]>
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; |
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 | ||
30322309 FW |
353 | #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK) |
354 | ||
0838aa7f | 355 | /* Released via destroy_conntrack() */ |
308ac914 DB |
356 | struct nf_conn *nf_ct_tmpl_alloc(struct net *net, |
357 | const struct nf_conntrack_zone *zone, | |
358 | gfp_t flags) | |
0838aa7f | 359 | { |
30322309 | 360 | struct nf_conn *tmpl, *p; |
0838aa7f | 361 | |
30322309 FW |
362 | if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) { |
363 | tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags); | |
364 | if (!tmpl) | |
365 | return NULL; | |
366 | ||
367 | p = tmpl; | |
368 | tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p); | |
369 | if (tmpl != p) { | |
370 | tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p); | |
371 | tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p; | |
372 | } | |
373 | } else { | |
374 | tmpl = kzalloc(sizeof(*tmpl), flags); | |
375 | if (!tmpl) | |
376 | return NULL; | |
377 | } | |
0838aa7f PNA |
378 | |
379 | tmpl->status = IPS_TEMPLATE; | |
380 | write_pnet(&tmpl->ct_net, net); | |
6c8dee98 | 381 | nf_ct_zone_add(tmpl, zone); |
0838aa7f PNA |
382 | atomic_set(&tmpl->ct_general.use, 0); |
383 | ||
384 | return tmpl; | |
0838aa7f PNA |
385 | } |
386 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc); | |
387 | ||
9cf94eab | 388 | void nf_ct_tmpl_free(struct nf_conn *tmpl) |
0838aa7f PNA |
389 | { |
390 | nf_ct_ext_destroy(tmpl); | |
391 | nf_ct_ext_free(tmpl); | |
30322309 FW |
392 | |
393 | if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) | |
394 | kfree((char *)tmpl - tmpl->proto.tmpl_padto); | |
395 | else | |
396 | kfree(tmpl); | |
0838aa7f | 397 | } |
9cf94eab | 398 | EXPORT_SYMBOL_GPL(nf_ct_tmpl_free); |
0838aa7f | 399 | |
9fb9cbb1 YK |
400 | static void |
401 | destroy_conntrack(struct nf_conntrack *nfct) | |
402 | { | |
403 | struct nf_conn *ct = (struct nf_conn *)nfct; | |
605dcad6 | 404 | struct nf_conntrack_l4proto *l4proto; |
9fb9cbb1 | 405 | |
0d53778e | 406 | pr_debug("destroy_conntrack(%p)\n", ct); |
9fb9cbb1 | 407 | NF_CT_ASSERT(atomic_read(&nfct->use) == 0); |
9fb9cbb1 | 408 | |
0838aa7f PNA |
409 | if (unlikely(nf_ct_is_template(ct))) { |
410 | nf_ct_tmpl_free(ct); | |
411 | return; | |
412 | } | |
923f4902 | 413 | rcu_read_lock(); |
5e8fbe2a | 414 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); |
4b4ceb9d | 415 | if (l4proto->destroy) |
605dcad6 | 416 | l4proto->destroy(ct); |
9fb9cbb1 | 417 | |
982d9a9c | 418 | rcu_read_unlock(); |
9fb9cbb1 | 419 | |
ca7433df | 420 | local_bh_disable(); |
9fb9cbb1 YK |
421 | /* Expectations will have been removed in clean_from_lists, |
422 | * except TFTP can create an expectation on the first packet, | |
423 | * before connection is in the list, so we need to clean here, | |
ca7433df JDB |
424 | * too. |
425 | */ | |
c1d10adb | 426 | nf_ct_remove_expectations(ct); |
9fb9cbb1 | 427 | |
b7779d06 | 428 | nf_ct_del_from_dying_or_unconfirmed_list(ct); |
9fb9cbb1 | 429 | |
ca7433df | 430 | local_bh_enable(); |
9fb9cbb1 YK |
431 | |
432 | if (ct->master) | |
433 | nf_ct_put(ct->master); | |
434 | ||
0d53778e | 435 | pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); |
9fb9cbb1 YK |
436 | nf_conntrack_free(ct); |
437 | } | |
438 | ||
02982c27 | 439 | static void nf_ct_delete_from_lists(struct nf_conn *ct) |
9fb9cbb1 | 440 | { |
0d55af87 | 441 | struct net *net = nf_ct_net(ct); |
93bb0ceb | 442 | unsigned int hash, reply_hash; |
93bb0ceb | 443 | unsigned int sequence; |
9fb9cbb1 | 444 | |
9858a3ae | 445 | nf_ct_helper_destroy(ct); |
93bb0ceb JDB |
446 | |
447 | local_bh_disable(); | |
448 | do { | |
a3efd812 | 449 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 450 | hash = hash_conntrack(net, |
93bb0ceb | 451 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 452 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
453 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
454 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
455 | ||
9fb9cbb1 | 456 | clean_from_lists(ct); |
93bb0ceb JDB |
457 | nf_conntrack_double_unlock(hash, reply_hash); |
458 | ||
b7779d06 | 459 | nf_ct_add_to_dying_list(ct); |
93bb0ceb | 460 | |
93bb0ceb | 461 | local_bh_enable(); |
dd7669a9 | 462 | } |
dd7669a9 | 463 | |
02982c27 | 464 | bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report) |
dd7669a9 | 465 | { |
a992ca2a PNA |
466 | struct nf_conn_tstamp *tstamp; |
467 | ||
f330a7fd FW |
468 | if (test_and_set_bit(IPS_DYING_BIT, &ct->status)) |
469 | return false; | |
470 | ||
a992ca2a PNA |
471 | tstamp = nf_conn_tstamp_find(ct); |
472 | if (tstamp && tstamp->stop == 0) | |
d2de875c | 473 | tstamp->stop = ktime_get_real_ns(); |
dd7669a9 | 474 | |
9500507c FW |
475 | if (nf_conntrack_event_report(IPCT_DESTROY, ct, |
476 | portid, report) < 0) { | |
f330a7fd FW |
477 | /* destroy event was not delivered. nf_ct_put will |
478 | * be done by event cache worker on redelivery. | |
479 | */ | |
dd7669a9 | 480 | nf_ct_delete_from_lists(ct); |
9500507c | 481 | nf_conntrack_ecache_delayed_work(nf_ct_net(ct)); |
02982c27 | 482 | return false; |
dd7669a9 | 483 | } |
9500507c FW |
484 | |
485 | nf_conntrack_ecache_work(nf_ct_net(ct)); | |
dd7669a9 | 486 | nf_ct_delete_from_lists(ct); |
9fb9cbb1 | 487 | nf_ct_put(ct); |
02982c27 FW |
488 | return true; |
489 | } | |
490 | EXPORT_SYMBOL_GPL(nf_ct_delete); | |
491 | ||
c6825c09 AV |
492 | static inline bool |
493 | nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, | |
308ac914 | 494 | const struct nf_conntrack_tuple *tuple, |
e0c7d472 FW |
495 | const struct nf_conntrack_zone *zone, |
496 | const struct net *net) | |
c6825c09 AV |
497 | { |
498 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
499 | ||
500 | /* A conntrack can be recreated with the equal tuple, | |
501 | * so we need to check that the conntrack is confirmed | |
502 | */ | |
503 | return nf_ct_tuple_equal(tuple, &h->tuple) && | |
deedb590 | 504 | nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) && |
e0c7d472 FW |
505 | nf_ct_is_confirmed(ct) && |
506 | net_eq(net, nf_ct_net(ct)); | |
c6825c09 AV |
507 | } |
508 | ||
f330a7fd FW |
509 | /* caller must hold rcu readlock and none of the nf_conntrack_locks */ |
510 | static void nf_ct_gc_expired(struct nf_conn *ct) | |
511 | { | |
512 | if (!atomic_inc_not_zero(&ct->ct_general.use)) | |
513 | return; | |
514 | ||
515 | if (nf_ct_should_gc(ct)) | |
516 | nf_ct_kill(ct); | |
517 | ||
518 | nf_ct_put(ct); | |
519 | } | |
520 | ||
ea781f19 ED |
521 | /* |
522 | * Warning : | |
523 | * - Caller must take a reference on returned object | |
524 | * and recheck nf_ct_tuple_equal(tuple, &h->tuple) | |
ea781f19 | 525 | */ |
99f07e91 | 526 | static struct nf_conntrack_tuple_hash * |
308ac914 | 527 | ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 528 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
529 | { |
530 | struct nf_conntrack_tuple_hash *h; | |
5e3c61f9 | 531 | struct hlist_nulls_head *ct_hash; |
ea781f19 | 532 | struct hlist_nulls_node *n; |
92e47ba8 | 533 | unsigned int bucket, hsize; |
9fb9cbb1 | 534 | |
ea781f19 | 535 | begin: |
92e47ba8 LZ |
536 | nf_conntrack_get_ht(&ct_hash, &hsize); |
537 | bucket = reciprocal_scale(hash, hsize); | |
5e3c61f9 FW |
538 | |
539 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) { | |
f330a7fd FW |
540 | struct nf_conn *ct; |
541 | ||
542 | ct = nf_ct_tuplehash_to_ctrack(h); | |
543 | if (nf_ct_is_expired(ct)) { | |
544 | nf_ct_gc_expired(ct); | |
545 | continue; | |
546 | } | |
547 | ||
548 | if (nf_ct_is_dying(ct)) | |
549 | continue; | |
550 | ||
8e8118f8 | 551 | if (nf_ct_key_equal(h, tuple, zone, net)) |
9fb9cbb1 | 552 | return h; |
9fb9cbb1 | 553 | } |
ea781f19 ED |
554 | /* |
555 | * if the nulls value we got at the end of this lookup is | |
556 | * not the expected one, we must restart lookup. | |
557 | * We probably met an item that was moved to another chain. | |
558 | */ | |
99f07e91 | 559 | if (get_nulls_value(n) != bucket) { |
2cf12348 | 560 | NF_CT_STAT_INC_ATOMIC(net, search_restart); |
ea781f19 | 561 | goto begin; |
af740b2c | 562 | } |
9fb9cbb1 YK |
563 | |
564 | return NULL; | |
565 | } | |
99f07e91 | 566 | |
9fb9cbb1 | 567 | /* Find a connection corresponding to a tuple. */ |
99f07e91 | 568 | static struct nf_conntrack_tuple_hash * |
308ac914 | 569 | __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 | 570 | const struct nf_conntrack_tuple *tuple, u32 hash) |
9fb9cbb1 YK |
571 | { |
