Linux Audio

Check our new training course

Loading...
v4.17
 
  1/*
  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
 11#include <linux/kernel.h>
 12#include <linux/slab.h>
 13#include <linux/module.h>
 14#include <net/inet_hashtables.h>
 15#include <net/inet_timewait_sock.h>
 16#include <net/ip.h>
 17
 18
 19/**
 20 *	inet_twsk_bind_unhash - unhash a timewait socket from bind hash
 21 *	@tw: timewait socket
 22 *	@hashinfo: hashinfo pointer
 23 *
 24 *	unhash a timewait socket from bind hash, if hashed.
 25 *	bind hash lock must be held by caller.
 26 *	Returns 1 if caller should call inet_twsk_put() after lock release.
 27 */
 28void inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
 29			  struct inet_hashinfo *hashinfo)
 30{
 
 31	struct inet_bind_bucket *tb = tw->tw_tb;
 32
 33	if (!tb)
 34		return;
 35
 36	__hlist_del(&tw->tw_bind_node);
 37	tw->tw_tb = NULL;
 
 
 38	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
 
 39	__sock_put((struct sock *)tw);
 40}
 41
 42/* Must be called with locally disabled BHs. */
 43static void inet_twsk_kill(struct inet_timewait_sock *tw)
 44{
 45	struct inet_hashinfo *hashinfo = tw->tw_dr->hashinfo;
 46	spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash);
 47	struct inet_bind_hashbucket *bhead;
 48
 49	spin_lock(lock);
 50	sk_nulls_del_node_init_rcu((struct sock *)tw);
 51	spin_unlock(lock);
 52
 53	/* Disassociate with bind bucket. */
 54	bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num,
 55			hashinfo->bhash_size)];
 
 
 56
 57	spin_lock(&bhead->lock);
 
 58	inet_twsk_bind_unhash(tw, hashinfo);
 
 59	spin_unlock(&bhead->lock);
 60
 61	atomic_dec(&tw->tw_dr->tw_count);
 62	inet_twsk_put(tw);
 63}
 64
 65void inet_twsk_free(struct inet_timewait_sock *tw)
 66{
 67	struct module *owner = tw->tw_prot->owner;
 68	twsk_destructor((struct sock *)tw);
 69#ifdef SOCK_REFCNT_DEBUG
 70	pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw);
 71#endif
 72	kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
 73	module_put(owner);
 74}
 75
 76void inet_twsk_put(struct inet_timewait_sock *tw)
 77{
 78	if (refcount_dec_and_test(&tw->tw_refcnt))
 79		inet_twsk_free(tw);
 80}
 81EXPORT_SYMBOL_GPL(inet_twsk_put);
 82
 83static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
 84				   struct hlist_nulls_head *list)
 85{
 86	hlist_nulls_add_head_rcu(&tw->tw_node, list);
 87}
 88
 89static void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
 90				    struct hlist_head *list)
 91{
 92	hlist_add_head(&tw->tw_bind_node, list);
 93}
 94
 95/*
 96 * Enter the time wait state. This is called with locally disabled BH.
 97 * Essentially we whip up a timewait bucket, copy the relevant info into it
 98 * from the SK, and mess with hash chains and list linkage.
 
 
 99 */
100void inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
101			   struct inet_hashinfo *hashinfo)
 
 
102{
103	const struct inet_sock *inet = inet_sk(sk);
104	const struct inet_connection_sock *icsk = inet_csk(sk);
105	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
106	spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
107	struct inet_bind_hashbucket *bhead;
 
108	/* Step 1: Put TW into bind hash. Original socket stays there too.
109	   Note, that any socket with inet->num != 0 MUST be bound in
110	   binding cache, even if it is closed.
111	 */
112	bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num,
113			hashinfo->bhash_size)];
 
 
 
114	spin_lock(&bhead->lock);
 
 
115	tw->tw_tb = icsk->icsk_bind_hash;
116	WARN_ON(!icsk->icsk_bind_hash);
117	inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
 
 
 
 
 
