blob: 6d592f8555fb8bf15506d828b53c6582386f7bec [file] [log] [blame]
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic TIME_WAIT sockets functions
7 *
8 * From code orinally in TCP
9 */
10
Ilpo Järvinen172589c2007-08-28 15:50:33 -070011#include <linux/kernel.h>
Vegard Nossum9e337b02008-10-18 17:37:51 +020012#include <linux/kmemcheck.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090013#include <linux/slab.h>
Paul Gortmaker3a9a2312011-05-27 09:12:25 -040014#include <linux/module.h>
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070015#include <net/inet_hashtables.h>
16#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -070017#include <net/ip.h>
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070018
Eric Dumazet13475a32009-12-02 22:31:19 +000019
Eric Dumazet2a8875e2009-12-08 20:19:53 -080020/**
21 * inet_twsk_unhash - unhash a timewait socket from established hash
22 * @tw: timewait socket
23 *
24 * unhash a timewait socket from established hash, if hashed.
25 * ehash lock must be held by caller.
26 * Returns 1 if caller should call inet_twsk_put() after lock release.
Eric Dumazet13475a32009-12-02 22:31:19 +000027 */
28int inet_twsk_unhash(struct inet_timewait_sock *tw)
29{
30 if (hlist_nulls_unhashed(&tw->tw_node))
31 return 0;
32
33 hlist_nulls_del_rcu(&tw->tw_node);
34 sk_nulls_node_init(&tw->tw_node);
Eric Dumazet2a8875e2009-12-08 20:19:53 -080035 /*
36 * We cannot call inet_twsk_put() ourself under lock,
37 * caller must call it for us.
38 */
Eric Dumazet13475a32009-12-02 22:31:19 +000039 return 1;
40}
41
Eric Dumazet2a8875e2009-12-08 20:19:53 -080042/**
43 * inet_twsk_bind_unhash - unhash a timewait socket from bind hash
44 * @tw: timewait socket
45 * @hashinfo: hashinfo pointer
46 *
47 * unhash a timewait socket from bind hash, if hashed.
48 * bind hash lock must be held by caller.
49 * Returns 1 if caller should call inet_twsk_put() after lock release.
Eric Dumazet3cdaeda2009-12-04 03:47:42 +000050 */
51int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
52 struct inet_hashinfo *hashinfo)
53{
54 struct inet_bind_bucket *tb = tw->tw_tb;
55
56 if (!tb)
57 return 0;
58
59 __hlist_del(&tw->tw_bind_node);
60 tw->tw_tb = NULL;
61 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
Eric Dumazet2a8875e2009-12-08 20:19:53 -080062 /*
63 * We cannot call inet_twsk_put() ourself under lock,
64 * caller must call it for us.
65 */
Eric Dumazet3cdaeda2009-12-04 03:47:42 +000066 return 1;
67}
68
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070069/* Must be called with locally disabled BHs. */
Adrian Bunkacd159b2007-07-14 19:00:59 -070070static void __inet_twsk_kill(struct inet_timewait_sock *tw,
71 struct inet_hashinfo *hashinfo)
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070072{
73 struct inet_bind_hashbucket *bhead;
Eric Dumazet13475a32009-12-02 22:31:19 +000074 int refcnt;
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070075 /* Unlink from established hashes. */
Eric Dumazet9db66bd2008-11-20 20:39:09 -080076 spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070077
Eric Dumazet9db66bd2008-11-20 20:39:09 -080078 spin_lock(lock);
Eric Dumazet13475a32009-12-02 22:31:19 +000079 refcnt = inet_twsk_unhash(tw);
Eric Dumazet9db66bd2008-11-20 20:39:09 -080080 spin_unlock(lock);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070081
82 /* Disassociate with bind bucket. */
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -070083 bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num,
84 hashinfo->bhash_size)];
Eric Dumazet3cdaeda2009-12-04 03:47:42 +000085
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070086 spin_lock(&bhead->lock);