572 | struct nf_conntrack_tuple_hash *h; | |
76507f69 | 573 | struct nf_conn *ct; |
9fb9cbb1 | 574 | |
76507f69 | 575 | rcu_read_lock(); |
ea781f19 | 576 | begin: |
99f07e91 | 577 | h = ____nf_conntrack_find(net, zone, tuple, hash); |
76507f69 PM |
578 | if (h) { |
579 | ct = nf_ct_tuplehash_to_ctrack(h); | |
8d8890b7 PM |
580 | if (unlikely(nf_ct_is_dying(ct) || |
581 | !atomic_inc_not_zero(&ct->ct_general.use))) | |
76507f69 | 582 | h = NULL; |
ea781f19 | 583 | else { |
e0c7d472 | 584 | if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) { |
ea781f19 ED |
585 | nf_ct_put(ct); |
586 | goto begin; | |
587 | } | |
588 | } | |
76507f69 PM |
589 | } |
590 | rcu_read_unlock(); | |
9fb9cbb1 YK |
591 | |
592 | return h; | |
593 | } | |
99f07e91 CG |
594 | |
595 | struct nf_conntrack_tuple_hash * | |
308ac914 | 596 | nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, |
99f07e91 CG |
597 | const struct nf_conntrack_tuple *tuple) |
598 | { | |
599 | return __nf_conntrack_find_get(net, zone, tuple, | |
1b8c8a9f | 600 | hash_conntrack_raw(tuple, net)); |
99f07e91 | 601 | } |
13b18339 | 602 | EXPORT_SYMBOL_GPL(nf_conntrack_find_get); |
9fb9cbb1 | 603 | |
c1d10adb PNA |
604 | static void __nf_conntrack_hash_insert(struct nf_conn *ct, |
605 | unsigned int hash, | |
b476b72a | 606 | unsigned int reply_hash) |
c1d10adb | 607 | { |
ea781f19 | 608 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, |
56d52d48 | 609 | &nf_conntrack_hash[hash]); |
ea781f19 | 610 | hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, |
56d52d48 | 611 | &nf_conntrack_hash[reply_hash]); |
c1d10adb PNA |
612 | } |
613 | ||
7d367e06 JK |
614 | int |
615 | nf_conntrack_hash_check_insert(struct nf_conn *ct) | |
c1d10adb | 616 | { |
308ac914 | 617 | const struct nf_conntrack_zone *zone; |
d696c7bd | 618 | struct net *net = nf_ct_net(ct); |
b476b72a | 619 | unsigned int hash, reply_hash; |
7d367e06 JK |
620 | struct nf_conntrack_tuple_hash *h; |
621 | struct hlist_nulls_node *n; | |
93bb0ceb | 622 | unsigned int sequence; |
c1d10adb | 623 | |
5d0aa2cc | 624 | zone = nf_ct_zone(ct); |
7d367e06 | 625 | |
93bb0ceb JDB |
626 | local_bh_disable(); |
627 | do { | |
a3efd812 | 628 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
deedb590 | 629 | hash = hash_conntrack(net, |
93bb0ceb | 630 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
deedb590 | 631 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
632 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
633 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
7d367e06 JK |
634 | |
635 | /* See if there's one in the list already, including reverse */ | |
56d52d48 | 636 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 637 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 638 | zone, net)) |
7d367e06 | 639 | goto out; |
86804348 | 640 | |
56d52d48 | 641 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 642 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 643 | zone, net)) |
7d367e06 | 644 | goto out; |
c1d10adb | 645 | |
e53376be PNA |
646 | smp_wmb(); |
647 | /* The caller holds a reference to this object */ | |
648 | atomic_set(&ct->ct_general.use, 2); | |
b476b72a | 649 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 650 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 651 | NF_CT_STAT_INC(net, insert); |
93bb0ceb | 652 | local_bh_enable(); |
7d367e06 JK |
653 | return 0; |
654 | ||
655 | out: | |
93bb0ceb | 656 | nf_conntrack_double_unlock(hash, reply_hash); |
7d367e06 | 657 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 658 | local_bh_enable(); |
7d367e06 | 659 | return -EEXIST; |
c1d10adb | 660 | } |
7d367e06 | 661 | EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); |
c1d10adb | 662 | |
ba76738c PNA |
663 | static inline void nf_ct_acct_update(struct nf_conn *ct, |
664 | enum ip_conntrack_info ctinfo, | |
665 | unsigned int len) | |
666 | { | |
667 | struct nf_conn_acct *acct; | |
668 | ||
669 | acct = nf_conn_acct_find(ct); | |
670 | if (acct) { | |
671 | struct nf_conn_counter *counter = acct->counter; | |
672 | ||
673 | atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); | |
674 | atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes); | |
675 | } | |
676 | } | |
677 | ||
71d8c47f PNA |
678 | static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo, |
679 | const struct nf_conn *loser_ct) | |
680 | { | |
681 | struct nf_conn_acct *acct; | |
682 | ||
683 | acct = nf_conn_acct_find(loser_ct); | |
684 | if (acct) { | |
685 | struct nf_conn_counter *counter = acct->counter; | |
71d8c47f PNA |
686 | unsigned int bytes; |
687 | ||
688 | /* u32 should be fine since we must have seen one packet. */ | |
689 | bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes); | |
690 | nf_ct_acct_update(ct, ctinfo, bytes); | |
691 | } | |
692 | } | |
693 | ||
694 | /* Resolve race on insertion if this protocol allows this. */ | |
695 | static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb, | |
696 | enum ip_conntrack_info ctinfo, | |
697 | struct nf_conntrack_tuple_hash *h) | |
698 | { | |
699 | /* This is the conntrack entry already in hashes that won race. */ | |
700 | struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); | |
701 | struct nf_conntrack_l4proto *l4proto; | |
702 | ||
703 | l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); | |
704 | if (l4proto->allow_clash && | |
590b52e1 | 705 | !nfct_nat(ct) && |
71d8c47f PNA |
706 | !nf_ct_is_dying(ct) && |
707 | atomic_inc_not_zero(&ct->ct_general.use)) { | |
97a6ad13 FW |
708 | enum ip_conntrack_info oldinfo; |
709 | struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo); | |
710 | ||
711 | nf_ct_acct_merge(ct, ctinfo, loser_ct); | |
712 | nf_conntrack_put(&loser_ct->ct_general); | |
c74454fa | 713 | nf_ct_set(skb, ct, oldinfo); |
71d8c47f PNA |
714 | return NF_ACCEPT; |
715 | } | |
716 | NF_CT_STAT_INC(net, drop); | |
717 | return NF_DROP; | |
718 | } | |
719 | ||
9fb9cbb1 YK |
720 | /* Confirm a connection given skb; places it in hash table */ |
721 | int | |
3db05fea | 722 | __nf_conntrack_confirm(struct sk_buff *skb) |
9fb9cbb1 | 723 | { |
308ac914 | 724 | const struct nf_conntrack_zone *zone; |
b476b72a | 725 | unsigned int hash, reply_hash; |
df0933dc | 726 | struct nf_conntrack_tuple_hash *h; |
9fb9cbb1 | 727 | struct nf_conn *ct; |
df0933dc | 728 | struct nf_conn_help *help; |
a992ca2a | 729 | struct nf_conn_tstamp *tstamp; |
ea781f19 | 730 | struct hlist_nulls_node *n; |
9fb9cbb1 | 731 | enum ip_conntrack_info ctinfo; |
400dad39 | 732 | struct net *net; |
93bb0ceb | 733 | unsigned int sequence; |
71d8c47f | 734 | int ret = NF_DROP; |
9fb9cbb1 | 735 | |
3db05fea | 736 | ct = nf_ct_get(skb, &ctinfo); |
400dad39 | 737 | net = nf_ct_net(ct); |
9fb9cbb1 YK |
738 | |
739 | /* ipt_REJECT uses nf_conntrack_attach to attach related | |
740 | ICMP/TCP RST packets in other direction. Actual packet | |
741 | which created connection will be IP_CT_NEW or for an | |
742 | expected connection, IP_CT_RELATED. */ | |
743 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) | |
744 | return NF_ACCEPT; | |
745 | ||
5d0aa2cc | 746 | zone = nf_ct_zone(ct); |
93bb0ceb JDB |
747 | local_bh_disable(); |
748 | ||
749 | do { | |
a3efd812 | 750 | sequence = read_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb JDB |
751 | /* reuse the hash saved before */ |
752 | hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; | |
56d52d48 | 753 | hash = scale_hash(hash); |
deedb590 | 754 | reply_hash = hash_conntrack(net, |
93bb0ceb JDB |
755 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
756 | ||
757 | } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); | |
9fb9cbb1 YK |
758 | |
759 | /* We're not in hash table, and we refuse to set up related | |
93bb0ceb JDB |
760 | * connections for unconfirmed conns. But packet copies and |
761 | * REJECT will give spurious warnings here. | |
762 | */ | |
9fb9cbb1 YK |
763 | /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ |
764 | ||
25985edc | 765 | /* No external references means no one else could have |
93bb0ceb JDB |
766 | * confirmed us. |
767 | */ | |
9fb9cbb1 | 768 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); |
0d53778e | 769 | pr_debug("Confirming conntrack %p\n", ct); |
8ca3f5e9 PNA |
770 | /* We have to check the DYING flag after unlink to prevent |
771 | * a race against nf_ct_get_next_corpse() possibly called from | |
772 | * user context, else we insert an already 'dead' hash, blocking | |
773 | * further use of that particular connection -JM. | |
774 | */ | |
