |
552 | * not the expected one, we must restart lookup. | |
553 | * We probably met an item that was moved to another chain. | |
554 | */ | |
99f07e91 | 555 | if (get_nulls_value(n) != bucket) { |
2cf12348 | 556 | NF_CT_STAT_INC_ATOMIC(net, search_restart); |
ea781f19 | 557 | goto begin; |
af740b2c | 558 | } |
9fb9cbb1 YK |
559 | |
560 | return NULL; | |
561 | } | |
99f07e91 | 562 | |
9fb9cbb1 | 563 | /* Find a connection corresponding to a tuple. */ |
99f07e91 | 564 | static struct nf_conntrack_tuple_hash * |
308ac914 | 565 | __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 566 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
567 | { |
568 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 569 | struct nf_conn *ct; |
9fb9cbb1 | 570 | |
76507f69 | 571 | rcu_read_lock(); |
ea781f19 | 572 | begin: |
99f07e91 | 573 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
574 | if (h) { |
575 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
576 | if (unlikely(nf_ct_is_dying(ct) || |
577 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 578 | h = NULL; |
ea781f19 | 579 | else { |
e0c7d472 | 580 | if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) { |
ea781f19 ED |
581 | nf_ct_put(ct); |
582 | goto begin; | |
583 | } | |
584 | } | |
76507f69 PM |
585 | } |
586 | rcu_read_unlock(); | |
9fb9cbb1 YK |
587 | |
588 | return h; | |
589 | } | |
99f07e91 CG |
590 | |
591 | struct nf_conntrack_tuple_hash * | |
308ac914 | 592 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 CG |
593 | const struct nf_conntrack_tuple *tuple) |
594 | { | |
595 | return __nf_conntrack_find_get(net, zone, tuple, | |
1b8c8a9f | 596 | hash_conntrack_raw(tuple, net)); |
99f07e91 | 597 | } |
13b18339 | 598 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 599 | |
c1d10adb PNA |
600 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
601 | unsigned int hash, | |
b476b72a | 602 | unsigned int reply_hash) |
c1d10adb | 603 | { |
ea781f19 | 604 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
56d52d48 | 605 | &nf_conntrack_hash[hash]); |
ea781f19 | 606 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
56d52d48 | 607 | &nf_conntrack_hash[reply_hash]); |
c1d10adb PNA |
608 | } |
609 | ||
7d367e06 JK |
610 | int |
611 | nf_conntrack_hash_check_insert(struct nf_conn *ct) | |
c1d10adb | 612 | { |
308ac914 | 613 | const struct nf_conntrack_zone *zone; |
d696c7bd | 614 | struct net *net = nf_ct_net(ct); |
b476b72a | 615 | unsigned int hash, reply_hash; |
7d367e06 JK |
616 | struct nf_conntrack_tuple_hash *h; |
617 | struct hlist_nulls_node *n; | |
93bb0ceb | 618 | unsigned int sequence; |
c1d10adb | 619 | |
5d0aa2cc | 620 | zone = nf_ct_zone(ct); |
7d367e06 | 621 | |
93bb0ceb JDB |
622 | local_bh_disable(); |
623 | do { | |
a3efd812 | 624 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 625 | hash = hash_conntrack(net, |
93bb0ceb | 626 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 627 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
628 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
629 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
7d367e06 JK |
630 | |
631 | /* See if there's one in the list already, including reverse */ | |
56d52d48 | 632 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 633 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 634 | zone, net)) |
7d367e06 | 635 | goto out; |
86804348 | 636 | |
56d52d48 | 637 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 638 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 639 | zone, net)) |
7d367e06 | 640 | goto out; |
c1d10adb | 641 | |
e53376be PNA |
642 | smp_wmb(); |
643 | /* The caller holds a reference to this object */ | |
644 | atomic_set(&ct->ct_general.use, 2); | |
b476b72a | 645 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 646 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 647 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 648 | local_bh_enable(); |
7d367e06 JK |
649 | return 0; |
650 | ||
651 | out: | |
93bb0ceb | 652 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 653 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 654 | local_bh_enable(); |
7d367e06 | 655 | return -EEXIST; |
c1d10adb | 656 | } |
7d367e06 | 657 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); |
c1d10adb | 658 | |
ba76738c PNA |
659 | static inline void nf_ct_acct_update(struct nf_conn *ct, |
660 | enum ip_conntrack_info ctinfo, | |
661 | unsigned int len) | |
662 | { | |
663 | struct nf_conn_acct *acct; | |
664 | ||
665 | acct = nf_conn_acct_find(ct); | |
666 | if (acct) { | |
667 | struct nf_conn_counter *counter = acct->counter; | |
668 | ||
669 | atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); | |
670 | atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes); | |
671 | } | |
672 | } | |
673 | ||
71d8c47f PNA |
674 | static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo, |
675 | const struct nf_conn *loser_ct) | |
676 | { | |
677 | struct nf_conn_acct *acct; | |
678 | ||
679 | acct = nf_conn_acct_find(loser_ct); | |
680 | if (acct) { | |
681 | struct nf_conn_counter *counter = acct->counter; | |
71d8c47f PNA |
682 | unsigned int bytes; |
683 | ||
684 | /* u32 should be fine since we must have seen one packet. */ | |
685 | bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes); | |
686 | nf_ct_acct_update(ct, ctinfo, bytes); | |
687 | } | |
688 | } | |
689 | ||
690 | /* Resolve race on insertion if this protocol allows this. */ | |
691 | static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb, | |
692 | enum ip_conntrack_info ctinfo, | |
693 | struct nf_conntrack_tuple_hash *h) | |
694 | { | |
695 | /* This is the conntrack entry already in hashes that won race. */ | |
696 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
697 | struct nf_conntrack_l4proto *l4proto; | |
698 | ||
699 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); | |
700 | if (l4proto->allow_clash && | |
6e699867 | 701 | ((ct->status & IPS_NAT_DONE_MASK) == 0) && |
71d8c47f PNA |
702 | !nf_ct_is_dying(ct) && |
703 | atomic_inc_not_zero(&ct->ct_general.use)) { | |
97a6ad13 FW |
704 | enum ip_conntrack_info oldinfo; |
705 | struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo); | |
706 | ||
707 | nf_ct_acct_merge(ct, ctinfo, loser_ct); | |
708 | nf_conntrack_put(&loser_ct->ct_general); | |
c74454fa | 709 | nf_ct_set(skb, ct, oldinfo); |
71d8c47f PNA |
710 | return NF_ACCEPT; |
711 | } | |
712 | NF_CT_STAT_INC(net, drop); | |
713 | return NF_DROP; | |
714 | } | |
715 | ||
9fb9cbb1 YK |
716 | /* Confirm a connection given skb; places it in hash table */ |
717 | int | |
3db05fea | 718 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 | 719 | { |
308ac914 | 720 | const struct nf_conntrack_zone *zone; |
b476b72a | 721 | unsigned int hash, reply_hash; |
df0933dc | 722 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 723 | struct nf_conn *ct; |
df0933dc | 724 | struct nf_conn_help *help; |
a992ca2a | 725 | struct nf_conn_tstamp *tstamp; |
ea781f19 | 726 | struct hlist_nulls_node *n; |
9fb9cbb1 | 727 | enum ip_conntrack_info ctinfo; |
400dad39 | 728 | struct net *net; |
93bb0ceb | 729 | unsigned int sequence; |
71d8c47f | 730 | int ret = NF_DROP; |
9fb9cbb1 | 731 | |