Eric Dumazet3cdaeda2009-12-04 03:47:42 +000087 refcnt += inet_twsk_bind_unhash(tw, hashinfo);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070088 spin_unlock(&bhead->lock);
Eric Dumazet3cdaeda2009-12-04 03:47:42 +000089
Eric Dumazet05dbc7b2013-10-03 00:22:02 -070090 BUG_ON(refcnt >= atomic_read(&tw->tw_refcnt));
91 atomic_sub(refcnt, &tw->tw_refcnt);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070092}
93
Eric Dumazet05dbc7b2013-10-03 00:22:02 -070094void inet_twsk_free(struct inet_timewait_sock *tw)
Arnaldo Carvalho de Melo4dbc8ef2009-05-06 16:50:52 -070095{
96 struct module *owner = tw->tw_prot->owner;
97 twsk_destructor((struct sock *)tw);
98#ifdef SOCK_REFCNT_DEBUG
99 pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw);
100#endif
101 release_net(twsk_net(tw));
102 kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
103 module_put(owner);
104}
105
Pavel Emelyanov7054fb92007-12-20 15:32:54 -0800106void inet_twsk_put(struct inet_timewait_sock *tw)
107{
Arnaldo Carvalho de Melo4dbc8ef2009-05-06 16:50:52 -0700108 if (atomic_dec_and_test(&tw->tw_refcnt))
109 inet_twsk_free(tw);
Pavel Emelyanov7054fb92007-12-20 15:32:54 -0800110}
111EXPORT_SYMBOL_GPL(inet_twsk_put);
112
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700113static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
114 struct hlist_nulls_head *list)
115{
116 hlist_nulls_add_head_rcu(&tw->tw_node, list);
117}
118
119static void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
120 struct hlist_head *list)
121{
122 hlist_add_head(&tw->tw_bind_node, list);
123}
124
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700125/*
126 * Enter the time wait state. This is called with locally disabled BH.
127 * Essentially we whip up a timewait bucket, copy the relevant info into it
128 * from the SK, and mess with hash chains and list linkage.
129 */
130void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
131 struct inet_hashinfo *hashinfo)
132{
133 const struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700134 const struct inet_connection_sock *icsk = inet_csk(sk);
Eric Dumazet81c3d542005-10-03 14:13:38 -0700135 struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800136 spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700137 struct inet_bind_hashbucket *bhead;
138 /* Step 1: Put TW into bind hash. Original socket stays there too.
139 Note, that any socket with inet->num != 0 MUST be bound in
140 binding cache, even if it is closed.
141 */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000142 bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num,
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700143 hashinfo->bhash_size)];
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700144 spin_lock(&bhead->lock);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700145 tw->tw_tb = icsk->icsk_bind_hash;
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700146 WARN_ON(!icsk->icsk_bind_hash);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700147 inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
148 spin_unlock(&bhead->lock);
149
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800150 spin_lock(lock);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700151
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800152 /*
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700153 * Step 2: Hash TW into tcp ehash chain.
154 * Notes :
155 * - tw_refcnt is set to 3 because :
156 * - We have one reference from bhash chain.
157 * - We have one reference from ehash chain.
158 * We can use atomic_set() because prior spin_lock()/spin_unlock()
159 * committed into memory all tw fields.