775 | nf_ct_del_from_dying_or_unconfirmed_list(ct); | |
776 | ||
71d8c47f PNA |
777 | if (unlikely(nf_ct_is_dying(ct))) { |
778 | nf_ct_add_to_dying_list(ct); | |
779 | goto dying; | |
780 | } | |
fc350777 | 781 | |
9fb9cbb1 YK |
782 | /* See if there's one in the list already, including reverse: |
783 | NAT could have grabbed it without realizing, since we're | |
784 | not in the hash. If there is, we lost race. */ | |
56d52d48 | 785 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) |
86804348 | 786 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, |
e0c7d472 | 787 | zone, net)) |
df0933dc | 788 | goto out; |
86804348 | 789 | |
56d52d48 | 790 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) |
86804348 | 791 | if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, |
e0c7d472 | 792 | zone, net)) |
df0933dc | 793 | goto out; |
9fb9cbb1 | 794 | |
df0933dc PM |
795 | /* Timer relative to confirmation time, not original |
796 | setting time, otherwise we'd get timer wrap in | |
797 | weird delay cases. */ | |
f330a7fd | 798 | ct->timeout += nfct_time_stamp; |
df0933dc | 799 | atomic_inc(&ct->ct_general.use); |
45eec341 | 800 | ct->status |= IPS_CONFIRMED; |
5c8ec910 | 801 | |
a992ca2a PNA |
802 | /* set conntrack timestamp, if enabled. */ |
803 | tstamp = nf_conn_tstamp_find(ct); | |
804 | if (tstamp) { | |
2456e855 | 805 | if (skb->tstamp == 0) |
e3192690 | 806 | __net_timestamp(skb); |
a992ca2a PNA |
807 | |
808 | tstamp->start = ktime_to_ns(skb->tstamp); | |
809 | } | |
5c8ec910 PM |
810 | /* Since the lookup is lockless, hash insertion must be done after |
811 | * starting the timer and setting the CONFIRMED bit. The RCU barriers | |
812 | * guarantee that no other CPU can find the conntrack before the above | |
813 | * stores are visible. | |
814 | */ | |
b476b72a | 815 | __nf_conntrack_hash_insert(ct, hash, reply_hash); |
93bb0ceb | 816 | nf_conntrack_double_unlock(hash, reply_hash); |
93bb0ceb | 817 | local_bh_enable(); |
5c8ec910 | 818 | |
df0933dc PM |
819 | help = nfct_help(ct); |
820 | if (help && help->helper) | |
a71996fc | 821 | nf_conntrack_event_cache(IPCT_HELPER, ct); |
17e6e4ea | 822 | |
df0933dc | 823 | nf_conntrack_event_cache(master_ct(ct) ? |
a71996fc | 824 | IPCT_RELATED : IPCT_NEW, ct); |
df0933dc | 825 | return NF_ACCEPT; |
9fb9cbb1 | 826 | |
df0933dc | 827 | out: |
8ca3f5e9 | 828 | nf_ct_add_to_dying_list(ct); |
71d8c47f PNA |
829 | ret = nf_ct_resolve_clash(net, skb, ctinfo, h); |
830 | dying: | |
93bb0ceb | 831 | nf_conntrack_double_unlock(hash, reply_hash); |
0d55af87 | 832 | NF_CT_STAT_INC(net, insert_failed); |
93bb0ceb | 833 | local_bh_enable(); |
71d8c47f | 834 | return ret; |
9fb9cbb1 | 835 | } |
13b18339 | 836 | EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); |
9fb9cbb1 YK |
837 | |
838 | /* Returns true if a connection correspondings to the tuple (required | |
839 | for NAT). */ | |
840 | int | |
841 | nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, | |
842 | const struct nf_conn *ignored_conntrack) | |
843 | { | |
400dad39 | 844 | struct net *net = nf_ct_net(ignored_conntrack); |
308ac914 | 845 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 | 846 | struct nf_conntrack_tuple_hash *h; |
5e3c61f9 | 847 | struct hlist_nulls_head *ct_hash; |
92e47ba8 | 848 | unsigned int hash, hsize; |
ea781f19 | 849 | struct hlist_nulls_node *n; |
5d0aa2cc | 850 | struct nf_conn *ct; |
308ac914 DB |
851 | |
852 | zone = nf_ct_zone(ignored_conntrack); | |
9fb9cbb1 | 853 | |
2cf12348 | 854 | rcu_read_lock(); |
95a8d19f | 855 | begin: |
92e47ba8 LZ |
856 | nf_conntrack_get_ht(&ct_hash, &hsize); |
857 | hash = __hash_conntrack(net, tuple, hsize); | |
5e3c61f9 FW |
858 | |
859 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) { | |
5d0aa2cc | 860 | ct = nf_ct_tuplehash_to_ctrack(h); |
f330a7fd FW |
861 | |
862 | if (ct == ignored_conntrack) | |
863 | continue; | |
864 | ||
865 | if (nf_ct_is_expired(ct)) { | |
866 | nf_ct_gc_expired(ct); | |
867 | continue; | |
868 | } | |
869 | ||
870 | if (nf_ct_key_equal(h, tuple, zone, net)) { | |
2cf12348 FW |
871 | NF_CT_STAT_INC_ATOMIC(net, found); |
872 | rcu_read_unlock(); | |
ba419aff PM |
873 | return 1; |
874 | } | |
ba419aff | 875 | } |
95a8d19f FW |
876 | |
877 | if (get_nulls_value(n) != hash) { | |
878 | NF_CT_STAT_INC_ATOMIC(net, search_restart); | |
879 | goto begin; | |
880 | } | |
881 | ||
2cf12348 | 882 | rcu_read_unlock(); |
9fb9cbb1 | 883 | |
ba419aff | 884 | return 0; |
9fb9cbb1 | 885 | } |
13b18339 | 886 | EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); |
9fb9cbb1 | 887 | |
7ae7730f PM |
888 | #define NF_CT_EVICTION_RANGE 8 |
889 | ||
9fb9cbb1 YK |
890 | /* There's a small race here where we may free a just-assured |
891 | connection. Too bad: we're in trouble anyway. */ | |
242922a0 FW |
892 | static unsigned int early_drop_list(struct net *net, |
893 | struct hlist_nulls_head *head) | |
9fb9cbb1 | 894 | { |
9fb9cbb1 | 895 | struct nf_conntrack_tuple_hash *h; |
ea781f19 | 896 | struct hlist_nulls_node *n; |
242922a0 FW |
897 | unsigned int drops = 0; |
898 | struct nf_conn *tmp; | |
3e86638e | 899 | |
242922a0 FW |
900 | hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) { |
901 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
9fb9cbb1 | 902 | |
f330a7fd FW |
903 | if (nf_ct_is_expired(tmp)) { |
904 | nf_ct_gc_expired(tmp); | |
905 | continue; | |
906 | } | |
907 | ||
242922a0 FW |
908 | if (test_bit(IPS_ASSURED_BIT, &tmp->status) || |
909 | !net_eq(nf_ct_net(tmp), net) || | |
910 | nf_ct_is_dying(tmp)) | |
911 | continue; | |
76507f69 | 912 | |
242922a0 FW |
913 | if (!atomic_inc_not_zero(&tmp->ct_general.use)) |
914 | continue; | |
76507f69 | 915 | |
242922a0 | 916 | /* kill only if still in same netns -- might have moved due to |
5f0d5a3a | 917 | * SLAB_TYPESAFE_BY_RCU rules. |
242922a0 FW |
918 | * |
919 | * We steal the timer reference. If that fails timer has | |
920 | * already fired or someone else deleted it. Just drop ref | |
921 | * and move to next entry. | |
922 | */ | |
923 | if (net_eq(nf_ct_net(tmp), net) && | |
924 | nf_ct_is_confirmed(tmp) && | |
242922a0 FW |
925 | nf_ct_delete(tmp, 0, 0)) |
926 | drops++; | |
927 | ||
928 | nf_ct_put(tmp); | |
9fb9cbb1 | 929 | } |
3e86638e | 930 | |
242922a0 FW |
931 | return drops; |
932 | } | |
9fb9cbb1 | 933 | |
242922a0 FW |
934 | static noinline int early_drop(struct net *net, unsigned int _hash) |
935 | { | |
936 | unsigned int i; | |
9fb9cbb1 | 937 | |
242922a0 FW |
938 | for (i = 0; i < NF_CT_EVICTION_RANGE; i++) { |
939 | struct hlist_nulls_head *ct_hash; | |
92e47ba8 | 940 | unsigned int hash, hsize, drops; |
242922a0 | 941 | |
3101e0fc | 942 | rcu_read_lock(); |
92e47ba8 LZ |
943 | nf_conntrack_get_ht(&ct_hash, &hsize); |
944 | hash = reciprocal_scale(_hash++, hsize); | |
242922a0 FW |
945 | |
946 | drops = early_drop_list(net, &ct_hash[hash]); | |
3101e0fc LZ |
947 | rcu_read_unlock(); |
948 | ||
242922a0 FW |
949 | if (drops) { |
950 | NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops); | |
951 | return true; | |
74138511 | 952 | } |
9fb9cbb1 | 953 | } |
3e86638e | 954 | |
242922a0 | 955 | return false; |
9fb9cbb1 YK |
956 | } |
957 | ||
b87a2f91 FW |
958 | static void gc_worker(struct work_struct *work) |
959 | { | |
e5072053 | 960 | unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u); |
b87a2f91 | 961 | unsigned int i, goal, buckets = 0, expired_count = 0; |
b87a2f91 | 962 | struct conntrack_gc_work *gc_work; |
e0df8cae FW |
963 | unsigned int ratio, scanned = 0; |
964 | unsigned long next_run; | |
b87a2f91 FW |
965 | |
966 | gc_work = container_of(work, struct conntrack_gc_work, dwork.work); | |
967 | ||
e0df8cae | 968 | goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV; |
b87a2f91 FW |
969 | i = gc_work->last_bucket; |
970 | ||
971 | do { | |
972 | struct nf_conntrack_tuple_hash *h; | |
973 | struct hlist_nulls_head *ct_hash; | |
974 | struct hlist_nulls_node *n; | |
975 | unsigned int hashsz; | |
976 | struct nf_conn *tmp; | |
977 | ||
978 | i++; | |
979 | rcu_read_lock(); | |
980 | ||
981 | nf_conntrack_get_ht(&ct_hash, &hashsz); | |
982 | if (i >= hashsz) | |
983 | i = 0; | |
984 | ||
985 | hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) { | |
986 | tmp = nf_ct_tuplehash_to_ctrack(h); | |
987 | ||
c023c0e4 | 988 | scanned++; |
b87a2f91 FW |
989 | if (nf_ct_is_expired(tmp)) { |
990 | nf_ct_gc_expired(tmp); | |
991 | expired_count++; | |
992 | continue; | |
993 | } | |
994 | } | |
995 | ||
996 | /* could check get_nulls_value() here and restart if ct | |