118	spin_unlock(&bhead->lock);
119
120	spin_lock(lock);
121
 
122	inet_twsk_add_node_rcu(tw, &ehead->chain);
123
124	/* Step 3: Remove SK from hash chain */
125	if (__sk_nulls_del_node_init_rcu(sk))
126		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
127
128	spin_unlock(lock);
129
 
 
 
 
 
130	/* tw_refcnt is set to 3 because we have :
131	 * - one reference for bhash chain.
132	 * - one reference for ehash chain.
133	 * - one reference for timer.
134	 * We can use atomic_set() because prior spin_lock()/spin_unlock()
135	 * committed into memory all tw fields.
136	 * Also note that after this point, we lost our implicit reference
137	 * so we are not allowed to use tw anymore.
138	 */
139	refcount_set(&tw->tw_refcnt, 3);
 
 
 
 
 
140}
141EXPORT_SYMBOL_GPL(inet_twsk_hashdance);
142
143static void tw_timer_handler(struct timer_list *t)
144{
145	struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer);
146
147	if (tw->tw_kill)
148		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED);
149	else
150		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_TIMEWAITED);
151	inet_twsk_kill(tw);
152}
153
154struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
155					   struct inet_timewait_death_row *dr,
156					   const int state)
157{
158	struct inet_timewait_sock *tw;
159
160	if (atomic_read(&dr->tw_count) >= dr->sysctl_max_tw_buckets)
 
161		return NULL;
162
163	tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
164			      GFP_ATOMIC);
165	if (tw) {
166		const struct inet_sock *inet = inet_sk(sk);
167
168		tw->tw_dr	    = dr;
169		/* Give us an identity. */
170		tw->tw_daddr	    = inet->inet_daddr;
171		tw->tw_rcv_saddr    = inet->inet_rcv_saddr;
172		tw->tw_bound_dev_if = sk->sk_bound_dev_if;
173		tw->tw_tos	    = inet->tos;
174		tw->tw_num	    = inet->inet_num;
175		tw->tw_state	    = TCP_TIME_WAIT;
176		tw->tw_substate	    = state;
177		tw->tw_sport	    = inet->inet_sport;
178		tw->tw_dport	    = inet->inet_dport;
179		tw->tw_family	    = sk->sk_family;
180		tw->tw_reuse	    = sk->sk_reuse;
181		tw->tw_reuseport    = sk->sk_reuseport;
182		tw->tw_hash	    = sk->sk_hash;
183		tw->tw_ipv6only	    = 0;
184		tw->tw_transparent  = inet->transparent;
185		tw->tw_prot	    = sk->sk_prot_creator;
186		atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
187		twsk_net_set(tw, sock_net(sk));
188		timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED);
189		/*
190		 * Because we use RCU lookups, we should not set tw_refcnt
191		 * to a non null value before everything is setup for this
192		 * timewait socket.
193		 */
194		refcount_set(&tw->tw_refcnt, 0);
195
196		__module_get(tw->tw_prot->owner);
197	}
198
199	return tw;
200}
201EXPORT_SYMBOL_GPL(inet_twsk_alloc);
202
203/* These are always called from BH context.  See callers in
204 * tcp_input.c to verify this.
205 */
206
207/* This is for handling early-kills of TIME_WAIT sockets.
208 * Warning : consume reference.
209 * Caller should not access tw anymore.
210 */
211void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
212{
213	if (del_timer_sync(&tw->tw_timer))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
214		inet_twsk_kill(tw);
215	inet_twsk_put(tw);
216}
217EXPORT_SYMBOL(inet_twsk_deschedule_put);
218
219void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
220{
221	/* timeout := RTO * 3.5
222	 *
223	 * 3.5 = 1+2+0.5 to wait for two retransmits.
224	 *
225	 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
226	 * our ACK acking that FIN can be lost. If N subsequent retransmitted
227	 * FINs (or previous seqments) are lost (probability of such event
228	 * is p^(N+1), where p is probability to lose single packet and
229	 * time to detect the loss is about RTO*(2^N - 1) with exponential
230	 * backoff). Normal timewait length is calculated so, that we
231	 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
232	 * [ BTW Linux. following BSD, violates this requirement waiting
233	 *   only for 60sec, we should wait at least for 240 secs.
234	 *   Well, 240 consumes too much of resources 8)
235	 * ]
236	 * This interval is not reduced to catch old duplicate and
237	 * responces to our wandering segments living for two MSLs.
238	 * However, if we use PAWS to detect
239	 * old duplicates, we can reduce the interval to bounds required
240	 * by RTO, rather than MSL. So, if peer understands PAWS, we
241	 * kill tw bucket after 3.5*RTO (it is important that this number
242	 * is greater than TS tick!) and detect old duplicates with help
243	 * of PAWS.
244	 */
245
246	tw->tw_kill = timeo <= 4*HZ;
247	if (!rearm) {
 