3db05fea | 732 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 733 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
734 | |
735 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
736 | ICMP/TCP RST packets in other direction. Actual packet | |
737 | which created connection will be IP_CT_NEW or for an | |
738 | expected connection, IP_CT_RELATED. */ | |
739 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
740 | return NF_ACCEPT; | |
741 | ||
5d0aa2cc | 742 | zone = nf_ct_zone(ct); |
93bb0ceb JDB |
743 | local_bh_disable(); |
744 | ||
745 | do { | |
a3efd812 | 746 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb JDB |
747 | /* reuse the hash saved before */ |
748 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
56d52d48 | 749 | hash = scale_hash(hash); |
deedb590 | 750 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
751 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
752 | ||
753 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
9fb9cbb1 YK |
754 | |
755 | /* We're not in hash table, and we refuse to set up related | |
93bb0ceb JDB |
756 | * connections for unconfirmed conns. But packet copies and |
757 | * REJECT will give spurious warnings here. | |
758 | */ | |
9fb9cbb1 YK |
759 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ |
760 | ||
25985edc | 761 | /* No external references means no one else could have |
93bb0ceb JDB |
762 | * confirmed us. |
763 | */ | |
9fb9cbb1 | 764 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); |
0d53778e | 765 | pr_debug("Confirming conntrack %p\n", ct); |
8ca3f5e9 PNA |
766 | /* We have to check the DYING flag after unlink to prevent |
767 | * a race against nf_ct_get_next_corpse() possibly called from | |
768 | * user context, else we insert an already 'dead' hash, blocking | |
769 | * further use of that particular connection -JM. | |
770 | */ | |
771 | nf_ct_del_from_dying_or_unconfirmed_list(ct); | |
772 | ||
71d8c47f PNA |
773 | if (unlikely(nf_ct_is_dying(ct))) { |
774 | nf_ct_add_to_dying_list(ct); | |
775 | goto dying; | |
776 | } | |
fc350777 | 777 | |
9fb9cbb1 YK |
778 | /* See if there's one in the list already, including reverse: |
779 | NAT could have grabbed it without realizing, since we're | |
780 | not in the hash. If there is, we lost race. */ | |
56d52d48 | 781 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 782 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 783 | zone, net)) |
df0933dc | 784 | goto out; |
86804348 | 785 | |
56d52d48 | 786 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 787 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 788 | zone, net)) |
df0933dc | 789 | goto out; |
9fb9cbb1 | 790 | |
df0933dc PM |
791 | /* Timer relative to confirmation time, not original |
792 | setting time, otherwise we'd get timer wrap in | |
793 | weird delay cases. */ | |
f330a7fd | 794 | ct->timeout += nfct_time_stamp; |
df0933dc | 795 | atomic_inc(&ct->ct_general.use); |
45eec341 | 796 | ct->status |= IPS_CONFIRMED; |
5c8ec910 | 797 | |
a992ca2a PNA |
798 | /* set conntrack timestamp, if enabled. */ |
799 | tstamp = nf_conn_tstamp_find(ct); | |
800 | if (tstamp) { | |
2456e855 | 801 | if (skb->tstamp == 0) |
e3192690 | 802 | __net_timestamp(skb); |
a992ca2a PNA |
803 | |
804 | tstamp->start = ktime_to_ns(skb->tstamp); | |
805 | } | |
5c8ec910 PM |
806 | /* Since the lookup is lockless, hash insertion must be done after |
807 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
808 | * guarantee that no other CPU can find the conntrack before the above | |
809 | * stores are visible. | |
810 | */ | |
b476b72a | 811 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 812 | nf_conntrack_double_unlock(hash, reply_hash); |
93bb0ceb | 813 | local_bh_enable(); |
5c8ec910 | 814 | |
df0933dc PM |
815 | help = nfct_help(ct); |
816 | if (help && help->helper) | |
a71996fc | 817 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 818 | |
df0933dc | 819 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 820 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 821 | return NF_ACCEPT; |
9fb9cbb1 | 822 | |
df0933dc | 823 | out: |
8ca3f5e9 | 824 | nf_ct_add_to_dying_list(ct); |
71d8c47f PNA |
825 | ret = nf_ct_resolve_clash(net, skb, ctinfo, h); |
826 | dying: | |
93bb0ceb | 827 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 828 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 829 | local_bh_enable(); |
71d8c47f | 830 | return ret; |
9fb9cbb1 | 831 | } |
13b18339 | 832 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
833 | |
834 | /* Returns true if a connection correspondings to the tuple (required | |
835 | for NAT). */ | |
836 | int | |
837 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
838 | const struct nf_conn *ignored_conntrack) | |
839 | { | |
400dad39 | 840 | struct net *net = nf_ct_net(ignored_conntrack); |
308ac914 | 841 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 | 842 | struct nf_conntrack_tuple_hash *h; |
5e3c61f9 | 843 | struct hlist_nulls_head *ct_hash; |
92e47ba8 | 844 | unsigned int hash, hsize; |
ea781f19 | 845 | struct hlist_nulls_node *n; |
5d0aa2cc | 846 | struct nf_conn *ct; |
308ac914 DB |
847 | |
848 | zone = nf_ct_zone(ignored_conntrack); | |
9fb9cbb1 | 849 | |
2cf12348 | 850 | rcu_read_lock(); |
95a8d19f | 851 | begin: |
92e47ba8 LZ |
852 | nf_conntrack_get_ht(&ct_hash, &hsize); |
853 | hash = __hash_conntrack(net, tuple, hsize); | |
5e3c61f9 FW |
854 | |
855 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) { | |
5d0aa2cc | 856 | ct = nf_ct_tuplehash_to_ctrack(h); |
f330a7fd FW |
857 | |
858 | if (ct == ignored_conntrack) | |
859 | continue; | |
860 | ||
861 | if (nf_ct_is_expired(ct)) { | |
862 | nf_ct_gc_expired(ct); | |
863 | continue; | |
864 | } | |
865 | ||
866 | if (nf_ct_key_equal(h, tuple, zone, net)) { | |
2cf12348 FW |
867 | NF_CT_STAT_INC_ATOMIC(net, found); |
868 | rcu_read_unlock(); | |
ba419aff PM |
869 | return 1; |
870 | } | |
ba419aff | 871 | } |
95a8d19f FW |
872 | |
873 | if (get_nulls_value(n) != hash) { | |
874 | NF_CT_STAT_INC_ATOMIC(net, search_restart); | |
875 | goto begin; | |
876 | } | |
877 | ||
2cf12348 | 878 | rcu_read_unlock(); |
9fb9cbb1 | 879 | |
ba419aff | 880 | return 0; |
9fb9cbb1 | 881 | } |
13b18339 | 882 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 883 | |
7ae7730f PM |
884 | #define NF_CT_EVICTION_RANGE 8 |
885 | ||
9fb9cbb1 YK |
886 | /* There's a small race here where we may free a just-assured |
887 | connection. Too bad: we're in trouble anyway. */ | |
242922a0 FW |
888 | static unsigned int early_drop_list(struct net *net, |
889 | struct hlist_nulls_head *head) | |
9fb9cbb1 | 890 | { |
9fb9cbb1 | 891 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 892 | struct hlist_nulls_node *n; |
242922a0 FW |
893 | unsigned int drops = 0; |
894 | struct nf_conn *tmp; | |
3e86638e | 895 | |
242922a0 FW |
896 | hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) { |
897 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
9fb9cbb1 | 898 | |
f330a7fd FW |
899 | if (nf_ct_is_expired(tmp)) { |
900 | nf_ct_gc_expired(tmp); | |
901 | continue; | |
902 | } | |
903 | ||
242922a0 FW |
904 | if (test_bit(IPS_ASSURED_BIT, &tmp->status) || |
905 | !net_eq(nf_ct_net(tmp), net) || | |