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800160 */
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700161 atomic_set(&tw->tw_refcnt, 1 + 1 + 1);
162 inet_twsk_add_node_rcu(tw, &ehead->chain);
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800163
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700164 /* Step 3: Remove SK from hash chain */
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800165 if (__sk_nulls_del_node_init_rcu(sk))
166 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700167
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800168 spin_unlock(lock);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700169}
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700170EXPORT_SYMBOL_GPL(__inet_twsk_hashdance);
171
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700172struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, const int state)
173{
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800174 struct inet_timewait_sock *tw =
175 kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
Christoph Lameter54e6ecb2006-12-06 20:33:16 -0800176 GFP_ATOMIC);
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700177 if (tw != NULL) {
178 const struct inet_sock *inet = inet_sk(sk);
179
Vegard Nossum9e337b02008-10-18 17:37:51 +0200180 kmemcheck_annotate_bitfield(tw, flags);
181
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700182 /* Give us an identity. */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000183 tw->tw_daddr = inet->inet_daddr;
184 tw->tw_rcv_saddr = inet->inet_rcv_saddr;
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700185 tw->tw_bound_dev_if = sk->sk_bound_dev_if;
Eric Dumazet66b13d92011-10-24 03:06:21 -0400186 tw->tw_tos = inet->tos;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000187 tw->tw_num = inet->inet_num;
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700188 tw->tw_state = TCP_TIME_WAIT;
189 tw->tw_substate = state;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000190 tw->tw_sport = inet->inet_sport;
191 tw->tw_dport = inet->inet_dport;
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700192 tw->tw_family = sk->sk_family;
193 tw->tw_reuse = sk->sk_reuse;
Eric Dumazet81c3d542005-10-03 14:13:38 -0700194 tw->tw_hash = sk->sk_hash;
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700195 tw->tw_ipv6only = 0;
KOVACS Krisztianf5715ae2008-10-01 07:30:02 -0700196 tw->tw_transparent = inet->transparent;
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700197 tw->tw_prot = sk->sk_prot_creator;
Denis V. Lunevcd5342d2008-04-16 02:00:28 -0700198 twsk_net_set(tw, hold_net(sock_net(sk)));
Eric Dumazet47e1c322009-12-03 00:49:01 +0000199 /*
200 * Because we use RCU lookups, we should not set tw_refcnt
201 * to a non null value before everything is setup for this
202 * timewait socket.
203 */
204 atomic_set(&tw->tw_refcnt, 0);
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700205 inet_twsk_dead_node_init(tw);
Arnaldo Carvalho de Meloeeb2b852005-10-10 21:25:23 -0700206 __module_get(tw->tw_prot->owner);
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700207 }
208
209 return tw;
210}
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700211EXPORT_SYMBOL_GPL(inet_twsk_alloc);
212
213/* Returns non-zero if quota exceeded. */
214static int inet_twdr_do_twkill_work(struct inet_timewait_death_row *twdr,
215 const int slot)
216{
217 struct inet_timewait_sock *tw;
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700218 unsigned int killed;
219 int ret;
220
221 /* NOTE: compare this to previous version where lock
222 * was released after detaching chain. It was racy,
223 * because tw buckets are scheduled in not serialized context
224 * in 2.3 (with netfilter), and with softnet it is common, because
225 * soft irqs are not sequenced.
226 */
227 killed = 0;
228 ret = 0;
229rescan:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800230 inet_twsk_for_each_inmate(tw, &twdr->cells[slot]) {
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700231 __inet_twsk_del_dead_node(tw);
232 spin_unlock(&twdr->death_lock);
233 __inet_twsk_kill(tw, twdr->hashinfo);
Pavel Emelyanovf2bf4152008-07-16 20:32:25 -0700234#ifdef CONFIG_NET_NS
235 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITED);
236#endif
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700237 inet_twsk_put(tw);
238 killed++;
239 spin_lock(&twdr->death_lock);
240 if (killed > INET_TWDR_TWKILL_QUOTA) {
241 ret = 1;
242 break;
243 }
244
245 /* While we dropped twdr->death_lock, another cpu may have
246 * killed off the next TW bucket in the list, therefore
247 * do a fresh re-read of the hlist head node with the
248 * lock reacquired. We still use the hlist traversal
249 * macro in order to get the prefetches.