 
 
 
248		BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
249		atomic_inc(&tw->tw_dr->tw_count);
250	} else {
251		mod_timer_pending(&tw->tw_timer, jiffies + timeo);
252	}
253}
254EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
255
256void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
 
257{
258	struct inet_timewait_sock *tw;
259	struct sock *sk;
260	struct hlist_nulls_node *node;
261	unsigned int slot;
 
 
 
 
 
262
263	for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
264		struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
265restart_rcu:
266		cond_resched();
267		rcu_read_lock();
268restart:
269		sk_nulls_for_each_rcu(sk, node, &head->chain) {
270			if (sk->sk_state != TCP_TIME_WAIT)
 
 
 
271				continue;
272			tw = inet_twsk(sk);
273			if ((tw->tw_family != family) ||
274				refcount_read(&twsk_net(tw)->count))
275				continue;
276
277			if (unlikely(!refcount_inc_not_zero(&tw->tw_refcnt)))
278				continue;
279
280			if (unlikely((tw->tw_family != family) ||
281				     refcount_read(&twsk_net(tw)->count))) {
282				inet_twsk_put(tw);
283				goto restart;
284			}
285
286			rcu_read_unlock();
287			local_bh_disable();
288			inet_twsk_deschedule_put(tw);
 
 
 
 
 
 
 
289			local_bh_enable();
290			goto restart_rcu;
291		}
292		/* If the nulls value we got at the end of this lookup is
293		 * not the expected one, we must restart lookup.
294		 * We probably met an item that was moved to another chain.
295		 */
296		if (get_nulls_value(node) != slot)
297			goto restart;
298		rcu_read_unlock();
299	}
300}
301EXPORT_SYMBOL_GPL(inet_twsk_purge);
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
  4 *		operating system.  INET is implemented using the  BSD Socket
  5 *		interface as the means of communication with the user level.
  6 *
  7 *		Generic TIME_WAIT sockets functions
  8 *
  9 *		From code orinally in TCP
 10 */
 11
 12#include <linux/kernel.h>
 13#include <linux/slab.h>
 14#include <linux/module.h>
 15#include <net/inet_hashtables.h>
 16#include <net/inet_timewait_sock.h>
 17#include <net/ip.h>
 18
 19
 20/**
 21 *	inet_twsk_bind_unhash - unhash a timewait socket from bind hash
 22 *	@tw: timewait socket
 23 *	@hashinfo: hashinfo pointer
 24 *
 25 *	unhash a timewait socket from bind hash, if hashed.
 26 *	bind hash lock must be held by caller.
 27 *	Returns 1 if caller should call inet_twsk_put() after lock release.
 28 */
 29void inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
 30			  struct inet_hashinfo *hashinfo)
 31{
 32	struct inet_bind2_bucket *tb2 = tw->tw_tb2;
 33	struct inet_bind_bucket *tb = tw->tw_tb;
 34
 35	if (!tb)
 36		return;
 37
 38	__sk_del_bind_node((struct sock *)tw);
 39	tw->tw_tb = NULL;
 40	tw->tw_tb2 = NULL;
 41	inet_bind2_bucket_destroy(hashinfo->bind2_bucket_cachep, tb2);
 42	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
 43
 44	__sock_put((struct sock *)tw);
 45}
 46
 47/* Must be called with locally disabled BHs. */
 48static void inet_twsk_kill(struct inet_timewait_sock *tw)
 49{
 50	struct inet_hashinfo *hashinfo = tw->tw_dr->hashinfo;
 51	spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash);
 52	struct inet_bind_hashbucket *bhead, *bhead2;
 53
 54	spin_lock(lock);
 55	sk_nulls_del_node_init_rcu((struct sock *)tw);
 56	spin_unlock(lock);
 57
 58	/* Disassociate with bind bucket. */
 59	bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num,
 60			hashinfo->bhash_size)];
 61	bhead2 = inet_bhashfn_portaddr(hashinfo, (struct sock *)tw,
 62				       twsk_net(tw), tw->tw_num);
 63
 64	spin_lock(&bhead->lock);
 65	spin_lock(&bhead2->lock);
 66	inet_twsk_bind_unhash(tw, hashinfo);
 67	spin_unlock(&bhead2->lock);
 68	spin_unlock(&bhead->lock);
 69
 70	refcount_dec(&tw->tw_dr->tw_refcount);
 71	inet_twsk_put(tw);
 72}
 73
 74void inet_twsk_free(struct inet_timewait_sock *tw)
 75{
 76	struct module *owner = tw->tw_prot->owner;
 77	twsk_destructor((struct sock *)tw);
 