997 | * was moved to another chain. But given gc is best-effort | |
998 | * we will just continue with next hash slot. | |
999 | */ | |
1000 | rcu_read_unlock(); | |
1001 | cond_resched_rcu_qs(); | |
524b698d | 1002 | } while (++buckets < goal); |
b87a2f91 FW |
1003 | |
1004 | if (gc_work->exiting) | |
1005 | return; | |
1006 | ||
e0df8cae FW |
1007 | /* |
1008 | * Eviction will normally happen from the packet path, and not | |
1009 | * from this gc worker. | |
1010 | * | |
1011 | * This worker is only here to reap expired entries when system went | |
1012 | * idle after a busy period. | |
1013 | * | |
1014 | * The heuristics below are supposed to balance conflicting goals: | |
1015 | * | |
1016 | * 1. Minimize time until we notice a stale entry | |
1017 | * 2. Maximize scan intervals to not waste cycles | |
1018 | * | |
e5072053 | 1019 | * Normally, expire ratio will be close to 0. |
e0df8cae | 1020 | * |
e5072053 FW |
1021 | * As soon as a sizeable fraction of the entries have expired |
1022 | * increase scan frequency. | |
e0df8cae | 1023 | */ |
c023c0e4 | 1024 | ratio = scanned ? expired_count * 100 / scanned : 0; |
e5072053 FW |
1025 | if (ratio > GC_EVICT_RATIO) { |
1026 | gc_work->next_gc_run = min_interval; | |
e0df8cae | 1027 | } else { |
e5072053 | 1028 | unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV; |
e0df8cae | 1029 | |
e5072053 FW |
1030 | BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0); |
1031 | ||
1032 | gc_work->next_gc_run += min_interval; | |
1033 | if (gc_work->next_gc_run > max) | |
1034 | gc_work->next_gc_run = max; | |
e0df8cae | 1035 | } |
c023c0e4 | 1036 | |
e5072053 | 1037 | next_run = gc_work->next_gc_run; |
b87a2f91 | 1038 | gc_work->last_bucket = i; |
e0df8cae | 1039 | queue_delayed_work(system_long_wq, &gc_work->dwork, next_run); |
b87a2f91 FW |
1040 | } |
1041 | ||
1042 | static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work) | |
1043 | { | |
1044 | INIT_DELAYED_WORK(&gc_work->dwork, gc_worker); | |
e5072053 | 1045 | gc_work->next_gc_run = HZ; |
b87a2f91 FW |
1046 | gc_work->exiting = false; |
1047 | } | |
1048 | ||
99f07e91 | 1049 | static struct nf_conn * |
308ac914 DB |
1050 | __nf_conntrack_alloc(struct net *net, |
1051 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1052 | const struct nf_conntrack_tuple *orig, |
1053 | const struct nf_conntrack_tuple *repl, | |
1054 | gfp_t gfp, u32 hash) | |
9fb9cbb1 | 1055 | { |
cd7fcbf1 | 1056 | struct nf_conn *ct; |
9fb9cbb1 | 1057 | |
5251e2d2 | 1058 | /* We don't want any race condition at early drop stage */ |
49ac8713 | 1059 | atomic_inc(&net->ct.count); |
5251e2d2 | 1060 | |
76eb9460 | 1061 | if (nf_conntrack_max && |
49ac8713 | 1062 | unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { |
93bb0ceb | 1063 | if (!early_drop(net, hash)) { |
49ac8713 | 1064 | atomic_dec(&net->ct.count); |
e87cc472 | 1065 | net_warn_ratelimited("nf_conntrack: table full, dropping packet\n"); |
9fb9cbb1 YK |
1066 | return ERR_PTR(-ENOMEM); |
1067 | } | |
1068 | } | |
1069 | ||
941297f4 ED |
1070 | /* |
1071 | * Do not use kmem_cache_zalloc(), as this cache uses | |
5f0d5a3a | 1072 | * SLAB_TYPESAFE_BY_RCU. |
941297f4 | 1073 | */ |
0c5366b3 | 1074 | ct = kmem_cache_alloc(nf_conntrack_cachep, gfp); |
5e8018fc DB |
1075 | if (ct == NULL) |
1076 | goto out; | |
1077 | ||
440f0d58 | 1078 | spin_lock_init(&ct->lock); |
c88130bc | 1079 | ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; |
941297f4 | 1080 | ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; |
c88130bc | 1081 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; |
99f07e91 CG |
1082 | /* save hash for reusing when confirming */ |
1083 | *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; | |
c41884ce | 1084 | ct->status = 0; |
c2d9ba9b | 1085 | write_pnet(&ct->ct_net, net); |
c41884ce FW |
1086 | memset(&ct->__nfct_init_offset[0], 0, |
1087 | offsetof(struct nf_conn, proto) - | |
1088 | offsetof(struct nf_conn, __nfct_init_offset[0])); | |
5e8018fc | 1089 | |
6c8dee98 | 1090 | nf_ct_zone_add(ct, zone); |
5e8018fc | 1091 | |
e53376be PNA |
1092 | /* Because we use RCU lookups, we set ct_general.use to zero before |
1093 | * this is inserted in any list. | |
941297f4 | 1094 | */ |
e53376be | 1095 | atomic_set(&ct->ct_general.use, 0); |
c88130bc | 1096 | return ct; |
5e8018fc DB |
1097 | out: |
1098 | atomic_dec(&net->ct.count); | |
5d0aa2cc | 1099 | return ERR_PTR(-ENOMEM); |
9fb9cbb1 | 1100 | } |
99f07e91 | 1101 | |
308ac914 DB |
1102 | struct nf_conn *nf_conntrack_alloc(struct net *net, |
1103 | const struct nf_conntrack_zone *zone, | |
99f07e91 CG |
1104 | const struct nf_conntrack_tuple *orig, |
1105 | const struct nf_conntrack_tuple *repl, | |
1106 | gfp_t gfp) | |
1107 | { | |
1108 | return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); | |
1109 | } | |
13b18339 | 1110 | EXPORT_SYMBOL_GPL(nf_conntrack_alloc); |
9fb9cbb1 | 1111 | |
c88130bc | 1112 | void nf_conntrack_free(struct nf_conn *ct) |
76507f69 | 1113 | { |
1d45209d ED |
1114 | struct net *net = nf_ct_net(ct); |
1115 | ||
e53376be | 1116 | /* A freed object has refcnt == 0, that's |
5f0d5a3a | 1117 | * the golden rule for SLAB_TYPESAFE_BY_RCU |
e53376be PNA |
1118 | */ |
1119 | NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0); | |
1120 | ||
ceeff754 | 1121 | nf_ct_ext_destroy(ct); |
ea781f19 | 1122 | nf_ct_ext_free(ct); |
0c5366b3 | 1123 | kmem_cache_free(nf_conntrack_cachep, ct); |
4e857c58 | 1124 | smp_mb__before_atomic(); |
0c3c6c00 | 1125 | atomic_dec(&net->ct.count); |
76507f69 | 1126 | } |
13b18339 | 1127 | EXPORT_SYMBOL_GPL(nf_conntrack_free); |
9fb9cbb1 | 1128 | |
c539f017 | 1129 | |
9fb9cbb1 YK |
1130 | /* Allocate a new conntrack: we return -ENOMEM if classification |
1131 | failed due to stress. Otherwise it really is unclassifiable. */ | |
1132 | static struct nf_conntrack_tuple_hash * | |
b2a15a60 | 1133 | init_conntrack(struct net *net, struct nf_conn *tmpl, |
5a1fb391 | 1134 | const struct nf_conntrack_tuple *tuple, |
9fb9cbb1 | 1135 | struct nf_conntrack_l3proto *l3proto, |
605dcad6 | 1136 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 1137 | struct sk_buff *skb, |
60b5f8f7 | 1138 | unsigned int dataoff, u32 hash) |
9fb9cbb1 | 1139 | { |
c88130bc | 1140 | struct nf_conn *ct; |
3c158f7f | 1141 | struct nf_conn_help *help; |
9fb9cbb1 | 1142 | struct nf_conntrack_tuple repl_tuple; |
b2a15a60 | 1143 | struct nf_conntrack_ecache *ecache; |
ca7433df | 1144 | struct nf_conntrack_expect *exp = NULL; |
308ac914 | 1145 | const struct nf_conntrack_zone *zone; |
60b5f8f7 | 1146 | struct nf_conn_timeout *timeout_ext; |
5e8018fc | 1147 | struct nf_conntrack_zone tmp; |
60b5f8f7 | 1148 | unsigned int *timeouts; |
9fb9cbb1 | 1149 | |
605dcad6 | 1150 | if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { |
0d53778e | 1151 | pr_debug("Can't invert tuple.\n"); |
9fb9cbb1 YK |
1152 | return NULL; |
1153 | } | |
1154 | ||
5e8018fc | 1155 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
99f07e91 CG |
1156 | ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, |
1157 | hash); | |
0a9ee813 | 1158 | if (IS_ERR(ct)) |
c88130bc | 1159 | return (struct nf_conntrack_tuple_hash *)ct; |
9fb9cbb1 | 1160 | |
4440a2ab GF |
1161 | if (!nf_ct_add_synproxy(ct, tmpl)) { |
1162 | nf_conntrack_free(ct); | |
1163 | return ERR_PTR(-ENOMEM); | |
48b1de4c PM |
1164 | } |
1165 | ||
60b5f8f7 | 1166 | timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; |
ae2d708e PNA |
1167 | if (timeout_ext) { |
1168 | timeouts = nf_ct_timeout_data(timeout_ext); | |
1169 | if (unlikely(!timeouts)) | |
1170 | timeouts = l4proto->get_timeouts(net); | |
1171 | } else { | |
60b5f8f7 | 1172 | timeouts = l4proto->get_timeouts(net); |
ae2d708e | 1173 | } |
60b5f8f7 | 1174 | |
2c8503f5 | 1175 | if (!l4proto->new(ct, skb, dataoff, timeouts)) { |
c88130bc | 1176 | nf_conntrack_free(ct); |
ccd63c20 | 1177 | pr_debug("can't track with proto module\n"); |
9fb9cbb1 YK |
1178 | return NULL; |
1179 | } | |
1180 | ||
60b5f8f7 | 1181 | if (timeout_ext) |
ae2d708e PNA |
1182 | nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout), |
1183 | GFP_ATOMIC); | |
60b5f8f7 | 1184 | |
58401572 | 1185 | nf_ct_acct_ext_add(ct, GFP_ATOMIC); |
a992ca2a | 1186 | nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); |
c539f017 | 1187 | nf_ct_labels_ext_add(ct); |
b2a15a60 PM |
1188 | |
1189 | ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; | |
1190 | nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, | |
1191 | ecache ? ecache->expmask : 0, | |
1192 | GFP_ATOMIC); | |