906 | nf_ct_is_dying(tmp)) | |
907 | continue; | |
76507f69 | 908 | |
242922a0 FW |
909 | if (!atomic_inc_not_zero(&tmp->ct_general.use)) |
910 | continue; | |
76507f69 | 911 | |
242922a0 FW |
912 | /* kill only if still in same netns -- might have moved due to |
913 | * SLAB_DESTROY_BY_RCU rules. | |
914 | * | |
915 | * We steal the timer reference. If that fails timer has | |
916 | * already fired or someone else deleted it. Just drop ref | |
917 | * and move to next entry. | |
918 | */ | |
919 | if (net_eq(nf_ct_net(tmp), net) && | |
920 | nf_ct_is_confirmed(tmp) && | |
242922a0 FW |
921 | nf_ct_delete(tmp, 0, 0)) |
922 | drops++; | |
923 | ||
924 | nf_ct_put(tmp); | |
9fb9cbb1 | 925 | } |
3e86638e | 926 | |
242922a0 FW |
927 | return drops; |
928 | } | |
9fb9cbb1 | 929 | |
242922a0 FW |
930 | static noinline int early_drop(struct net *net, unsigned int _hash) |
931 | { | |
932 | unsigned int i; | |
9fb9cbb1 | 933 | |
242922a0 FW |
934 | for (i = 0; i < NF_CT_EVICTION_RANGE; i++) { |
935 | struct hlist_nulls_head *ct_hash; | |
92e47ba8 | 936 | unsigned int hash, hsize, drops; |
242922a0 | 937 | |
3101e0fc | 938 | rcu_read_lock(); |
92e47ba8 LZ |
939 | nf_conntrack_get_ht(&ct_hash, &hsize); |
940 | hash = reciprocal_scale(_hash++, hsize); | |
242922a0 FW |
941 | |
942 | drops = early_drop_list(net, &ct_hash[hash]); | |
3101e0fc LZ |
943 | rcu_read_unlock(); |
944 | ||
242922a0 FW |
945 | if (drops) { |
946 | NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops); | |
947 | return true; | |
74138511 | 948 | } |
9fb9cbb1 | 949 | } |
3e86638e | 950 | |
242922a0 | 951 | return false; |
9fb9cbb1 YK |
952 | } |
953 | ||
b87a2f91 FW |
954 | static void gc_worker(struct work_struct *work) |
955 | { | |
e5072053 | 956 | unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u); |
b87a2f91 | 957 | unsigned int i, goal, buckets = 0, expired_count = 0; |
b87a2f91 | 958 | struct conntrack_gc_work *gc_work; |
e0df8cae FW |
959 | unsigned int ratio, scanned = 0; |
960 | unsigned long next_run; | |
b87a2f91 FW |
961 | |
962 | gc_work = container_of(work, struct conntrack_gc_work, dwork.work); | |
963 | ||
e0df8cae | 964 | goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV; |
b87a2f91 FW |
965 | i = gc_work->last_bucket; |
966 | ||
967 | do { | |
968 | struct nf_conntrack_tuple_hash *h; | |
969 | struct hlist_nulls_head *ct_hash; | |
970 | struct hlist_nulls_node *n; | |
971 | unsigned int hashsz; | |
972 | struct nf_conn *tmp; | |
973 | ||
974 | i++; | |
975 | rcu_read_lock(); | |
976 | ||
977 | nf_conntrack_get_ht(&ct_hash, &hashsz); | |
978 | if (i >= hashsz) | |
979 | i = 0; | |
980 | ||
981 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) { | |
982 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
983 | ||
c023c0e4 | 984 | scanned++; |
b87a2f91 FW |
985 | if (nf_ct_is_expired(tmp)) { |
986 | nf_ct_gc_expired(tmp); | |
987 | expired_count++; | |
988 | continue; | |
989 | } | |
990 | } | |
991 | ||
992 | /* could check get_nulls_value() here and restart if ct | |
993 | * was moved to another chain. But given gc is best-effort | |
994 | * we will just continue with next hash slot. | |
995 | */ | |
996 | rcu_read_unlock(); | |
997 | cond_resched_rcu_qs(); | |
524b698d | 998 | } while (++buckets < goal); |
b87a2f91 FW |
999 | |
1000 | if (gc_work->exiting) | |
1001 | return; | |
1002 | ||
e0df8cae FW |
1003 | /* |
1004 | * Eviction will normally happen from the packet path, and not | |
1005 | * from this gc worker. | |
1006 | * | |
1007 | * This worker is only here to reap expired entries when system went | |
1008 | * idle after a busy period. | |
1009 | * | |
1010 | * The heuristics below are supposed to balance conflicting goals: | |
1011 | * | |
1012 | * 1. Minimize time until we notice a stale entry | |
1013 | * 2. Maximize scan intervals to not waste cycles | |
1014 | * | |
e5072053 | 1015 | * Normally, expire ratio will be close to 0. |
e0df8cae | 1016 | * |
e5072053 FW |
1017 | * As soon as a sizeable fraction of the entries have expired |
1018 | * increase scan frequency. | |
e0df8cae | 1019 | */ |
c023c0e4 | 1020 | ratio = scanned ? expired_count * 100 / scanned : 0; |
e5072053 FW |
1021 | if (ratio > GC_EVICT_RATIO) { |
1022 | gc_work->next_gc_run = min_interval; | |
e0df8cae | 1023 | } else { |
e5072053 | 1024 | unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV; |
e0df8cae | 1025 | |
e5072053 FW |
1026 | BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0); |
1027 | ||
1028 | gc_work->next_gc_run += min_interval; | |
1029 | if (gc_work->next_gc_run > max) | |
1030 | gc_work->next_gc_run = max; | |
e0df8cae | 1031 | } |
c023c0e4 | 1032 | |
e5072053 | 1033 | next_run = gc_work->next_gc_run; |
b87a2f91 | 1034 | gc_work->last_bucket = i; |
e0df8cae | 1035 | queue_delayed_work(system_long_wq, &gc_work->dwork, next_run); |
b87a2f91 FW |
1036 | } |
1037 | ||
1038 | static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work) | |
1039 | { | |
1040 | INIT_DELAYED_WORK(&gc_work->dwork, gc_worker); | |
e5072053 | 1041 | gc_work->next_gc_run = HZ; |
b87a2f91 FW |
1042 | gc_work->exiting = false; |
1043 | } | |
1044 | ||
99f07e91 | 1045 | static struct nf_conn * |
308ac914 DB |
1046 | __nf_conntrack_alloc(struct net *net, |
1047 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1048 | const struct nf_conntrack_tuple *orig, |
1049 | const struct nf_conntrack_tuple *repl, | |
1050 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 1051 | { |
cd7fcbf1 | 1052 | struct nf_conn *ct; |
9fb9cbb1 | 1053 | |
5251e2d2 | 1054 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 1055 | atomic_inc(&net->ct.count); |
5251e2d2 | 1056 | |
76eb9460 | 1057 | if (nf_conntrack_max && |
49ac8713 | 1058 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
93bb0ceb | 1059 | if (!early_drop(net, hash)) { |
49ac8713 | 1060 | atomic_dec(&net->ct.count); |
e87cc472 | 1061 | net_warn_ratelimited("nf_conntrack: table full, dropping packet\n"); |
9fb9cbb1 YK |
1062 | return ERR_PTR(-ENOMEM); |
1063 | } | |
1064 | } | |
1065 | ||
941297f4 ED |
1066 | /* |
1067 | * Do not use kmem_cache_zalloc(), as this cache uses | |
1068 | * SLAB_DESTROY_BY_RCU. | |
1069 | */ | |
0c5366b3 | 1070 | ct = kmem_cache_alloc(nf_conntrack_cachep, gfp); |
5e8018fc DB |
1071 | if (ct == NULL) |
1072 | goto out; | |
1073 | ||
440f0d58 | 1074 | spin_lock_init(&ct->lock); |
c88130bc | 1075 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 1076 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 1077 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
1078 | /* save hash for reusing when confirming */ |
1079 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
c41884ce | 1080 | ct->status = 0; |
c2d9ba9b | 1081 | write_pnet(&ct->ct_net, net); |
c41884ce FW |
1082 | memset(&ct->__nfct_init_offset[0], 0, |
1083 | offsetof(struct nf_conn, proto) - | |
1084 | offsetof(struct nf_conn, __nfct_init_offset[0])); | |
5e8018fc | 1085 | |
6c8dee98 | 1086 | nf_ct_zone_add(ct, zone); |
5e8018fc | 1087 | |
e53376be PNA |
1088 | /* Because we use RCU lookups, we set ct_general.use to zero before |
1089 | * this is inserted in any list. | |
941297f4 | 1090 | */ |
e53376be | 1091 | atomic_set(&ct->ct_general.use, 0); |