 
 
 78	kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
 79	module_put(owner);
 80}
 81
 82void inet_twsk_put(struct inet_timewait_sock *tw)
 83{
 84	if (refcount_dec_and_test(&tw->tw_refcnt))
 85		inet_twsk_free(tw);
 86}
 87EXPORT_SYMBOL_GPL(inet_twsk_put);
 88
 89static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
 90				   struct hlist_nulls_head *list)
 91{
 92	hlist_nulls_add_head_rcu(&tw->tw_node, list);
 93}
 94
 95static void inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo)
 
 96{
 97	__inet_twsk_schedule(tw, timeo, false);
 98}
 99
100/*
101 * Enter the time wait state.
102 * Essentially we whip up a timewait bucket, copy the relevant info into it
103 * from the SK, and mess with hash chains and list linkage.
104 *
105 * The caller must not access @tw anymore after this function returns.
106 */
107void inet_twsk_hashdance_schedule(struct inet_timewait_sock *tw,
108				  struct sock *sk,
109				  struct inet_hashinfo *hashinfo,
110				  int timeo)
111{
112	const struct inet_sock *inet = inet_sk(sk);
113	const struct inet_connection_sock *icsk = inet_csk(sk);
114	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
115	spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
116	struct inet_bind_hashbucket *bhead, *bhead2;
117
118	/* Step 1: Put TW into bind hash. Original socket stays there too.
119	   Note, that any socket with inet->num != 0 MUST be bound in
120	   binding cache, even if it is closed.
121	 */
122	bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num,
123			hashinfo->bhash_size)];
124	bhead2 = inet_bhashfn_portaddr(hashinfo, sk, twsk_net(tw), inet->inet_num);
125
126	local_bh_disable();
127	spin_lock(&bhead->lock);
128	spin_lock(&bhead2->lock);
129
130	tw->tw_tb = icsk->icsk_bind_hash;
131	WARN_ON(!icsk->icsk_bind_hash);
132
133	tw->tw_tb2 = icsk->icsk_bind2_hash;
134	WARN_ON(!icsk->icsk_bind2_hash);
135	sk_add_bind_node((struct sock *)tw, &tw->tw_tb2->owners);
136
137	spin_unlock(&bhead2->lock);
138	spin_unlock(&bhead->lock);
139
140	spin_lock(lock);
141
142	/* Step 2: Hash TW into tcp ehash chain */
143	inet_twsk_add_node_rcu(tw, &ehead->chain);
144
145	/* Step 3: Remove SK from hash chain */
146	if (__sk_nulls_del_node_init_rcu(sk))
147		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
148
 
149
150	/* Ensure above writes are committed into memory before updating the
151	 * refcount.
152	 * Provides ordering vs later refcount_inc().
153	 */
154	smp_wmb();
155	/* tw_refcnt is set to 3 because we have :
156	 * - one reference for bhash chain.
157	 * - one reference for ehash chain.
158	 * - one reference for timer.
 