58401572 | 1193 | |
ca7433df JDB |
1194 | local_bh_disable(); |
1195 | if (net->ct.expect_count) { | |
1196 | spin_lock(&nf_conntrack_expect_lock); | |
1197 | exp = nf_ct_find_expectation(net, zone, tuple); | |
1198 | if (exp) { | |
ccd63c20 | 1199 | pr_debug("expectation arrives ct=%p exp=%p\n", |
ca7433df JDB |
1200 | ct, exp); |
1201 | /* Welcome, Mr. Bond. We've been expecting you... */ | |
1202 | __set_bit(IPS_EXPECTED_BIT, &ct->status); | |
1203 | /* exp->master safe, refcnt bumped in nf_ct_find_expectation */ | |
1204 | ct->master = exp->master; | |
1205 | if (exp->helper) { | |
1206 | help = nf_ct_helper_ext_add(ct, exp->helper, | |
1207 | GFP_ATOMIC); | |
1208 | if (help) | |
1209 | rcu_assign_pointer(help->helper, exp->helper); | |
1210 | } | |
ceceae1b | 1211 | |
9fb9cbb1 | 1212 | #ifdef CONFIG_NF_CONNTRACK_MARK |
ca7433df | 1213 | ct->mark = exp->master->mark; |
7c9728c3 JM |
1214 | #endif |
1215 | #ifdef CONFIG_NF_CONNTRACK_SECMARK | |
ca7433df | 1216 | ct->secmark = exp->master->secmark; |
9fb9cbb1 | 1217 | #endif |
ca7433df JDB |
1218 | NF_CT_STAT_INC(net, expect_new); |
1219 | } | |
1220 | spin_unlock(&nf_conntrack_expect_lock); | |
1221 | } | |
8e8118f8 | 1222 | if (!exp) |
b2a15a60 | 1223 | __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); |
9fb9cbb1 | 1224 | |
e53376be PNA |
1225 | /* Now it is inserted into the unconfirmed list, bump refcount */ |
1226 | nf_conntrack_get(&ct->ct_general); | |
b7779d06 | 1227 | nf_ct_add_to_unconfirmed_list(ct); |
9fb9cbb1 | 1228 | |
ca7433df | 1229 | local_bh_enable(); |
9fb9cbb1 YK |
1230 | |
1231 | if (exp) { | |
1232 | if (exp->expectfn) | |
c88130bc | 1233 | exp->expectfn(ct, exp); |
6823645d | 1234 | nf_ct_expect_put(exp); |
9fb9cbb1 YK |
1235 | } |
1236 | ||
c88130bc | 1237 | return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; |
9fb9cbb1 YK |
1238 | } |
1239 | ||
a9e419dc | 1240 | /* On success, returns conntrack ptr, sets skb->_nfct | ctinfo */ |
9fb9cbb1 | 1241 | static inline struct nf_conn * |
b2a15a60 | 1242 | resolve_normal_ct(struct net *net, struct nf_conn *tmpl, |
a702a65f | 1243 | struct sk_buff *skb, |
9fb9cbb1 YK |
1244 | unsigned int dataoff, |
1245 | u_int16_t l3num, | |
1246 | u_int8_t protonum, | |
1247 | struct nf_conntrack_l3proto *l3proto, | |
605dcad6 | 1248 | struct nf_conntrack_l4proto *l4proto, |
9fb9cbb1 | 1249 | int *set_reply, |
60b5f8f7 | 1250 | enum ip_conntrack_info *ctinfo) |
9fb9cbb1 | 1251 | { |
308ac914 | 1252 | const struct nf_conntrack_zone *zone; |
9fb9cbb1 YK |
1253 | struct nf_conntrack_tuple tuple; |
1254 | struct nf_conntrack_tuple_hash *h; | |
5e8018fc | 1255 | struct nf_conntrack_zone tmp; |
9fb9cbb1 | 1256 | struct nf_conn *ct; |
99f07e91 | 1257 | u32 hash; |
9fb9cbb1 | 1258 | |
bbe735e4 | 1259 | if (!nf_ct_get_tuple(skb, skb_network_offset(skb), |
a31f1adc | 1260 | dataoff, l3num, protonum, net, &tuple, l3proto, |
605dcad6 | 1261 | l4proto)) { |
ccd63c20 | 1262 | pr_debug("Can't get tuple\n"); |
9fb9cbb1 YK |
1263 | return NULL; |
1264 | } | |
1265 | ||
1266 | /* look for tuple match */ | |
5e8018fc | 1267 | zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); |
1b8c8a9f | 1268 | hash = hash_conntrack_raw(&tuple, net); |
99f07e91 | 1269 | h = __nf_conntrack_find_get(net, zone, &tuple, hash); |
9fb9cbb1 | 1270 | if (!h) { |
b2a15a60 | 1271 | h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, |
60b5f8f7 | 1272 | skb, dataoff, hash); |
9fb9cbb1 YK |
1273 | if (!h) |
1274 | return NULL; | |
1275 | if (IS_ERR(h)) | |
1276 | return (void *)h; | |
1277 | } | |
1278 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1279 | ||
1280 | /* It exists; we have (non-exclusive) reference. */ | |
1281 | if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { | |
fb048833 | 1282 | *ctinfo = IP_CT_ESTABLISHED_REPLY; |
9fb9cbb1 YK |
1283 | /* Please set reply bit if this packet OK */ |
1284 | *set_reply = 1; | |
1285 | } else { | |
1286 | /* Once we've had two way comms, always ESTABLISHED. */ | |
1287 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { | |
ccd63c20 | 1288 | pr_debug("normal packet for %p\n", ct); |
9fb9cbb1 YK |
1289 | *ctinfo = IP_CT_ESTABLISHED; |
1290 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { | |
ccd63c20 | 1291 | pr_debug("related packet for %p\n", ct); |
9fb9cbb1 YK |
1292 | *ctinfo = IP_CT_RELATED; |
1293 | } else { | |
ccd63c20 | 1294 | pr_debug("new packet for %p\n", ct); |
9fb9cbb1 YK |
1295 | *ctinfo = IP_CT_NEW; |
1296 | } | |
1297 | *set_reply = 0; | |
1298 | } | |
c74454fa | 1299 | nf_ct_set(skb, ct, *ctinfo); |
9fb9cbb1 YK |
1300 | return ct; |
1301 | } | |
1302 | ||
1303 | unsigned int | |
a702a65f AD |
1304 | nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, |
1305 | struct sk_buff *skb) | |
9fb9cbb1 | 1306 | { |
97a6ad13 | 1307 | struct nf_conn *ct, *tmpl; |
9fb9cbb1 YK |
1308 | enum ip_conntrack_info ctinfo; |
1309 | struct nf_conntrack_l3proto *l3proto; | |
605dcad6 | 1310 | struct nf_conntrack_l4proto *l4proto; |
2c8503f5 | 1311 | unsigned int *timeouts; |
9fb9cbb1 YK |
1312 | unsigned int dataoff; |
1313 | u_int8_t protonum; | |
1314 | int set_reply = 0; | |
1315 | int ret; | |
1316 | ||
97a6ad13 FW |
1317 | tmpl = nf_ct_get(skb, &ctinfo); |
1318 | if (tmpl) { | |
b2a15a60 | 1319 | /* Previously seen (loopback or untracked)? Ignore. */ |
b2a15a60 PM |
1320 | if (!nf_ct_is_template(tmpl)) { |
1321 | NF_CT_STAT_INC_ATOMIC(net, ignore); | |
1322 | return NF_ACCEPT; | |
1323 | } | |
a9e419dc | 1324 | skb->_nfct = 0; |
9fb9cbb1 YK |
1325 | } |
1326 | ||
e2361cb9 | 1327 | /* rcu_read_lock()ed by nf_hook_thresh */ |
76108cea | 1328 | l3proto = __nf_ct_l3proto_find(pf); |
3db05fea | 1329 | ret = l3proto->get_l4proto(skb, skb_network_offset(skb), |
ffc30690 YK |
1330 | &dataoff, &protonum); |
1331 | if (ret <= 0) { | |
25985edc | 1332 | pr_debug("not prepared to track yet or error occurred\n"); |
0d55af87 AD |
1333 | NF_CT_STAT_INC_ATOMIC(net, error); |
1334 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1335 | ret = -ret; |
1336 | goto out; | |
9fb9cbb1 YK |
1337 | } |
1338 | ||
76108cea | 1339 | l4proto = __nf_ct_l4proto_find(pf, protonum); |
9fb9cbb1 YK |
1340 | |
1341 | /* It may be an special packet, error, unclean... | |
1342 | * inverse of the return code tells to the netfilter | |
1343 | * core what to do with the packet. */ | |
74c51a14 | 1344 | if (l4proto->error != NULL) { |
11df4b76 | 1345 | ret = l4proto->error(net, tmpl, skb, dataoff, pf, hooknum); |
74c51a14 | 1346 | if (ret <= 0) { |
0d55af87 AD |
1347 | NF_CT_STAT_INC_ATOMIC(net, error); |
1348 | NF_CT_STAT_INC_ATOMIC(net, invalid); | |
b2a15a60 PM |
1349 | ret = -ret; |
1350 | goto out; | |
74c51a14 | 1351 | } |
88ed01d1 | 1352 | /* ICMP[v6] protocol trackers may assign one conntrack. */ |
a9e419dc | 1353 | if (skb->_nfct) |
88ed01d1 | 1354 | goto out; |
9fb9cbb1 | 1355 | } |
08733a0c | 1356 | repeat: |
b2a15a60 | 1357 | ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, |
60b5f8f7 | 1358 | l3proto, l4proto, &set_reply, &ctinfo); |
9fb9cbb1 YK |
1359 | if (!ct) { |
1360 | /* Not valid part of a connection */ | |
0d55af87 | 1361 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
b2a15a60 PM |
1362 | ret = NF_ACCEPT; |
1363 | goto out; | |
9fb9cbb1 YK |
1364 | } |
1365 | ||
1366 | if (IS_ERR(ct)) { | |
1367 | /* Too stressed to deal. */ | |
0d55af87 | 1368 | NF_CT_STAT_INC_ATOMIC(net, drop); |
b2a15a60 PM |
1369 | ret = NF_DROP; |
1370 | goto out; | |
9fb9cbb1 YK |
1371 | } |
1372 | ||
cb9c6836 | 1373 | NF_CT_ASSERT(skb_nfct(skb)); |
9fb9cbb1 | 1374 | |
60b5f8f7 | 1375 | /* Decide what timeout policy we want to apply to this flow. */ |
84b5ee93 | 1376 | timeouts = nf_ct_timeout_lookup(net, ct, l4proto); |
60b5f8f7 | 1377 | |
2c8503f5 | 1378 | ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); |
ec8d5409 | 1379 | if (ret <= 0) { |
9fb9cbb1 YK |
1380 | /* Invalid: inverse of the return code tells |
1381 | * the netfilter core what to do */ | |
0d53778e | 1382 | pr_debug("nf_conntrack_in: Can't track with proto module\n"); |
97a6ad13 | 1383 | nf_conntrack_put(&ct->ct_general); |
a9e419dc | 1384 | skb->_nfct = 0; |
0d55af87 | 1385 | NF_CT_STAT_INC_ATOMIC(net, invalid); |
7d1e0459 PNA |
1386 | if (ret == -NF_DROP) |
1387 | NF_CT_STAT_INC_ATOMIC(net, drop); | |
56a62e22 AB |
1388 | /* Special case: TCP tracker reports an attempt to reopen a |
1389 | * closed/aborted connection. We have to go back and create a | |
1390 | * fresh conntrack. | |
1391 | */ | |
1392 | if (ret == -NF_REPEAT) | |
1393 | goto repeat; | |
b2a15a60 PM |
1394 | ret = -ret; |