250 */
251 goto rescan;
252 }
253
254 twdr->tw_count -= killed;
Pavel Emelyanovf2bf4152008-07-16 20:32:25 -0700255#ifndef CONFIG_NET_NS
256 NET_ADD_STATS_BH(&init_net, LINUX_MIB_TIMEWAITED, killed);
257#endif
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700258 return ret;
259}
260
261void inet_twdr_hangman(unsigned long data)
262{
263 struct inet_timewait_death_row *twdr;
Eric Dumazet95c96172012-04-15 05:58:06 +0000264 unsigned int need_timer;
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700265
266 twdr = (struct inet_timewait_death_row *)data;
267 spin_lock(&twdr->death_lock);
268
269 if (twdr->tw_count == 0)
270 goto out;
271
272 need_timer = 0;
273 if (inet_twdr_do_twkill_work(twdr, twdr->slot)) {
274 twdr->thread_slots |= (1 << twdr->slot);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700275 schedule_work(&twdr->twkill_work);
276 need_timer = 1;
277 } else {
278 /* We purged the entire slot, anything left? */
279 if (twdr->tw_count)
280 need_timer = 1;
Octavian Purdila80a10962009-08-29 00:00:35 -0700281 twdr->slot = ((twdr->slot + 1) & (INET_TWDR_TWKILL_SLOTS - 1));
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700282 }
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700283 if (need_timer)
284 mod_timer(&twdr->tw_timer, jiffies + twdr->period);
285out:
286 spin_unlock(&twdr->death_lock);
287}
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700288EXPORT_SYMBOL_GPL(inet_twdr_hangman);
289
David Howells65f27f32006-11-22 14:55:48 +0000290void inet_twdr_twkill_work(struct work_struct *work)
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700291{
David Howells65f27f32006-11-22 14:55:48 +0000292 struct inet_timewait_death_row *twdr =
293 container_of(work, struct inet_timewait_death_row, twkill_work);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700294 int i;
295
Pavel Emelyanov95c93822007-12-11 02:12:36 -0800296 BUILD_BUG_ON((INET_TWDR_TWKILL_SLOTS - 1) >
297 (sizeof(twdr->thread_slots) * 8));
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700298
299 while (twdr->thread_slots) {
300 spin_lock_bh(&twdr->death_lock);
301 for (i = 0; i < INET_TWDR_TWKILL_SLOTS; i++) {
302 if (!(twdr->thread_slots & (1 << i)))
303 continue;
304
305 while (inet_twdr_do_twkill_work(twdr, i) != 0) {
306 if (need_resched()) {
307 spin_unlock_bh(&twdr->death_lock);
308 schedule();
309 spin_lock_bh(&twdr->death_lock);
310 }
311 }
312
313 twdr->thread_slots &= ~(1 << i);
314 }
315 spin_unlock_bh(&twdr->death_lock);
316 }
317}
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700318EXPORT_SYMBOL_GPL(inet_twdr_twkill_work);
319
320/* These are always called from BH context. See callers in
321 * tcp_input.c to verify this.
322 */
323
324/* This is for handling early-kills of TIME_WAIT sockets. */
325void inet_twsk_deschedule(struct inet_timewait_sock *tw,
326 struct inet_timewait_death_row *twdr)
327{
328 spin_lock(&twdr->death_lock);
329 if (inet_twsk_del_dead_node(tw)) {
330 inet_twsk_put(tw);
331 if (--twdr->tw_count == 0)
332 del_timer(&twdr->tw_timer);
333 }
334 spin_unlock(&twdr->death_lock);
335 __inet_twsk_kill(tw, twdr->hashinfo);
336}
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700337EXPORT_SYMBOL(inet_twsk_deschedule);
338
339void inet_twsk_schedule(struct inet_timewait_sock *tw,
340 struct inet_timewait_death_row *twdr,
341 const int timeo, const int timewait_len)
342{
343 struct hlist_head *list;
344 int slot;
345
346 /* timeout := RTO * 3.5
347 *
348 * 3.5 = 1+2+0.5 to wait for two retransmits.
349 *
350 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
351 * our ACK acking that FIN can be lost. If N subsequent retransmitted
352 * FINs (or previous seqments) are lost (probability of such event
353 * is p^(N+1), where p is probability to lose single packet and
354 * time to detect the loss is about RTO*(2^N - 1) with exponential
355 * backoff). Normal timewait length is calculated so, that we
356 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
357 * [ BTW Linux. following BSD, violates this requirement waiting
358 * only for 60sec, we should wait at least for 240 secs.
359 * Well, 240 consumes too much of resources 8)
360 * ]
361 * This interval is not reduced to catch old duplicate and
362 * responces to our wandering segments living for two MSLs.