 
159	 * Also note that after this point, we lost our implicit reference
160	 * so we are not allowed to use tw anymore.
161	 */
162	refcount_set(&tw->tw_refcnt, 3);
163
164	inet_twsk_schedule(tw, timeo);
165
166	spin_unlock(lock);
167	local_bh_enable();
168}
169EXPORT_SYMBOL_GPL(inet_twsk_hashdance_schedule);
170
171static void tw_timer_handler(struct timer_list *t)
172{
173	struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer);
174
 
 
 
 
175	inet_twsk_kill(tw);
176}
177
178struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
179					   struct inet_timewait_death_row *dr,
180					   const int state)
181{
182	struct inet_timewait_sock *tw;
183
184	if (refcount_read(&dr->tw_refcount) - 1 >=
185	    READ_ONCE(dr->sysctl_max_tw_buckets))
186		return NULL;
187
188	tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
189			      GFP_ATOMIC);
190	if (tw) {
191		const struct inet_sock *inet = inet_sk(sk);
192
193		tw->tw_dr	    = dr;
194		/* Give us an identity. */
195		tw->tw_daddr	    = inet->inet_daddr;
196		tw->tw_rcv_saddr    = inet->inet_rcv_saddr;
197		tw->tw_bound_dev_if = sk->sk_bound_dev_if;
198		tw->tw_tos	    = inet->tos;
199		tw->tw_num	    = inet->inet_num;
200		tw->tw_state	    = TCP_TIME_WAIT;
201		tw->tw_substate	    = state;
202		tw->tw_sport	    = inet->inet_sport;
203		tw->tw_dport	    = inet->inet_dport;
204		tw->tw_family	    = sk->sk_family;
205		tw->tw_reuse	    = sk->sk_reuse;
206		tw->tw_reuseport    = sk->sk_reuseport;
207		tw->tw_hash	    = sk->sk_hash;
208		tw->tw_ipv6only	    = 0;
209		tw->tw_transparent  = inet_test_bit(TRANSPARENT, sk);
210		tw->tw_prot	    = sk->sk_prot_creator;
211		atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
212		twsk_net_set(tw, sock_net(sk));
213		timer_setup(&tw->tw_timer, tw_timer_handler, 0);
214		/*
215		 * Because we use RCU lookups, we should not set tw_refcnt
216		 * to a non null value before everything is setup for this
217		 * timewait socket.
218		 */
219		refcount_set(&tw->tw_refcnt, 0);
220
221		__module_get(tw->tw_prot->owner);
222	}
223
224	return tw;
225}
226EXPORT_SYMBOL_GPL(inet_twsk_alloc);
227
228/* These are always called from BH context.  See callers in
229 * tcp_input.c to verify this.
230 */
231
232/* This is for handling early-kills of TIME_WAIT sockets.
233 * Warning : consume reference.
234 * Caller should not access tw anymore.
235 */
236void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
237{
238	struct inet_hashinfo *hashinfo = tw->tw_dr->hashinfo;
239	spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash);
240
241	/* inet_twsk_purge() walks over all sockets, including tw ones,
242	 * and removes them via inet_twsk_deschedule_put() after a
243	 * refcount_inc_not_zero().
244	 *
245	 * inet_twsk_hashdance_schedule() must (re)init the refcount before
246	 * arming the timer, i.e. inet_twsk_purge can obtain a reference to
247	 * a twsk that did not yet schedule the timer.
248	 *
249	 * The ehash lock synchronizes these two:
250	 * After acquiring the lock, the timer is always scheduled (else
251	 * timer_shutdown returns false), because hashdance_schedule releases
252	 * the ehash lock only after completing the timer initialization.
253	 *
254	 * Without grabbing the ehash lock, we get:
255	 * 1) cpu x sets twsk refcount to 3
256	 * 2) cpu y bumps refcount to 4
257	 * 3) cpu y calls inet_twsk_deschedule_put() and shuts timer down
258	 * 4) cpu x tries to start timer, but mod_timer is a noop post-shutdown
259	 * -> timer refcount is never decremented.
260	 */
261	spin_lock(lock);
262	/*  Makes sure hashdance_schedule() has completed */
263	spin_unlock(lock);
264
265	if (timer_shutdown_sync(&tw->tw_timer))
266		inet_twsk_kill(tw);
267	inet_twsk_put(tw);
268}
269EXPORT_SYMBOL(inet_twsk_deschedule_put);
270
271void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
272{
273	/* timeout := RTO * 3.5
274	 *
275	 * 3.5 = 1+2+0.5 to wait for two retransmits.
276	 *
277	 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
278	 * our ACK acking that FIN can be lost. If N subsequent retransmitted
279	 * FINs (or previous seqments) are lost (probability of such event
280	 * is p^(N+1), where p is probability to lose single packet and
281	 * time to detect the loss is about RTO*(2^N - 1) with exponential
282	 * backoff). Normal timewait length is calculated so, that we
283	 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
284	 * [ BTW Linux. following BSD, violates this requirement waiting
285	 *   only for 60sec, we should wait at least for 240 secs.
286	 *   Well, 240 consumes too much of resources 8)
287	 * ]
288	 * This interval is not reduced to catch old duplicate and
289	 * responces to our wandering segments living for two MSLs.
290	 * However, if we use PAWS to detect
291	 * old duplicates, we can reduce the interval to bounds required
292	 * by RTO, rather than MSL. So, if peer understands PAWS, we
293	 * kill tw bucket after 3.5*RTO (it is important that this number
294	 * is greater than TS tick!) and detect old duplicates with help
295	 * of PAWS.
296	 */
297
 