1395 | goto out; | |
9fb9cbb1 YK |
1396 | } |
1397 | ||
1398 | if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) | |
858b3133 | 1399 | nf_conntrack_event_cache(IPCT_REPLY, ct); |
b2a15a60 | 1400 | out: |
56a62e22 AB |
1401 | if (tmpl) |
1402 | nf_ct_put(tmpl); | |
9fb9cbb1 YK |
1403 | |
1404 | return ret; | |
1405 | } | |
13b18339 | 1406 | EXPORT_SYMBOL_GPL(nf_conntrack_in); |
9fb9cbb1 | 1407 | |
5f2b4c90 JE |
1408 | bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, |
1409 | const struct nf_conntrack_tuple *orig) | |
9fb9cbb1 | 1410 | { |
5f2b4c90 | 1411 | bool ret; |
923f4902 PM |
1412 | |
1413 | rcu_read_lock(); | |
1414 | ret = nf_ct_invert_tuple(inverse, orig, | |
1415 | __nf_ct_l3proto_find(orig->src.l3num), | |
1416 | __nf_ct_l4proto_find(orig->src.l3num, | |
1417 | orig->dst.protonum)); | |
1418 | rcu_read_unlock(); | |
1419 | return ret; | |
9fb9cbb1 | 1420 | } |
13b18339 | 1421 | EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); |
9fb9cbb1 | 1422 | |
5b1158e9 JK |
1423 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
1424 | implicitly racy: see __nf_conntrack_confirm */ | |
1425 | void nf_conntrack_alter_reply(struct nf_conn *ct, | |
1426 | const struct nf_conntrack_tuple *newreply) | |
1427 | { | |
1428 | struct nf_conn_help *help = nfct_help(ct); | |
1429 | ||
5b1158e9 JK |
1430 | /* Should be unconfirmed, so not in hash table yet */ |
1431 | NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); | |
1432 | ||
0d53778e | 1433 | pr_debug("Altering reply tuple of %p to ", ct); |
3c9fba65 | 1434 | nf_ct_dump_tuple(newreply); |
5b1158e9 JK |
1435 | |
1436 | ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; | |
ef1a5a50 | 1437 | if (ct->master || (help && !hlist_empty(&help->expectations))) |
c52fbb41 | 1438 | return; |
ceceae1b | 1439 | |
c52fbb41 | 1440 | rcu_read_lock(); |
b2a15a60 | 1441 | __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); |
c52fbb41 | 1442 | rcu_read_unlock(); |
5b1158e9 | 1443 | } |
13b18339 | 1444 | EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); |
5b1158e9 | 1445 | |
9fb9cbb1 YK |
1446 | /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ |
1447 | void __nf_ct_refresh_acct(struct nf_conn *ct, | |
1448 | enum ip_conntrack_info ctinfo, | |
1449 | const struct sk_buff *skb, | |
1450 | unsigned long extra_jiffies, | |
1451 | int do_acct) | |
1452 | { | |
9fb9cbb1 YK |
1453 | NF_CT_ASSERT(skb); |
1454 | ||
997ae831 | 1455 | /* Only update if this is not a fixed timeout */ |
47d95045 PM |
1456 | if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) |
1457 | goto acct; | |
997ae831 | 1458 | |
9fb9cbb1 | 1459 | /* If not in hash table, timer will not be active yet */ |
f330a7fd FW |
1460 | if (nf_ct_is_confirmed(ct)) |
1461 | extra_jiffies += nfct_time_stamp; | |
9fb9cbb1 | 1462 | |
f330a7fd | 1463 | ct->timeout = extra_jiffies; |
47d95045 | 1464 | acct: |
ba76738c PNA |
1465 | if (do_acct) |
1466 | nf_ct_acct_update(ct, ctinfo, skb->len); | |
9fb9cbb1 | 1467 | } |
13b18339 | 1468 | EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); |
9fb9cbb1 | 1469 | |
ad66713f FW |
1470 | bool nf_ct_kill_acct(struct nf_conn *ct, |
1471 | enum ip_conntrack_info ctinfo, | |
1472 | const struct sk_buff *skb) | |
51091764 | 1473 | { |
ad66713f | 1474 | nf_ct_acct_update(ct, ctinfo, skb->len); |
58401572 | 1475 | |
f330a7fd | 1476 | return nf_ct_delete(ct, 0, 0); |
51091764 | 1477 | } |
ad66713f | 1478 | EXPORT_SYMBOL_GPL(nf_ct_kill_acct); |
51091764 | 1479 | |
c0cd1156 | 1480 | #if IS_ENABLED(CONFIG_NF_CT_NETLINK) |
c1d10adb PNA |
1481 | |
1482 | #include <linux/netfilter/nfnetlink.h> | |
1483 | #include <linux/netfilter/nfnetlink_conntrack.h> | |
57b47a53 IM |
1484 | #include <linux/mutex.h> |
1485 | ||
c1d10adb PNA |
1486 | /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be |
1487 | * in ip_conntrack_core, since we don't want the protocols to autoload | |
1488 | * or depend on ctnetlink */ | |
fdf70832 | 1489 | int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, |
c1d10adb PNA |
1490 | const struct nf_conntrack_tuple *tuple) |
1491 | { | |
bae65be8 DM |
1492 | if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) || |
1493 | nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port)) | |
1494 | goto nla_put_failure; | |
c1d10adb PNA |
1495 | return 0; |
1496 | ||
df6fb868 | 1497 | nla_put_failure: |
c1d10adb PNA |
1498 | return -1; |
1499 | } | |
fdf70832 | 1500 | EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); |
c1d10adb | 1501 | |
f73e924c PM |
1502 | const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { |
1503 | [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, | |
1504 | [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, | |
c1d10adb | 1505 | }; |
f73e924c | 1506 | EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); |
c1d10adb | 1507 | |
fdf70832 | 1508 | int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], |
c1d10adb PNA |
1509 | struct nf_conntrack_tuple *t) |
1510 | { | |
df6fb868 | 1511 | if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) |
c1d10adb PNA |
1512 | return -EINVAL; |
1513 | ||
77236b6e PM |
1514 | t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); |
1515 | t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); | |
c1d10adb PNA |
1516 | |
1517 | return 0; | |
1518 | } | |
fdf70832 | 1519 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); |
5c0de29d HE |
1520 | |
1521 | int nf_ct_port_nlattr_tuple_size(void) | |
1522 | { | |
1523 | return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); | |
1524 | } | |
1525 | EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); | |
c1d10adb PNA |
1526 | #endif |
1527 | ||
9fb9cbb1 | 1528 | /* Used by ipt_REJECT and ip6t_REJECT. */ |
312a0c16 | 1529 | static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb) |
9fb9cbb1 YK |
1530 | { |
1531 | struct nf_conn *ct; | |
1532 | enum ip_conntrack_info ctinfo; | |
1533 | ||
1534 | /* This ICMP is in reverse direction to the packet which caused it */ | |
1535 | ct = nf_ct_get(skb, &ctinfo); | |
1536 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) | |
fb048833 | 1537 | ctinfo = IP_CT_RELATED_REPLY; |
9fb9cbb1 YK |
1538 | else |
1539 | ctinfo = IP_CT_RELATED; | |
1540 | ||
1541 | /* Attach to new skbuff, and increment count */ | |
c74454fa | 1542 | nf_ct_set(nskb, ct, ctinfo); |
cb9c6836 | 1543 | nf_conntrack_get(skb_nfct(nskb)); |
9fb9cbb1 YK |
1544 | } |
1545 | ||
9fb9cbb1 | 1546 | /* Bring out ya dead! */ |
df0933dc | 1547 | static struct nf_conn * |
400dad39 | 1548 | get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), |
9fb9cbb1 YK |
1549 | void *data, unsigned int *bucket) |
1550 | { | |
df0933dc PM |
1551 | struct nf_conntrack_tuple_hash *h; |
1552 | struct nf_conn *ct; | |
ea781f19 | 1553 | struct hlist_nulls_node *n; |
b7779d06 | 1554 | int cpu; |
93bb0ceb | 1555 | spinlock_t *lockp; |
9fb9cbb1 | 1556 | |
56d52d48 | 1557 | for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { |
93bb0ceb JDB |
1558 | lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS]; |
1559 | local_bh_disable(); | |
b16c2919 | 1560 | nf_conntrack_lock(lockp); |
56d52d48 FW |
1561 | if (*bucket < nf_conntrack_htable_size) { |
1562 | hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) { | |
93bb0ceb JDB |
1563 | if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL) |
1564 | continue; | |
1565 | ct = nf_ct_tuplehash_to_ctrack(h); | |
e0c7d472 FW |
1566 | if (net_eq(nf_ct_net(ct), net) && |
1567 | iter(ct, data)) | |
93bb0ceb JDB |
1568 | goto found; |
1569 | } | |
df0933dc | 1570 | } |
93bb0ceb JDB |
1571 | spin_unlock(lockp); |
1572 | local_bh_enable(); | |
d93c6258 | 1573 | cond_resched(); |
601e68e1 | 1574 | } |
b7779d06 JDB |
1575 | |
1576 | for_each_possible_cpu(cpu) { | |
1577 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1578 | ||
1579 | spin_lock_bh(&pcpu->lock); | |
1580 | hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) { | |
1581 | ct = nf_ct_tuplehash_to_ctrack(h); | |
1582 | if (iter(ct, data)) | |
1583 | set_bit(IPS_DYING_BIT, &ct->status); | |
1584 | } | |
1585 | spin_unlock_bh(&pcpu->lock); | |
d93c6258 | 1586 | cond_resched(); |
b7779d06 | 1587 | } |
df0933dc PM |
1588 | return NULL; |
1589 | found: | |
c073e3fa | 1590 | atomic_inc(&ct->ct_general.use); |
93bb0ceb JDB |
1591 | spin_unlock(lockp); |
1592 | local_bh_enable(); | |
df0933dc | 1593 | return ct; |
9fb9cbb1 YK |
1594 | } |
1595 | ||
400dad39 AD |
1596 | void nf_ct_iterate_cleanup(struct net *net, |
1597 | int (*iter)(struct nf_conn *i, void *data), | |