363 * However, if we use PAWS to detect
364 * old duplicates, we can reduce the interval to bounds required
365 * by RTO, rather than MSL. So, if peer understands PAWS, we
366 * kill tw bucket after 3.5*RTO (it is important that this number
367 * is greater than TS tick!) and detect old duplicates with help
368 * of PAWS.
369 */
370 slot = (timeo + (1 << INET_TWDR_RECYCLE_TICK) - 1) >> INET_TWDR_RECYCLE_TICK;
371
372 spin_lock(&twdr->death_lock);
373
374 /* Unlink it, if it was scheduled */
375 if (inet_twsk_del_dead_node(tw))
376 twdr->tw_count--;
377 else
378 atomic_inc(&tw->tw_refcnt);
379
380 if (slot >= INET_TWDR_RECYCLE_SLOTS) {
381 /* Schedule to slow timer */
382 if (timeo >= timewait_len) {
383 slot = INET_TWDR_TWKILL_SLOTS - 1;
384 } else {
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700385 slot = DIV_ROUND_UP(timeo, twdr->period);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700386 if (slot >= INET_TWDR_TWKILL_SLOTS)
387 slot = INET_TWDR_TWKILL_SLOTS - 1;
388 }
Eric Dumazet96f817f2013-10-03 14:27:25 -0700389 tw->tw_ttd = inet_tw_time_stamp() + timeo;
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700390 slot = (twdr->slot + slot) & (INET_TWDR_TWKILL_SLOTS - 1);
391 list = &twdr->cells[slot];
392 } else {
Eric Dumazet96f817f2013-10-03 14:27:25 -0700393 tw->tw_ttd = inet_tw_time_stamp() + (slot << INET_TWDR_RECYCLE_TICK);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700394
395 if (twdr->twcal_hand < 0) {
396 twdr->twcal_hand = 0;
397 twdr->twcal_jiffie = jiffies;
398 twdr->twcal_timer.expires = twdr->twcal_jiffie +
399 (slot << INET_TWDR_RECYCLE_TICK);
400 add_timer(&twdr->twcal_timer);
401 } else {
402 if (time_after(twdr->twcal_timer.expires,
403 jiffies + (slot << INET_TWDR_RECYCLE_TICK)))
404 mod_timer(&twdr->twcal_timer,
405 jiffies + (slot << INET_TWDR_RECYCLE_TICK));
406 slot = (twdr->twcal_hand + slot) & (INET_TWDR_RECYCLE_SLOTS - 1);
407 }
408 list = &twdr->twcal_row[slot];
409 }
410
411 hlist_add_head(&tw->tw_death_node, list);
412
413 if (twdr->tw_count++ == 0)
414 mod_timer(&twdr->tw_timer, jiffies + twdr->period);
415 spin_unlock(&twdr->death_lock);
416}
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700417EXPORT_SYMBOL_GPL(inet_twsk_schedule);
418
419void inet_twdr_twcal_tick(unsigned long data)
420{
421 struct inet_timewait_death_row *twdr;
422 int n, slot;
423 unsigned long j;
424 unsigned long now = jiffies;
425 int killed = 0;
426 int adv = 0;
427
428 twdr = (struct inet_timewait_death_row *)data;
429
430 spin_lock(&twdr->death_lock);
431 if (twdr->twcal_hand < 0)
432 goto out;
433
434 slot = twdr->twcal_hand;
435 j = twdr->twcal_jiffie;
436
437 for (n = 0; n < INET_TWDR_RECYCLE_SLOTS; n++) {
438 if (time_before_eq(j, now)) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800439 struct hlist_node *safe;
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700440 struct inet_timewait_sock *tw;
441
Sasha Levinb67bfe02013-02-27 17:06:00 -0800442 inet_twsk_for_each_inmate_safe(tw, safe,
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700443 &twdr->twcal_row[slot]) {
444 __inet_twsk_del_dead_node(tw);
445 __inet_twsk_kill(tw, twdr->hashinfo);
Pavel Emelyanovf2bf4152008-07-16 20:32:25 -0700446#ifdef CONFIG_NET_NS
447 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED);
448#endif