c88130bc | 1092 | return ct; |
5e8018fc DB |
1093 | out: |
1094 | atomic_dec(&net->ct.count); | |
5d0aa2cc | 1095 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 1096 | } |
99f07e91 | 1097 | |
308ac914 DB |
1098 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
1099 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1100 | const struct nf_conntrack_tuple *orig, |
1101 | const struct nf_conntrack_tuple *repl, | |
1102 | gfp_t gfp) | |
1103 | { | |
1104 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
1105 | } | |
13b18339 | 1106 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 1107 | |
c88130bc | 1108 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 1109 | { |
1d45209d ED |
1110 | struct net *net = nf_ct_net(ct); |
1111 | ||
e53376be PNA |
1112 | /* A freed object has refcnt == 0, that's |
1113 | * the golden rule for SLAB_DESTROY_BY_RCU | |
1114 | */ | |
1115 | NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0); | |
1116 | ||
ceeff754 | 1117 | nf_ct_ext_destroy(ct); |
ea781f19 | 1118 | nf_ct_ext_free(ct); |
0c5366b3 | 1119 | kmem_cache_free(nf_conntrack_cachep, ct); |
4e857c58 | 1120 | smp_mb__before_atomic(); |
0c3c6c00 | 1121 | atomic_dec(&net->ct.count); |
76507f69 | 1122 | } |
13b18339 | 1123 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 | 1124 | |
c539f017 | 1125 | |
9fb9cbb1 YK |
1126 | /* Allocate a new conntrack: we return -ENOMEM if classification |
1127 | failed due to stress. Otherwise it really is unclassifiable. */ | |
fc09e4a7 | 1128 | static noinline struct nf_conntrack_tuple_hash * |
b2a15a60 | 1129 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 1130 | const struct nf_conntrack_tuple *tuple, |
9fb9cbb1 | 1131 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 1132 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 1133 | struct sk_buff *skb, |
60b5f8f7 | 1134 | unsigned int dataoff, u32 hash) |
9fb9cbb1 | 1135 | { |
c88130bc | 1136 | struct nf_conn *ct; |
3c158f7f | 1137 | struct nf_conn_help *help; |
9fb9cbb1 | 1138 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 1139 | struct nf_conntrack_ecache *ecache; |
ca7433df | 1140 | struct nf_conntrack_expect *exp = NULL; |
308ac914 | 1141 | const struct nf_conntrack_zone *zone; |
60b5f8f7 | 1142 | struct nf_conn_timeout *timeout_ext; |
5e8018fc | 1143 | struct nf_conntrack_zone tmp; |
60b5f8f7 | 1144 | unsigned int *timeouts; |
9fb9cbb1 | 1145 | |
605dcad6 | 1146 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 1147 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
1148 | return NULL; |
1149 | } | |
1150 | ||
5e8018fc | 1151 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
99f07e91 CG |
1152 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
1153 | hash); | |
0a9ee813 | 1154 | if (IS_ERR(ct)) |
c88130bc | 1155 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 | 1156 | |
4440a2ab GF |
1157 | if (!nf_ct_add_synproxy(ct, tmpl)) { |
1158 | nf_conntrack_free(ct); | |
1159 | return ERR_PTR(-ENOMEM); | |
48b1de4c PM |
1160 | } |
1161 | ||
60b5f8f7 | 1162 | timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; |
ae2d708e PNA |
1163 | if (timeout_ext) { |
1164 | timeouts = nf_ct_timeout_data(timeout_ext); | |
1165 | if (unlikely(!timeouts)) | |
1166 | timeouts = l4proto->get_timeouts(net); | |
1167 | } else { | |
60b5f8f7 | 1168 | timeouts = l4proto->get_timeouts(net); |
ae2d708e | 1169 | } |
60b5f8f7 | 1170 | |
2c8503f5 | 1171 | if (!l4proto->new(ct, skb, dataoff, timeouts)) { |
c88130bc | 1172 | nf_conntrack_free(ct); |
ccd63c20 | 1173 | pr_debug("can't track with proto module\n"); |
9fb9cbb1 YK |
1174 | return NULL; |
1175 | } | |
1176 | ||
60b5f8f7 | 1177 | if (timeout_ext) |
ae2d708e PNA |
1178 | nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout), |
1179 | GFP_ATOMIC); | |
60b5f8f7 | 1180 | |
58401572 | 1181 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a992ca2a | 1182 | nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); |
c539f017 | 1183 | nf_ct_labels_ext_add(ct); |
b2a15a60 PM |
1184 | |
1185 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
1186 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
1187 | ecache ? ecache->expmask : 0, | |
1188 | GFP_ATOMIC); | |
58401572 | 1189 | |
ca7433df JDB |
1190 | local_bh_disable(); |
1191 | if (net->ct.expect_count) { | |
1192 | spin_lock(&nf_conntrack_expect_lock); | |
1193 | exp = nf_ct_find_expectation(net, zone, tuple); | |
1194 | if (exp) { | |
ccd63c20 | 1195 | pr_debug("expectation arrives ct=%p exp=%p\n", |
ca7433df JDB |
1196 | ct, exp); |
1197 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
1198 | __set_bit(IPS_EXPECTED_BIT, &ct->status); | |
1199 | /* exp->master safe, refcnt bumped in nf_ct_find_expectation */ | |
1200 | ct->master = exp->master; | |
1201 | if (exp->helper) { | |
1202 | help = nf_ct_helper_ext_add(ct, exp->helper, | |
1203 | GFP_ATOMIC); | |
1204 | if (help) | |
1205 | rcu_assign_pointer(help->helper, exp->helper); | |
1206 | } | |
ceceae1b | 1207 | |
9fb9cbb1 | 1208 | #ifdef CONFIG_NF_CONNTRACK_MARK |
ca7433df | 1209 | ct->mark = exp->master->mark; |
7c9728c3 JM |
1210 | #endif |
1211 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
ca7433df | 1212 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 1213 | #endif |
ca7433df JDB |
1214 | NF_CT_STAT_INC(net, expect_new); |
1215 | } | |
1216 | spin_unlock(&nf_conntrack_expect_lock); | |
1217 | } | |
8e8118f8 | 1218 | if (!exp) |
b2a15a60 | 1219 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
9fb9cbb1 | 1220 | |
e53376be PNA |
1221 | /* Now it is inserted into the unconfirmed list, bump refcount */ |
1222 | nf_conntrack_get(&ct->ct_general); | |
b7779d06 | 1223 | nf_ct_add_to_unconfirmed_list(ct); |
9fb9cbb1 | 1224 | |
ca7433df | 1225 | local_bh_enable(); |
9fb9cbb1 YK |
1226 | |
1227 | if (exp) { | |
1228 | if (exp->expectfn) | |
c88130bc | 1229 | exp->expectfn(ct, exp); |
6823645d | 1230 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
1231 | } |
1232 | ||
c88130bc | 1233 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
1234 | } |
1235 | ||
fc09e4a7 FW |
1236 | /* On success, returns 0, sets skb->_nfct | ctinfo */ |
1237 | static int | |
b2a15a60 | 1238 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 1239 | struct sk_buff *skb, |
9fb9cbb1 YK |
1240 | unsigned int dataoff, |
1241 | u_int16_t l3num, | |
1242 | u_int8_t protonum, | |
1243 | struct nf_conntrack_l3proto *l3proto, | |
fc09e4a7 | 1244 | struct nf_conntrack_l4proto *l4proto) |
9fb9cbb1 | 1245 | { |
308ac914 | 1246 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 YK |
1247 | struct nf_conntrack_tuple tuple; |
1248 | struct nf_conntrack_tuple_hash *h; | |
fc09e4a7 | 1249 | enum ip_conntrack_info ctinfo; |
5e8018fc | 1250 | struct nf_conntrack_zone tmp; |
9fb9cbb1 | 1251 | struct nf_conn *ct; |
99f07e91 | 1252 | u32 hash; |
9fb9cbb1 | 1253 | |
bbe735e4 | 1254 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
a31f1adc | 1255 | dataoff, l3num, protonum, net, &tuple, l3proto, |
605dcad6 | 1256 | l4proto)) { |
ccd63c20 | 1257 | pr_debug("Can't get tuple\n"); |
fc09e4a7 | 1258 | return 0; |
9fb9cbb1 YK |
1259 | } |
1260 | ||
1261 | /* look for tuple match */ | |
5e8018fc | 1262 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