298	if (!rearm) {
299		bool kill = timeo <= 4*HZ;
300
301		__NET_INC_STATS(twsk_net(tw), kill ? LINUX_MIB_TIMEWAITKILLED :
302						     LINUX_MIB_TIMEWAITED);
303		BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
304		refcount_inc(&tw->tw_dr->tw_refcount);
305	} else {
306		mod_timer_pending(&tw->tw_timer, jiffies + timeo);
307	}
308}
309EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
310
311/* Remove all non full sockets (TIME_WAIT and NEW_SYN_RECV) for dead netns */
312void inet_twsk_purge(struct inet_hashinfo *hashinfo)
313{
314	struct inet_ehash_bucket *head = &hashinfo->ehash[0];
315	unsigned int ehash_mask = hashinfo->ehash_mask;
316	struct hlist_nulls_node *node;
317	unsigned int slot;
318	struct sock *sk;
319
320	for (slot = 0; slot <= ehash_mask; slot++, head++) {
321		if (hlist_nulls_empty(&head->chain))
322			continue;
323
 
 
324restart_rcu:
325		cond_resched();
326		rcu_read_lock();
327restart:
328		sk_nulls_for_each_rcu(sk, node, &head->chain) {
329			int state = inet_sk_state_load(sk);
330
331			if ((1 << state) & ~(TCPF_TIME_WAIT |
332					     TCPF_NEW_SYN_RECV))
333				continue;
334
335			if (refcount_read(&sock_net(sk)->ns.count))
 
336				continue;
337
338			if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
339				continue;
340
341			if (refcount_read(&sock_net(sk)->ns.count)) {
342				sock_gen_put(sk);
 
343				goto restart;
344			}
345
346			rcu_read_unlock();
347			local_bh_disable();
348			if (state == TCP_TIME_WAIT) {
349				inet_twsk_deschedule_put(inet_twsk(sk));
350			} else {
351				struct request_sock *req = inet_reqsk(sk);
352
353				inet_csk_reqsk_queue_drop_and_put(req->rsk_listener,
354								  req);
355			}
356			local_bh_enable();
357			goto restart_rcu;
358		}
359		/* If the nulls value we got at the end of this lookup is
360		 * not the expected one, we must restart lookup.
361		 * We probably met an item that was moved to another chain.
362		 */
363		if (get_nulls_value(node) != slot)
364			goto restart;
365		rcu_read_unlock();
366	}
367}
368EXPORT_SYMBOL_GPL(inet_twsk_purge);