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700449 inet_twsk_put(tw);
450 killed++;
451 }
452 } else {
453 if (!adv) {
454 adv = 1;
455 twdr->twcal_jiffie = j;
456 twdr->twcal_hand = slot;
457 }
458
459 if (!hlist_empty(&twdr->twcal_row[slot])) {
460 mod_timer(&twdr->twcal_timer, j);
461 goto out;
462 }
463 }
464 j += 1 << INET_TWDR_RECYCLE_TICK;
465 slot = (slot + 1) & (INET_TWDR_RECYCLE_SLOTS - 1);
466 }
467 twdr->twcal_hand = -1;
468
469out:
470 if ((twdr->tw_count -= killed) == 0)
471 del_timer(&twdr->tw_timer);
Pavel Emelyanovf2bf4152008-07-16 20:32:25 -0700472#ifndef CONFIG_NET_NS
473 NET_ADD_STATS_BH(&init_net, LINUX_MIB_TIMEWAITKILLED, killed);
474#endif
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700475 spin_unlock(&twdr->death_lock);
476}
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700477EXPORT_SYMBOL_GPL(inet_twdr_twcal_tick);
Daniel Lezcanod3154922008-09-08 13:17:27 -0700478
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000479void inet_twsk_purge(struct inet_hashinfo *hashinfo,
Daniel Lezcanod3154922008-09-08 13:17:27 -0700480 struct inet_timewait_death_row *twdr, int family)
481{
482 struct inet_timewait_sock *tw;
483 struct sock *sk;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800484 struct hlist_nulls_node *node;
Eric W. Biederman575f4cd2009-12-03 02:29:08 +0000485 unsigned int slot;
Daniel Lezcanod3154922008-09-08 13:17:27 -0700486
Eric W. Biederman575f4cd2009-12-03 02:29:08 +0000487 for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
488 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
489restart_rcu:
490 rcu_read_lock();
Daniel Lezcanod3154922008-09-08 13:17:27 -0700491restart:
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700492 sk_nulls_for_each_rcu(sk, node, &head->chain) {
493 if (sk->sk_state != TCP_TIME_WAIT)
494 continue;
Daniel Lezcanod3154922008-09-08 13:17:27 -0700495 tw = inet_twsk(sk);
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000496 if ((tw->tw_family != family) ||
497 atomic_read(&twsk_net(tw)->count))
Daniel Lezcanod3154922008-09-08 13:17:27 -0700498 continue;
499
Eric W. Biederman575f4cd2009-12-03 02:29:08 +0000500 if (unlikely(!atomic_inc_not_zero(&tw->tw_refcnt)))
501 continue;
502
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000503 if (unlikely((tw->tw_family != family) ||
504 atomic_read(&twsk_net(tw)->count))) {
Eric W. Biederman575f4cd2009-12-03 02:29:08 +0000505 inet_twsk_put(tw);
506 goto restart;
507 }
508
509 rcu_read_unlock();
Eric Dumazet91035f02011-02-18 22:35:56 +0000510 local_bh_disable();
Daniel Lezcanod3154922008-09-08 13:17:27 -0700511 inet_twsk_deschedule(tw, twdr);
Eric Dumazet91035f02011-02-18 22:35:56 +0000512 local_bh_enable();
Daniel Lezcanod3154922008-09-08 13:17:27 -0700513 inet_twsk_put(tw);
Eric W. Biederman575f4cd2009-12-03 02:29:08 +0000514 goto restart_rcu;
Daniel Lezcanod3154922008-09-08 13:17:27 -0700515 }
Eric W. Biederman575f4cd2009-12-03 02:29:08 +0000516 /* If the nulls value we got at the end of this lookup is
517 * not the expected one, we must restart lookup.
518 * We probably met an item that was moved to another chain.
519 */
520 if (get_nulls_value(node) != slot)
521 goto restart;
522 rcu_read_unlock();
Daniel Lezcanod3154922008-09-08 13:17:27 -0700523 }
Daniel Lezcanod3154922008-09-08 13:17:27 -0700524}
525EXPORT_SYMBOL_GPL(inet_twsk_purge);