1b8c8a9f | 1263 | hash = hash_conntrack_raw(&tuple, net); |
99f07e91 | 1264 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); |
9fb9cbb1 | 1265 | if (!h) { |
b2a15a60 | 1266 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
60b5f8f7 | 1267 | skb, dataoff, hash); |
9fb9cbb1 | 1268 | if (!h) |
fc09e4a7 | 1269 | return 0; |
9fb9cbb1 | 1270 | if (IS_ERR(h)) |
fc09e4a7 | 1271 | return PTR_ERR(h); |
9fb9cbb1 YK |
1272 | } |
1273 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1274 | ||
1275 | /* It exists; we have (non-exclusive) reference. */ | |
1276 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
fc09e4a7 | 1277 | ctinfo = IP_CT_ESTABLISHED_REPLY; |
9fb9cbb1 YK |
1278 | } else { |
1279 | /* Once we've had two way comms, always ESTABLISHED. */ | |
1280 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
ccd63c20 | 1281 | pr_debug("normal packet for %p\n", ct); |
fc09e4a7 | 1282 | ctinfo = IP_CT_ESTABLISHED; |
9fb9cbb1 | 1283 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { |
ccd63c20 | 1284 | pr_debug("related packet for %p\n", ct); |
fc09e4a7 | 1285 | ctinfo = IP_CT_RELATED; |
9fb9cbb1 | 1286 | } else { |
ccd63c20 | 1287 | pr_debug("new packet for %p\n", ct); |
fc09e4a7 | 1288 | ctinfo = IP_CT_NEW; |
9fb9cbb1 | 1289 | } |
9fb9cbb1 | 1290 | } |
fc09e4a7 FW |
1291 | nf_ct_set(skb, ct, ctinfo); |
1292 | return 0; | |
9fb9cbb1 YK |
1293 | } |
1294 | ||
1295 | unsigned int | |
a702a65f AD |
1296 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
1297 | struct sk_buff *skb) | |
9fb9cbb1 | 1298 | { |
97a6ad13 | 1299 | struct nf_conn *ct, *tmpl; |
9fb9cbb1 YK |
1300 | enum ip_conntrack_info ctinfo; |
1301 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 1302 | struct nf_conntrack_l4proto *l4proto; |
2c8503f5 | 1303 | unsigned int *timeouts; |
9fb9cbb1 YK |
1304 | unsigned int dataoff; |
1305 | u_int8_t protonum; | |
9fb9cbb1 YK |
1306 | int ret; |
1307 | ||
97a6ad13 | 1308 | tmpl = nf_ct_get(skb, &ctinfo); |
cc41c84b | 1309 | if (tmpl || ctinfo == IP_CT_UNTRACKED) { |
b2a15a60 | 1310 | /* Previously seen (loopback or untracked)? Ignore. */ |
cc41c84b FW |
1311 | if ((tmpl && !nf_ct_is_template(tmpl)) || |
1312 | ctinfo == IP_CT_UNTRACKED) { | |
b2a15a60 PM |
1313 | NF_CT_STAT_INC_ATOMIC(net, ignore); |
1314 | return NF_ACCEPT; | |
1315 | } | |
a9e419dc | 1316 | skb->_nfct = 0; |
9fb9cbb1 YK |
1317 | } |
1318 | ||
e2361cb9 | 1319 | /* rcu_read_lock()ed by nf_hook_thresh */ |
76108cea | 1320 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 1321 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
1322 | &dataoff, &protonum); |
1323 | if (ret <= 0) { | |
25985edc | 1324 | pr_debug("not prepared to track yet or error occurred\n"); |
0d55af87 AD |
1325 | NF_CT_STAT_INC_ATOMIC(net, error); |
1326 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1327 | ret = -ret; |
1328 | goto out; | |
9fb9cbb1 YK |
1329 | } |
1330 | ||
76108cea | 1331 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
1332 | |
1333 | /* It may be an special packet, error, unclean... | |
1334 | * inverse of the return code tells to the netfilter | |
1335 | * core what to do with the packet. */ | |
74c51a14 | 1336 | if (l4proto->error != NULL) { |
11df4b76 | 1337 | ret = l4proto->error(net, tmpl, skb, dataoff, pf, hooknum); |
74c51a14 | 1338 | if (ret <= 0) { |
0d55af87 AD |
1339 | NF_CT_STAT_INC_ATOMIC(net, error); |
1340 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1341 | ret = -ret; |
1342 | goto out; | |
74c51a14 | 1343 | } |
88ed01d1 | 1344 | /* ICMP[v6] protocol trackers may assign one conntrack. */ |
a9e419dc | 1345 | if (skb->_nfct) |
88ed01d1 | 1346 | goto out; |
9fb9cbb1 | 1347 | } |
08733a0c | 1348 | repeat: |
fc09e4a7 FW |
1349 | ret = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
1350 | l3proto, l4proto); | |
1351 | if (ret < 0) { | |
9fb9cbb1 | 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 | ||
fc09e4a7 FW |
1358 | ct = nf_ct_get(skb, &ctinfo); |
1359 | if (!ct) { | |
1360 | /* Not valid part of a connection */ | |
1361 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
1362 | ret = NF_ACCEPT; | |
1363 | goto out; | |
1364 | } | |
9fb9cbb1 | 1365 | |
60b5f8f7 | 1366 | /* Decide what timeout policy we want to apply to this flow. */ |
84b5ee93 | 1367 | timeouts = nf_ct_timeout_lookup(net, ct, l4proto); |
60b5f8f7 | 1368 | |
2c8503f5 | 1369 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); |
ec8d5409 | 1370 | if (ret <= 0) { |
9fb9cbb1 YK |
1371 | /* Invalid: inverse of the return code tells |
1372 | * the netfilter core what to do */ | |
0d53778e | 1373 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
97a6ad13 | 1374 | nf_conntrack_put(&ct->ct_general); |
a9e419dc | 1375 | skb->_nfct = 0; |
0d55af87 | 1376 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
1377 | if (ret == -NF_DROP) |
1378 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
56a62e22 AB |
1379 | /* Special case: TCP tracker reports an attempt to reopen a |
1380 | * closed/aborted connection. We have to go back and create a | |
1381 | * fresh conntrack. | |
1382 | */ | |
1383 | if (ret == -NF_REPEAT) | |
1384 | goto repeat; | |
b2a15a60 PM |
1385 | ret = -ret; |
1386 | goto out; | |
9fb9cbb1 YK |
1387 | } |
1388 | ||
fc09e4a7 FW |
1389 | if (ctinfo == IP_CT_ESTABLISHED_REPLY && |
1390 | !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 1391 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 | 1392 | out: |
56a62e22 AB |
1393 | if (tmpl) |
1394 | nf_ct_put(tmpl); | |
9fb9cbb1 YK |
1395 | |
1396 | return ret; | |
1397 | } | |
13b18339 | 1398 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 1399 | |
5f2b4c90 JE |
1400 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
1401 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 1402 | { |
5f2b4c90 | 1403 | bool ret; |
923f4902 PM |
1404 | |
1405 | rcu_read_lock(); | |
1406 | ret = nf_ct_invert_tuple(inverse, orig, | |
1407 | __nf_ct_l3proto_find(orig->src.l3num), | |
1408 | __nf_ct_l4proto_find(orig->src.l3num, | |
1409 | orig->dst.protonum)); | |
1410 | rcu_read_unlock(); | |
1411 | return ret; | |
9fb9cbb1 | 1412 | } |
13b18339 | 1413 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 1414 | |
5b1158e9 JK |
1415 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
1416 | implicitly racy: see __nf_conntrack_confirm */ | |
1417 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
1418 | const struct nf_conntrack_tuple *newreply) | |
1419 | { | |
1420 | struct nf_conn_help *help = nfct_help(ct); | |
1421 | ||
5b1158e9 JK |
1422 | /* Should be unconfirmed, so not in hash table yet */ |
1423 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
1424 | ||
0d53778e | 1425 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 1426 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
1427 | |
1428 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 1429 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 1430 | return; |
ceceae1b | 1431 | |
c52fbb41 | 1432 | rcu_read_lock(); |
b2a15a60 | 1433 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 1434 | rcu_read_unlock(); |
5b1158e9 | 1435 | } |
13b18339 | 1436 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 1437 | |
9fb9cbb1 YK |