c655bc68 | 1598 | void *data, u32 portid, int report) |
9fb9cbb1 | 1599 | { |
df0933dc | 1600 | struct nf_conn *ct; |
9fb9cbb1 YK |
1601 | unsigned int bucket = 0; |
1602 | ||
d93c6258 FW |
1603 | might_sleep(); |
1604 | ||
88b68bc5 FW |
1605 | if (atomic_read(&net->ct.count) == 0) |
1606 | return; | |
1607 | ||
400dad39 | 1608 | while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { |
9fb9cbb1 | 1609 | /* Time to push up daises... */ |
9fb9cbb1 | 1610 | |
f330a7fd | 1611 | nf_ct_delete(ct, portid, report); |
9fb9cbb1 | 1612 | nf_ct_put(ct); |
d93c6258 | 1613 | cond_resched(); |
9fb9cbb1 YK |
1614 | } |
1615 | } | |
13b18339 | 1616 | EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); |
9fb9cbb1 | 1617 | |
274d383b PNA |
1618 | static int kill_all(struct nf_conn *i, void *data) |
1619 | { | |
1620 | return 1; | |
1621 | } | |
1622 | ||
d862a662 | 1623 | void nf_ct_free_hashtable(void *hash, unsigned int size) |
9fb9cbb1 | 1624 | { |
d862a662 | 1625 | if (is_vmalloc_addr(hash)) |
9fb9cbb1 YK |
1626 | vfree(hash); |
1627 | else | |
601e68e1 | 1628 | free_pages((unsigned long)hash, |
f205c5e0 | 1629 | get_order(sizeof(struct hlist_head) * size)); |
9fb9cbb1 | 1630 | } |
ac565e5f | 1631 | EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); |
9fb9cbb1 | 1632 | |
b3c5163f ED |
1633 | static int untrack_refs(void) |
1634 | { | |
1635 | int cnt = 0, cpu; | |
1636 | ||
1637 | for_each_possible_cpu(cpu) { | |
1638 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
1639 | ||
1640 | cnt += atomic_read(&ct->ct_general.use) - 1; | |
1641 | } | |
1642 | return cnt; | |
1643 | } | |
1644 | ||
f94161c1 | 1645 | void nf_conntrack_cleanup_start(void) |
9fb9cbb1 | 1646 | { |
b87a2f91 | 1647 | conntrack_gc_work.exiting = true; |
f94161c1 G |
1648 | RCU_INIT_POINTER(ip_ct_attach, NULL); |
1649 | } | |
1650 | ||
1651 | void nf_conntrack_cleanup_end(void) | |
1652 | { | |
1653 | RCU_INIT_POINTER(nf_ct_destroy, NULL); | |
b3c5163f | 1654 | while (untrack_refs() > 0) |
9edd7ca0 PM |
1655 | schedule(); |
1656 | ||
b87a2f91 | 1657 | cancel_delayed_work_sync(&conntrack_gc_work.dwork); |
56d52d48 FW |
1658 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
1659 | ||
04d87001 | 1660 | nf_conntrack_proto_fini(); |
41d73ec0 | 1661 | nf_conntrack_seqadj_fini(); |
5f69b8f5 | 1662 | nf_conntrack_labels_fini(); |
5e615b22 | 1663 | nf_conntrack_helper_fini(); |
8684094c | 1664 | nf_conntrack_timeout_fini(); |
3fe0f943 | 1665 | nf_conntrack_ecache_fini(); |
73f4001a | 1666 | nf_conntrack_tstamp_fini(); |
b7ff3a1f | 1667 | nf_conntrack_acct_fini(); |
83b4dbe1 | 1668 | nf_conntrack_expect_fini(); |
77571149 FW |
1669 | |
1670 | kmem_cache_destroy(nf_conntrack_cachep); | |
08f6547d | 1671 | } |
9fb9cbb1 | 1672 | |
f94161c1 G |
1673 | /* |
1674 | * Mishearing the voices in his head, our hero wonders how he's | |
1675 | * supposed to kill the mall. | |
1676 | */ | |
1677 | void nf_conntrack_cleanup_net(struct net *net) | |
08f6547d | 1678 | { |
dece40e8 VD |
1679 | LIST_HEAD(single); |
1680 | ||
1681 | list_add(&net->exit_list, &single); | |
1682 | nf_conntrack_cleanup_net_list(&single); | |
1683 | } | |
1684 | ||
1685 | void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list) | |
1686 | { | |
1687 | int busy; | |
1688 | struct net *net; | |
1689 | ||
f94161c1 G |
1690 | /* |
1691 | * This makes sure all current packets have passed through | |
1692 | * netfilter framework. Roll on, two-stage module | |
1693 | * delete... | |
1694 | */ | |
1695 | synchronize_net(); | |
dece40e8 VD |
1696 | i_see_dead_people: |
1697 | busy = 0; | |
1698 | list_for_each_entry(net, net_exit_list, exit_list) { | |
c655bc68 | 1699 | nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0); |
dece40e8 VD |
1700 | if (atomic_read(&net->ct.count) != 0) |
1701 | busy = 1; | |
1702 | } | |
1703 | if (busy) { | |
9fb9cbb1 YK |
1704 | schedule(); |
1705 | goto i_see_dead_people; | |
1706 | } | |
1707 | ||
dece40e8 | 1708 | list_for_each_entry(net, net_exit_list, exit_list) { |
dece40e8 VD |
1709 | nf_conntrack_proto_pernet_fini(net); |
1710 | nf_conntrack_helper_pernet_fini(net); | |
1711 | nf_conntrack_ecache_pernet_fini(net); | |
1712 | nf_conntrack_tstamp_pernet_fini(net); | |
1713 | nf_conntrack_acct_pernet_fini(net); | |
1714 | nf_conntrack_expect_pernet_fini(net); | |
dece40e8 | 1715 | free_percpu(net->ct.stat); |
b7779d06 | 1716 | free_percpu(net->ct.pcpu_lists); |
dece40e8 | 1717 | } |
08f6547d AD |
1718 | } |
1719 | ||
d862a662 | 1720 | void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) |
9fb9cbb1 | 1721 | { |
ea781f19 ED |
1722 | struct hlist_nulls_head *hash; |
1723 | unsigned int nr_slots, i; | |
1724 | size_t sz; | |
9fb9cbb1 | 1725 | |
9cc1c73a FW |
1726 | if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head))) |
1727 | return NULL; | |
1728 | ||
ea781f19 ED |
1729 | BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); |
1730 | nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); | |
9cc1c73a FW |
1731 | |
1732 | if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head))) | |
1733 | return NULL; | |
1734 | ||
ea781f19 ED |
1735 | sz = nr_slots * sizeof(struct hlist_nulls_head); |
1736 | hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, | |
1737 | get_order(sz)); | |
f0ad4621 | 1738 | if (!hash) |
966567b7 | 1739 | hash = vzalloc(sz); |
9fb9cbb1 | 1740 | |
ea781f19 ED |
1741 | if (hash && nulls) |
1742 | for (i = 0; i < nr_slots; i++) | |
1743 | INIT_HLIST_NULLS_HEAD(&hash[i], i); | |
9fb9cbb1 YK |
1744 | |
1745 | return hash; | |
1746 | } | |
ac565e5f | 1747 | EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); |
9fb9cbb1 | 1748 | |
3183ab89 | 1749 | int nf_conntrack_hash_resize(unsigned int hashsize) |
9fb9cbb1 | 1750 | { |
3183ab89 FW |
1751 | int i, bucket; |
1752 | unsigned int old_size; | |
ea781f19 | 1753 | struct hlist_nulls_head *hash, *old_hash; |
9fb9cbb1 | 1754 | struct nf_conntrack_tuple_hash *h; |
5d0aa2cc | 1755 | struct nf_conn *ct; |
9fb9cbb1 | 1756 | |
9fb9cbb1 YK |
1757 | if (!hashsize) |
1758 | return -EINVAL; | |
1759 | ||
d862a662 | 1760 | hash = nf_ct_alloc_hashtable(&hashsize, 1); |
9fb9cbb1 YK |
1761 | if (!hash) |
1762 | return -ENOMEM; | |
1763 | ||
3183ab89 FW |
1764 | old_size = nf_conntrack_htable_size; |
1765 | if (old_size == hashsize) { | |
1766 | nf_ct_free_hashtable(hash, hashsize); | |
1767 | return 0; | |
1768 | } | |
1769 | ||
93bb0ceb JDB |
1770 | local_bh_disable(); |
1771 | nf_conntrack_all_lock(); | |
a3efd812 | 1772 | write_seqcount_begin(&nf_conntrack_generation); |
93bb0ceb | 1773 | |
76507f69 PM |
1774 | /* Lookups in the old hash might happen in parallel, which means we |
1775 | * might get false negatives during connection lookup. New connections | |
1776 | * created because of a false negative won't make it into the hash | |
93bb0ceb | 1777 | * though since that required taking the locks. |
76507f69 | 1778 | */ |
93bb0ceb | 1779 | |
56d52d48 FW |
1780 | for (i = 0; i < nf_conntrack_htable_size; i++) { |
1781 | while (!hlist_nulls_empty(&nf_conntrack_hash[i])) { | |
1782 | h = hlist_nulls_entry(nf_conntrack_hash[i].first, | |
1783 | struct nf_conntrack_tuple_hash, hnnode); | |
5d0aa2cc | 1784 | ct = nf_ct_tuplehash_to_ctrack(h); |
ea781f19 | 1785 | hlist_nulls_del_rcu(&h->hnnode); |
1b8c8a9f FW |
1786 | bucket = __hash_conntrack(nf_ct_net(ct), |
1787 | &h->tuple, hashsize); | |
ea781f19 | 1788 | hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); |
9fb9cbb1 YK |
1789 | } |
1790 | } | |
56d52d48 FW |
1791 | old_size = nf_conntrack_htable_size; |
1792 | old_hash = nf_conntrack_hash; | |
9fb9cbb1 | 1793 | |
56d52d48 FW |
1794 | nf_conntrack_hash = hash; |
1795 | nf_conntrack_htable_size = hashsize; | |
93bb0ceb | 1796 | |
a3efd812 | 1797 | write_seqcount_end(&nf_conntrack_generation); |
93bb0ceb JDB |
1798 | nf_conntrack_all_unlock(); |
1799 | local_bh_enable(); | |
9fb9cbb1 | 1800 | |
5e3c61f9 | 1801 | synchronize_net(); |
d862a662 | 1802 | nf_ct_free_hashtable(old_hash, old_size); |
9fb9cbb1 YK |
1803 | return 0; |
1804 | } | |
3183ab89 FW |
1805 | |
1806 | int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) | |
1807 | { | |
1808 | unsigned int hashsize; | |
1809 | int rc; | |
1810 | ||
1811 | if (current->nsproxy->net_ns != &init_net) | |
1812 | return -EOPNOTSUPP; | |
1813 | ||
1814 | /* On boot, we can set this without any fancy locking. */ | |
1815 | if (!nf_conntrack_htable_size) | |
1816 | return param_set_uint(val, kp); | |
1817 | ||
1818 | rc = kstrtouint(val, 0, &hashsize); | |
1819 | if (rc) | |