1438 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
1439 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
1440 | enum ip_conntrack_info ctinfo, | |
1441 | const struct sk_buff *skb, | |
1442 | unsigned long extra_jiffies, | |
1443 | int do_acct) | |
1444 | { | |
9fb9cbb1 YK |
1445 | NF_CT_ASSERT(skb); |
1446 | ||
997ae831 | 1447 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1448 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1449 | goto acct; | |
997ae831 | 1450 | |
9fb9cbb1 | 1451 | /* If not in hash table, timer will not be active yet */ |
f330a7fd FW |
1452 | if (nf_ct_is_confirmed(ct)) |
1453 | extra_jiffies += nfct_time_stamp; | |
9fb9cbb1 | 1454 | |
f330a7fd | 1455 | ct->timeout = extra_jiffies; |
47d95045 | 1456 | acct: |
ba76738c PNA |
1457 | if (do_acct) |
1458 | nf_ct_acct_update(ct, ctinfo, skb->len); | |
9fb9cbb1 | 1459 | } |
13b18339 | 1460 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1461 | |
ad66713f FW |
1462 | bool nf_ct_kill_acct(struct nf_conn *ct, |
1463 | enum ip_conntrack_info ctinfo, | |
1464 | const struct sk_buff *skb) | |
51091764 | 1465 | { |
ad66713f | 1466 | nf_ct_acct_update(ct, ctinfo, skb->len); |
58401572 | 1467 | |
f330a7fd | 1468 | return nf_ct_delete(ct, 0, 0); |
51091764 | 1469 | } |
ad66713f | 1470 | EXPORT_SYMBOL_GPL(nf_ct_kill_acct); |
51091764 | 1471 | |
c0cd1156 | 1472 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
c1d10adb PNA |
1473 | |
1474 | #include <linux/netfilter/nfnetlink.h> | |
1475 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1476 | #include <linux/mutex.h> |
1477 | ||
c1d10adb PNA |
1478 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1479 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1480 | * or depend on ctnetlink */ | |
fdf70832 | 1481 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1482 | const struct nf_conntrack_tuple *tuple) |
1483 | { | |
bae65be8 DM |
1484 | if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) || |
1485 | nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port)) | |
1486 | goto nla_put_failure; | |
c1d10adb PNA |
1487 | return 0; |
1488 | ||
df6fb868 | 1489 | nla_put_failure: |
c1d10adb PNA |
1490 | return -1; |
1491 | } | |
fdf70832 | 1492 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1493 | |
f73e924c PM |
1494 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1495 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1496 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1497 | }; |
f73e924c | 1498 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1499 | |
fdf70832 | 1500 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1501 | struct nf_conntrack_tuple *t) |
1502 | { | |
df6fb868 | 1503 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1504 | return -EINVAL; |
1505 | ||
77236b6e PM |
1506 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1507 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1508 | |
1509 | return 0; | |
1510 | } | |
fdf70832 | 1511 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1512 | |
1513 | int nf_ct_port_nlattr_tuple_size(void) | |
1514 | { | |
1515 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1516 | } | |
1517 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1518 | #endif |
1519 | ||
9fb9cbb1 | 1520 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
312a0c16 | 1521 | static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb) |
9fb9cbb1 YK |
1522 | { |
1523 | struct nf_conn *ct; | |
1524 | enum ip_conntrack_info ctinfo; | |
1525 | ||
1526 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1527 | ct = nf_ct_get(skb, &ctinfo); | |
1528 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
fb048833 | 1529 | ctinfo = IP_CT_RELATED_REPLY; |
9fb9cbb1 YK |
1530 | else |
1531 | ctinfo = IP_CT_RELATED; | |
1532 | ||
1533 | /* Attach to new skbuff, and increment count */ | |
c74454fa | 1534 | nf_ct_set(nskb, ct, ctinfo); |
cb9c6836 | 1535 | nf_conntrack_get(skb_nfct(nskb)); |
9fb9cbb1 YK |
1536 | } |
1537 | ||
9fb9cbb1 | 1538 | /* Bring out ya dead! */ |
df0933dc | 1539 | static struct nf_conn * |
400dad39 | 1540 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1541 | void *data, unsigned int *bucket) |
1542 | { | |
df0933dc PM |
1543 | struct nf_conntrack_tuple_hash *h; |
1544 | struct nf_conn *ct; | |
ea781f19 | 1545 | struct hlist_nulls_node *n; |
b7779d06 | 1546 | int cpu; |
93bb0ceb | 1547 | spinlock_t *lockp; |
9fb9cbb1 | 1548 | |
56d52d48 | 1549 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
93bb0ceb JDB |
1550 | lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS]; |
1551 | local_bh_disable(); | |
b16c2919 | 1552 | nf_conntrack_lock(lockp); |
56d52d48 FW |
1553 | if (*bucket < nf_conntrack_htable_size) { |
1554 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) { | |
93bb0ceb JDB |
1555 | if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL) |
1556 | continue; | |
1557 | ct = nf_ct_tuplehash_to_ctrack(h); | |
e0c7d472 FW |
1558 | if (net_eq(nf_ct_net(ct), net) && |
1559 | iter(ct, data)) | |
93bb0ceb JDB |
1560 | goto found; |
1561 | } | |
df0933dc | 1562 | } |
93bb0ceb JDB |
1563 | spin_unlock(lockp); |
1564 | local_bh_enable(); | |
d93c6258 | 1565 | cond_resched(); |
601e68e1 | 1566 | } |
b7779d06 JDB |
1567 | |
1568 | for_each_possible_cpu(cpu) { | |
1569 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1570 | ||
1571 | spin_lock_bh(&pcpu->lock); | |
1572 | hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) { | |
1573 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1574 | if (iter(ct, data)) | |
1575 | set_bit(IPS_DYING_BIT, &ct->status); | |
1576 | } | |
1577 | spin_unlock_bh(&pcpu->lock); | |
d93c6258 | 1578 | cond_resched(); |
b7779d06 | 1579 | } |
df0933dc PM |
1580 | return NULL; |
1581 | found: | |
c073e3fa | 1582 | atomic_inc(&ct->ct_general.use); |
93bb0ceb JDB |
1583 | spin_unlock(lockp); |
1584 | local_bh_enable(); | |
df0933dc | 1585 | return ct; |
9fb9cbb1 YK |
1586 | } |
1587 | ||
400dad39 AD |
1588 | void nf_ct_iterate_cleanup(struct net *net, |
1589 | int (*iter)(struct nf_conn *i, void *data), | |
c655bc68 | 1590 | void *data, u32 portid, int report) |
9fb9cbb1 | 1591 | { |
df0933dc | 1592 | struct nf_conn *ct; |
9fb9cbb1 YK |
1593 | unsigned int bucket = 0; |
1594 | ||
d93c6258 FW |
1595 | might_sleep(); |
1596 | ||
88b68bc5 FW |
1597 | if (atomic_read(&net->ct.count) == 0) |
1598 | return; | |
1599 | ||
400dad39 | 1600 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 | 1601 | /* Time to push up daises... */ |
9fb9cbb1 | 1602 | |
f330a7fd | 1603 | nf_ct_delete(ct, portid, report); |
9fb9cbb1 | 1604 | nf_ct_put(ct); |
d93c6258 | 1605 | cond_resched(); |
9fb9cbb1 YK |
1606 | } |
1607 | } | |
13b18339 | 1608 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1609 | |
274d383b PNA |
1610 | static int kill_all(struct nf_conn *i, void *data) |
1611 | { | |
1612 | return 1; | |
1613 | } | |
1614 | ||
d862a662 | 1615 | void nf_ct_free_hashtable(void *hash, unsigned int size) |
9fb9cbb1 | 1616 | { |
d862a662 | 1617 | if (is_vmalloc_addr(hash)) |
9fb9cbb1 YK |
1618 | vfree(hash); |
1619 | else | |
601e68e1 | 1620 | free_pages((unsigned long)hash, |
f205c5e0 | 1621 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1622 | } |