1820 | return rc; | |
1821 | ||
1822 | return nf_conntrack_hash_resize(hashsize); | |
1823 | } | |
fae718dd | 1824 | EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); |
9fb9cbb1 | 1825 | |
fae718dd | 1826 | module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, |
9fb9cbb1 YK |
1827 | &nf_conntrack_htable_size, 0600); |
1828 | ||
5bfddbd4 ED |
1829 | void nf_ct_untracked_status_or(unsigned long bits) |
1830 | { | |
b3c5163f ED |
1831 | int cpu; |
1832 | ||
1833 | for_each_possible_cpu(cpu) | |
1834 | per_cpu(nf_conntrack_untracked, cpu).status |= bits; | |
5bfddbd4 ED |
1835 | } |
1836 | EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); | |
1837 | ||
f94161c1 | 1838 | int nf_conntrack_init_start(void) |
9fb9cbb1 | 1839 | { |
f205c5e0 | 1840 | int max_factor = 8; |
0c5366b3 FW |
1841 | int ret = -ENOMEM; |
1842 | int i, cpu; | |
93bb0ceb | 1843 | |
a3efd812 FW |
1844 | seqcount_init(&nf_conntrack_generation); |
1845 | ||
d5d20912 | 1846 | for (i = 0; i < CONNTRACK_LOCKS; i++) |
93bb0ceb | 1847 | spin_lock_init(&nf_conntrack_locks[i]); |
9fb9cbb1 | 1848 | |
9fb9cbb1 | 1849 | if (!nf_conntrack_htable_size) { |
88eab472 ML |
1850 | /* Idea from tcp.c: use 1/16384 of memory. |
1851 | * On i386: 32MB machine has 512 buckets. | |
1852 | * >= 1GB machines have 16384 buckets. | |
1853 | * >= 4GB machines have 65536 buckets. | |
1854 | */ | |
9fb9cbb1 | 1855 | nf_conntrack_htable_size |
4481374c | 1856 | = (((totalram_pages << PAGE_SHIFT) / 16384) |
f205c5e0 | 1857 | / sizeof(struct hlist_head)); |
88eab472 ML |
1858 | if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE))) |
1859 | nf_conntrack_htable_size = 65536; | |
1860 | else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) | |
f205c5e0 PM |
1861 | nf_conntrack_htable_size = 16384; |
1862 | if (nf_conntrack_htable_size < 32) | |
1863 | nf_conntrack_htable_size = 32; | |
1864 | ||
1865 | /* Use a max. factor of four by default to get the same max as | |
1866 | * with the old struct list_heads. When a table size is given | |
1867 | * we use the old value of 8 to avoid reducing the max. | |
1868 | * entries. */ | |
1869 | max_factor = 4; | |
9fb9cbb1 | 1870 | } |
56d52d48 FW |
1871 | |
1872 | nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1); | |
1873 | if (!nf_conntrack_hash) | |
1874 | return -ENOMEM; | |
1875 | ||
f205c5e0 | 1876 | nf_conntrack_max = max_factor * nf_conntrack_htable_size; |
8e5105a0 | 1877 | |
0c5366b3 | 1878 | nf_conntrack_cachep = kmem_cache_create("nf_conntrack", |
a9e419dc FW |
1879 | sizeof(struct nf_conn), |
1880 | NFCT_INFOMASK + 1, | |
5f0d5a3a | 1881 | SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL); |
0c5366b3 FW |
1882 | if (!nf_conntrack_cachep) |
1883 | goto err_cachep; | |
1884 | ||
654d0fbd | 1885 | printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", |
8e5105a0 PM |
1886 | NF_CONNTRACK_VERSION, nf_conntrack_htable_size, |
1887 | nf_conntrack_max); | |
83b4dbe1 G |
1888 | |
1889 | ret = nf_conntrack_expect_init(); | |
1890 | if (ret < 0) | |
1891 | goto err_expect; | |
1892 | ||
b7ff3a1f G |
1893 | ret = nf_conntrack_acct_init(); |
1894 | if (ret < 0) | |
1895 | goto err_acct; | |
1896 | ||
73f4001a G |
1897 | ret = nf_conntrack_tstamp_init(); |
1898 | if (ret < 0) | |
1899 | goto err_tstamp; | |
1900 | ||
3fe0f943 G |
1901 | ret = nf_conntrack_ecache_init(); |
1902 | if (ret < 0) | |
1903 | goto err_ecache; | |
1904 | ||
8684094c G |
1905 | ret = nf_conntrack_timeout_init(); |
1906 | if (ret < 0) | |
1907 | goto err_timeout; | |
1908 | ||
5e615b22 G |
1909 | ret = nf_conntrack_helper_init(); |
1910 | if (ret < 0) | |
1911 | goto err_helper; | |
1912 | ||
5f69b8f5 G |
1913 | ret = nf_conntrack_labels_init(); |
1914 | if (ret < 0) | |
1915 | goto err_labels; | |
1916 | ||
41d73ec0 PM |
1917 | ret = nf_conntrack_seqadj_init(); |
1918 | if (ret < 0) | |
1919 | goto err_seqadj; | |
1920 | ||
04d87001 G |
1921 | ret = nf_conntrack_proto_init(); |
1922 | if (ret < 0) | |
1923 | goto err_proto; | |
1924 | ||
9edd7ca0 | 1925 | /* Set up fake conntrack: to never be deleted, not in any hashes */ |
b3c5163f ED |
1926 | for_each_possible_cpu(cpu) { |
1927 | struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); | |
b3c5163f ED |
1928 | write_pnet(&ct->ct_net, &init_net); |
1929 | atomic_set(&ct->ct_general.use, 1); | |
1930 | } | |
9edd7ca0 | 1931 | /* - and look it like as a confirmed connection */ |
5bfddbd4 | 1932 | nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); |
b87a2f91 FW |
1933 | |
1934 | conntrack_gc_work_init(&conntrack_gc_work); | |
e5072053 | 1935 | queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, HZ); |
b87a2f91 | 1936 | |
08f6547d AD |
1937 | return 0; |
1938 | ||
04d87001 | 1939 | err_proto: |
41d73ec0 PM |
1940 | nf_conntrack_seqadj_fini(); |
1941 | err_seqadj: | |
04d87001 | 1942 | nf_conntrack_labels_fini(); |
5f69b8f5 G |
1943 | err_labels: |
1944 | nf_conntrack_helper_fini(); | |
5e615b22 G |
1945 | err_helper: |
1946 | nf_conntrack_timeout_fini(); | |
8684094c G |
1947 | err_timeout: |
1948 | nf_conntrack_ecache_fini(); | |
3fe0f943 G |
1949 | err_ecache: |
1950 | nf_conntrack_tstamp_fini(); | |
73f4001a G |
1951 | err_tstamp: |
1952 | nf_conntrack_acct_fini(); | |
b7ff3a1f G |
1953 | err_acct: |
1954 | nf_conntrack_expect_fini(); | |
83b4dbe1 | 1955 | err_expect: |
0c5366b3 FW |
1956 | kmem_cache_destroy(nf_conntrack_cachep); |
1957 | err_cachep: | |
56d52d48 | 1958 | nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); |
08f6547d AD |
1959 | return ret; |
1960 | } | |
1961 | ||
f94161c1 G |
1962 | void nf_conntrack_init_end(void) |
1963 | { | |
1964 | /* For use by REJECT target */ | |
1965 | RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); | |
1966 | RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); | |
f94161c1 G |
1967 | } |
1968 | ||
8cc20198 ED |
1969 | /* |
1970 | * We need to use special "null" values, not used in hash table | |
1971 | */ | |
1972 | #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) | |
1973 | #define DYING_NULLS_VAL ((1<<30)+1) | |
252b3e8c | 1974 | #define TEMPLATE_NULLS_VAL ((1<<30)+2) |
8cc20198 | 1975 | |
f94161c1 | 1976 | int nf_conntrack_init_net(struct net *net) |
08f6547d | 1977 | { |
b7779d06 JDB |
1978 | int ret = -ENOMEM; |
1979 | int cpu; | |
ceceae1b | 1980 | |
08f6547d | 1981 | atomic_set(&net->ct.count, 0); |
b7779d06 JDB |
1982 | |
1983 | net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu); | |
1984 | if (!net->ct.pcpu_lists) | |
08f6547d | 1985 | goto err_stat; |
b7779d06 JDB |
1986 | |
1987 | for_each_possible_cpu(cpu) { | |
1988 | struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); | |
1989 | ||
1990 | spin_lock_init(&pcpu->lock); | |
1991 | INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL); | |
1992 | INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL); | |
08f6547d | 1993 | } |
5b3501fa | 1994 | |
b7779d06 JDB |
1995 | net->ct.stat = alloc_percpu(struct ip_conntrack_stat); |
1996 | if (!net->ct.stat) | |
1997 | goto err_pcpu_lists; | |
1998 | ||
83b4dbe1 | 1999 | ret = nf_conntrack_expect_pernet_init(net); |
08f6547d AD |
2000 | if (ret < 0) |
2001 | goto err_expect; | |
b7ff3a1f | 2002 | ret = nf_conntrack_acct_pernet_init(net); |
58401572 | 2003 | if (ret < 0) |
08f6547d | 2004 | goto err_acct; |
73f4001a | 2005 | ret = nf_conntrack_tstamp_pernet_init(net); |
a992ca2a PNA |
2006 | if (ret < 0) |
2007 | goto err_tstamp; | |
3fe0f943 | 2008 | ret = nf_conntrack_ecache_pernet_init(net); |
a0891aa6 PNA |
2009 | if (ret < 0) |
2010 | goto err_ecache; | |
5e615b22 | 2011 | ret = nf_conntrack_helper_pernet_init(net); |
a9006892 EL |
2012 | if (ret < 0) |
2013 | goto err_helper; | |
04d87001 | 2014 | ret = nf_conntrack_proto_pernet_init(net); |
f94161c1 G |
2015 | if (ret < 0) |
2016 | goto err_proto; | |
08f6547d | 2017 | return 0; |
c539f017 | 2018 | |
f94161c1 | 2019 | err_proto: |
5e615b22 | 2020 | nf_conntrack_helper_pernet_fini(net); |
a9006892 | 2021 | err_helper: |
3fe0f943 | 2022 | nf_conntrack_ecache_pernet_fini(net); |
a0891aa6 | 2023 | err_ecache: |
73f4001a | 2024 | nf_conntrack_tstamp_pernet_fini(net); |
a992ca2a | 2025 | err_tstamp: |
b7ff3a1f | 2026 | nf_conntrack_acct_pernet_fini(net); |
08f6547d | 2027 | err_acct: |
83b4dbe1 | 2028 | nf_conntrack_expect_pernet_fini(net); |
08f6547d | 2029 | err_expect: |
0d55af87 | 2030 | free_percpu(net->ct.stat); |
b7779d06 JDB |
2031 | err_pcpu_lists: |
2032 | free_percpu(net->ct.pcpu_lists); | |
0d55af87 | 2033 | err_stat: |
08f6547d AD |
2034 | return ret; |
2035 | } |