ac565e5f | 1623 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1624 | |
f94161c1 | 1625 | void nf_conntrack_cleanup_start(void) |
9fb9cbb1 | 1626 | { |
b87a2f91 | 1627 | conntrack_gc_work.exiting = true; |
f94161c1 G |
1628 | RCU_INIT_POINTER(ip_ct_attach, NULL); |
1629 | } | |
1630 | ||
1631 | void nf_conntrack_cleanup_end(void) | |
1632 | { | |
1633 | RCU_INIT_POINTER(nf_ct_destroy, NULL); | |
9edd7ca0 | 1634 | |
b87a2f91 | 1635 | cancel_delayed_work_sync(&conntrack_gc_work.dwork); |
56d52d48 FW |
1636 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
1637 | ||
04d87001 | 1638 | nf_conntrack_proto_fini(); |
41d73ec0 | 1639 | nf_conntrack_seqadj_fini(); |
5f69b8f5 | 1640 | nf_conntrack_labels_fini(); |
5e615b22 | 1641 | nf_conntrack_helper_fini(); |
8684094c | 1642 | nf_conntrack_timeout_fini(); |
3fe0f943 | 1643 | nf_conntrack_ecache_fini(); |
73f4001a | 1644 | nf_conntrack_tstamp_fini(); |
b7ff3a1f | 1645 | nf_conntrack_acct_fini(); |
83b4dbe1 | 1646 | nf_conntrack_expect_fini(); |
77571149 FW |
1647 | |
1648 | kmem_cache_destroy(nf_conntrack_cachep); | |
08f6547d | 1649 | } |
9fb9cbb1 | 1650 | |
f94161c1 G |
1651 | /* |
1652 | * Mishearing the voices in his head, our hero wonders how he's | |
1653 | * supposed to kill the mall. | |
1654 | */ | |
1655 | void nf_conntrack_cleanup_net(struct net *net) | |
08f6547d | 1656 | { |
dece40e8 VD |
1657 | LIST_HEAD(single); |
1658 | ||
1659 | list_add(&net->exit_list, &single); | |
1660 | nf_conntrack_cleanup_net_list(&single); | |
1661 | } | |
1662 | ||
1663 | void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list) | |
1664 | { | |
1665 | int busy; | |
1666 | struct net *net; | |
1667 | ||
f94161c1 G |
1668 | /* |
1669 | * This makes sure all current packets have passed through | |
1670 | * netfilter framework. Roll on, two-stage module | |
1671 | * delete... | |
1672 | */ | |
1673 | synchronize_net(); | |
dece40e8 VD |
1674 | i_see_dead_people: |
1675 | busy = 0; | |
1676 | list_for_each_entry(net, net_exit_list, exit_list) { | |
c655bc68 | 1677 | nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0); |
dece40e8 VD |
1678 | if (atomic_read(&net->ct.count) != 0) |
1679 | busy = 1; | |
1680 | } | |
1681 | if (busy) { | |
9fb9cbb1 YK |
1682 | schedule(); |
1683 | goto i_see_dead_people; | |
1684 | } | |
1685 | ||
dece40e8 | 1686 | list_for_each_entry(net, net_exit_list, exit_list) { |
dece40e8 VD |
1687 | nf_conntrack_proto_pernet_fini(net); |
1688 | nf_conntrack_helper_pernet_fini(net); | |
1689 | nf_conntrack_ecache_pernet_fini(net); | |
1690 | nf_conntrack_tstamp_pernet_fini(net); | |
1691 | nf_conntrack_acct_pernet_fini(net); | |
1692 | nf_conntrack_expect_pernet_fini(net); | |
dece40e8 | 1693 | free_percpu(net->ct.stat); |
b7779d06 | 1694 | free_percpu(net->ct.pcpu_lists); |
dece40e8 | 1695 | } |
08f6547d AD |
1696 | } |
1697 | ||
d862a662 | 1698 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) |
9fb9cbb1 | 1699 | { |
ea781f19 ED |
1700 | struct hlist_nulls_head *hash; |
1701 | unsigned int nr_slots, i; | |
1702 | size_t sz; | |
9fb9cbb1 | 1703 | |
9cc1c73a FW |
1704 | if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head))) |
1705 | return NULL; | |
1706 | ||
ea781f19 ED |
1707 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1708 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
9cc1c73a FW |
1709 | |
1710 | if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head))) | |
1711 | return NULL; | |
1712 | ||
ea781f19 ED |
1713 | sz = nr_slots * sizeof(struct hlist_nulls_head); |
1714 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1715 | get_order(sz)); | |
f0ad4621 | 1716 | if (!hash) |
966567b7 | 1717 | hash = vzalloc(sz); |
9fb9cbb1 | 1718 | |
ea781f19 ED |
1719 | if (hash && nulls) |
1720 | for (i = 0; i < nr_slots; i++) | |
1721 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1722 | |
1723 | return hash; | |
1724 | } | |
ac565e5f | 1725 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1726 | |
3183ab89 | 1727 | int nf_conntrack_hash_resize(unsigned int hashsize) |
9fb9cbb1 | 1728 | { |
3183ab89 FW |
1729 | int i, bucket; |
1730 | unsigned int old_size; | |
ea781f19 | 1731 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1732 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1733 | struct nf_conn *ct; |
9fb9cbb1 | 1734 | |
9fb9cbb1 YK |
1735 | if (!hashsize) |
1736 | return -EINVAL; | |
1737 | ||
d862a662 | 1738 | hash = nf_ct_alloc_hashtable(&hashsize, 1); |
9fb9cbb1 YK |
1739 | if (!hash) |
1740 | return -ENOMEM; | |
1741 | ||
3183ab89 FW |
1742 | old_size = nf_conntrack_htable_size; |
1743 | if (old_size == hashsize) { | |
1744 | nf_ct_free_hashtable(hash, hashsize); | |
1745 | return 0; | |
1746 | } | |
1747 | ||
93bb0ceb JDB |
1748 | local_bh_disable(); |
1749 | nf_conntrack_all_lock(); | |
a3efd812 | 1750 | write_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb | 1751 | |
76507f69 PM |
1752 | /* Lookups in the old hash might happen in parallel, which means we |
1753 | * might get false negatives during connection lookup. New connections | |
1754 | * created because of a false negative won't make it into the hash | |
93bb0ceb | 1755 | * though since that required taking the locks. |
76507f69 | 1756 | */ |
93bb0ceb | 1757 | |
56d52d48 FW |
1758 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
1759 | while (!hlist_nulls_empty(&nf_conntrack_hash[i])) { | |
1760 | h = hlist_nulls_entry(nf_conntrack_hash[i].first, | |
1761 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1762 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1763 | hlist_nulls_del_rcu(&h->hnnode); |
1b8c8a9f FW |
1764 | bucket = __hash_conntrack(nf_ct_net(ct), |
1765 | &h->tuple, hashsize); | |
ea781f19 | 1766 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1767 | } |
1768 | } | |
56d52d48 FW |
1769 | old_size = nf_conntrack_htable_size; |
1770 | old_hash = nf_conntrack_hash; | |
9fb9cbb1 | 1771 | |
56d52d48 FW |
1772 | nf_conntrack_hash = hash; |
1773 | nf_conntrack_htable_size = hashsize; | |
93bb0ceb | 1774 | |
a3efd812 | 1775 | write_seqcount_end(&nf_conntrack_generation); |
93bb0ceb JDB |
1776 | nf_conntrack_all_unlock(); |
1777 | local_bh_enable(); | |
9fb9cbb1 | 1778 | |
5e3c61f9 | 1779 | synchronize_net(); |
d862a662 | 1780 | nf_ct_free_hashtable(old_hash, old_size); |
9fb9cbb1 YK |
1781 | return 0; |
1782 | } | |
3183ab89 FW |
1783 | |
1784 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) | |
1785 | { | |
1786 | unsigned int hashsize; | |
1787 | int rc; | |
1788 | ||
1789 | if (current->nsproxy->net_ns != &init_net) | |
1790 | return -EOPNOTSUPP; | |
1791 | ||
1792 | /* On boot, we can set this without any fancy locking. */ | |
1793 | if (!nf_conntrack_htable_size) | |
1794 | return param_set_uint(val, kp); | |
1795 | ||
1796 | rc = kstrtouint(val, 0, &hashsize); | |
1797 | if (rc) | |
1798 | return rc; | |
1799 | ||
1800 | return nf_conntrack_hash_resize(hashsize); | |
1801 | } | |
fae718dd | 1802 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1803 | |
fae718dd | 1804 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1805 | &nf_conntrack_htable_size, 0600); |
1806 | ||
f94161c1 | 1807 | int nf_conntrack_init_start(void) |
9fb9cbb1 | 1808 | { |
f205c5e0 | 1809 | int max_factor = 8; |
0c5366b3 | 1810 | int ret = -ENOMEM; |
cc41c84b | 1811 | int i; |
93bb0ceb | 1812 | |
a3efd812 FW |
1813 | seqcount_init(&nf_conntrack_generation); |
1814 | ||
d5d20912 | 1815 | for (i = 0; i < CONNTRACK_LOCKS; i++) |
93bb0ceb | 1816 | spin_lock_init(&nf_conntrack_locks[i]); |
9fb9cbb1 | 1817 | |
9fb9cbb1 | 1818 | if (!nf_conntrack_htable_size) { |
88eab472 ML |
1819 | /* Idea from tcp.c: use 1/16384 of memory. |
1820 | * On i386: 32MB machine has 512 buckets. | |
1821 | * >= 1GB machines have 16384 buckets. | |
1822 | * >= 4GB machines have 65536 buckets. | |
1823 | */ | |
9fb9cbb1 | 1824 | nf_conntrack_htable_size |
4481374c | 1825 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1826 | / sizeof(struct hlist_head)); |
88eab472 ML |
1827 | if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE))) |
1828 | nf_conntrack_htable_size = 65536; | |
1829 | else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
f205c5e0 PM |
1830 | nf_conntrack_htable_size = 16384; |
1831 | if (nf_conntrack_htable_size < 32) | |
1832 | nf_conntrack_htable_size = 32; | |
1833 | ||
1834 | /* Use a max. factor of four by default to get the same max as | |
1835 | * with the old struct list_heads. When a table size is given | |
1836 | * we use the old value of 8 to avoid reducing the max. | |
1837 | * entries. */ | |
1838 | max_factor = 4; | |
9fb9cbb1 | 1839 | } |
56d52d48 FW |
1840 | |
1841 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1); | |
1842 | if (!nf_conntrack_hash) | |
1843 | return -ENOMEM; | |
1844 | ||
f205c5e0 | 1845 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 1846 | |
0c5366b3 | 1847 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
a9e419dc FW |
1848 | sizeof(struct nf_conn), |
1849 | NFCT_INFOMASK + 1, | |
5a75cdeb | 1850 | SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL); |
0c5366b3 FW |
1851 | if (!nf_conntrack_cachep) |
1852 | goto err_cachep; | |
1853 | ||
654d0fbd | 1854 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
1855 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
1856 | nf_conntrack_max); | |
83b4dbe1 G |
1857 | |
1858 | ret = nf_conntrack_expect_init(); | |
1859 | if (ret < 0) | |
1860 | goto err_expect; | |
1861 | ||
b7ff3a1f G |
1862 | ret = nf_conntrack_acct_init(); |
1863 | if (ret < 0) | |
1864 | goto err_acct; | |
1865 | ||
73f4001a G |
1866 | ret = nf_conntrack_tstamp_init(); |
1867 | if (ret < 0) | |
1868 | goto err_tstamp; | |
1869 | ||
3fe0f943 G |
1870 | ret = nf_conntrack_ecache_init(); |
1871 | if (ret < 0) | |
1872 | goto err_ecache; | |
1873 | ||
8684094c G |
1874 | ret = nf_conntrack_timeout_init(); |
1875 | if (ret < 0) | |
1876 | goto err_timeout; | |
1877 | ||
5e615b22 G |
1878 | ret = nf_conntrack_helper_init(); |
1879 | if (ret < 0) | |
1880 | goto err_helper; | |
1881 | ||
5f69b8f5 G |
1882 | ret = nf_conntrack_labels_init(); |
1883 | if (ret < 0) | |
1884 | goto err_labels; | |
1885 | ||
41d73ec0 PM |
1886 | ret = nf_conntrack_seqadj_init(); |
1887 | if (ret < 0) | |
1888 | goto err_seqadj; | |
1889 | ||
04d87001 G |
1890 | ret = nf_conntrack_proto_init(); |
1891 | if (ret < 0) | |
1892 | goto err_proto; | |
1893 | ||
b87a2f91 | 1894 | conntrack_gc_work_init(&conntrack_gc_work); |
e5072053 | 1895 | queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, HZ); |
b87a2f91 | 1896 | |
08f6547d AD |
1897 | return 0; |
1898 | ||
04d87001 | 1899 | err_proto: |
41d73ec0 PM |
1900 | nf_conntrack_seqadj_fini(); |
1901 | err_seqadj: | |
04d87001 | 1902 | nf_conntrack_labels_fini(); |
5f69b8f5 G |
1903 | err_labels: |
1904 | nf_conntrack_helper_fini(); | |
5e615b22 G |
1905 | err_helper: |
1906 | nf_conntrack_timeout_fini(); | |
8684094c G |
1907 | err_timeout: |
1908 | nf_conntrack_ecache_fini(); | |
3fe0f943 G |
1909 | err_ecache: |
1910 | nf_conntrack_tstamp_fini(); | |
73f4001a G |
1911 | err_tstamp: |
1912 | nf_conntrack_acct_fini(); | |
b7ff3a1f G |
1913 | err_acct: |
1914 | nf_conntrack_expect_fini(); | |
83b4dbe1 | 1915 | err_expect: |
0c5366b3 FW |
1916 | kmem_cache_destroy(nf_conntrack_cachep); |
1917 | err_cachep: | |
56d52d48 | 1918 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
08f6547d AD |
1919 | return ret; |
1920 | } | |
1921 | ||
f94161c1 G |
1922 | void nf_conntrack_init_end(void) |
1923 | { | |
1924 | /* For use by REJECT target */ | |
1925 | RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); | |
1926 | RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); | |
f94161c1 G |
1927 | } |
1928 | ||
8cc20198 ED |
1929 | /* |
1930 | * We need to use special "null" values, not used in hash table | |
1931 | */ | |
1932 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1933 | #define DYING_NULLS_VAL ((1<<30)+1) | |
252b3e8c | 1934 | #define TEMPLATE_NULLS_VAL ((1<<30)+2) |
8cc20198 | 1935 | |
f94161c1 | 1936 | int nf_conntrack_init_net(struct net *net) |
08f6547d | 1937 | { |
b7779d06 JDB |
1938 | int ret = -ENOMEM; |
1939 | int cpu; | |
ceceae1b | 1940 | |
cc41c84b | 1941 | BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER); |
08f6547d | 1942 | atomic_set(&net->ct.count, 0); |
b7779d06 JDB |
1943 | |
1944 | net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu); | |
1945 | if (!net->ct.pcpu_lists) | |
08f6547d | 1946 | goto err_stat; |
b7779d06 JDB |
1947 | |
1948 | for_each_possible_cpu(cpu) { | |
1949 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1950 | ||
1951 | spin_lock_init(&pcpu->lock); | |
1952 | INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL); | |
1953 | INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL); | |
08f6547d | 1954 | } |
5b3501fa | 1955 | |
b7779d06 JDB |
1956 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1957 | if (!net->ct.stat) | |
1958 | goto err_pcpu_lists; | |
1959 | ||
83b4dbe1 | 1960 | ret = nf_conntrack_expect_pernet_init(net); |
08f6547d AD |
1961 | if (ret < 0) |
1962 | goto err_expect; | |
b7ff3a1f | 1963 | ret = nf_conntrack_acct_pernet_init(net); |
58401572 | 1964 | if (ret < 0) |
08f6547d | 1965 | goto err_acct; |
73f4001a | 1966 | ret = nf_conntrack_tstamp_pernet_init(net); |
a992ca2a PNA |
1967 | if (ret < 0) |
1968 | goto err_tstamp; | |
3fe0f943 | 1969 | ret = nf_conntrack_ecache_pernet_init(net); |
a0891aa6 PNA |
1970 | if (ret < 0) |
1971 | goto err_ecache; | |
5e615b22 | 1972 | ret = nf_conntrack_helper_pernet_init(net); |
a9006892 EL |
1973 | if (ret < 0) |
1974 | goto err_helper; | |
04d87001 | 1975 | ret = nf_conntrack_proto_pernet_init(net); |
f94161c1 G |
1976 | if (ret < 0) |
1977 | goto err_proto; | |
08f6547d | 1978 | return 0; |
c539f017 | 1979 | |
f94161c1 | 1980 | err_proto: |
5e615b22 | 1981 | nf_conntrack_helper_pernet_fini(net); |
a9006892 | 1982 | err_helper: |
3fe0f943 | 1983 | nf_conntrack_ecache_pernet_fini(net); |
a0891aa6 | 1984 | err_ecache: |
73f4001a | 1985 | nf_conntrack_tstamp_pernet_fini(net); |
a992ca2a | 1986 | err_tstamp: |
b7ff3a1f | 1987 | nf_conntrack_acct_pernet_fini(net); |
08f6547d | 1988 | err_acct: |
83b4dbe1 | 1989 | nf_conntrack_expect_pernet_fini(net); |
08f6547d | 1990 | err_expect: |
0d55af87 | 1991 | free_percpu(net->ct.stat); |
b7779d06 JDB |
1992 | err_pcpu_lists: |
1993 | free_percpu(net->ct.pcpu_lists); | |
0d55af87 | 1994 | err_stat: |
08f6547d AD |
1995 | return ret; |
1996 | } |