[net-next-2.6.git] / net / ipv4 / inet_timewait_sock.c

/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		Generic TIME_WAIT sockets functions
 *
 *		From code orinally in TCP
 */


#include <net/inet_hashtables.h>
#include <net/inet_timewait_sock.h>
#include <net/ip.h>

/* Must be called with locally disabled BHs. */
void __inet_twsk_kill(struct inet_timewait_sock *tw, struct inet_hashinfo *hashinfo)
{
	struct inet_bind_hashbucket *bhead;
	struct inet_bind_bucket *tb;
	/* Unlink from established hashes. */
	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, tw->tw_hash);

	write_lock(&ehead->lock);
	if (hlist_unhashed(&tw->tw_node)) {
		write_unlock(&ehead->lock);
		return;
	}
	__hlist_del(&tw->tw_node);
	sk_node_init(&tw->tw_node);
	write_unlock(&ehead->lock);

	/* Disassociate with bind bucket. */
	bhead = &hashinfo->bhash[inet_bhashfn(tw->tw_num, hashinfo->bhash_size)];
	spin_lock(&bhead->lock);
	tb = tw->tw_tb;
	__hlist_del(&tw->tw_bind_node);
	tw->tw_tb = NULL;
	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
	spin_unlock(&bhead->lock);
#ifdef SOCK_REFCNT_DEBUG
	if (atomic_read(&tw->tw_refcnt) != 1) {
		printk(KERN_DEBUG "%s timewait_sock %p refcnt=%d\n",
		       tw->tw_prot->name, tw, atomic_read(&tw->tw_refcnt));
	}
#endif
	inet_twsk_put(tw);
}

EXPORT_SYMBOL_GPL(__inet_twsk_kill);

/*
 * Enter the time wait state. This is called with locally disabled BH.
 * Essentially we whip up a timewait bucket, copy the relevant info into it
 * from the SK, and mess with hash chains and list linkage.
 */
void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
			   struct inet_hashinfo *hashinfo)
{
	const struct inet_sock *inet = inet_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
	struct inet_bind_hashbucket *bhead;
	/* Step 1: Put TW into bind hash. Original socket stays there too.
	   Note, that any socket with inet->num != 0 MUST be bound in
	   binding cache, even if it is closed.
	 */
	bhead = &hashinfo->bhash[inet_bhashfn(inet->num, hashinfo->bhash_size)];
	spin_lock(&bhead->lock);
	tw->tw_tb = icsk->icsk_bind_hash;
	BUG_TRAP(icsk->icsk_bind_hash);
	inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
	spin_unlock(&bhead->lock);

	write_lock(&ehead->lock);

	/* Step 2: Remove SK from established hash. */
	if (__sk_del_node_init(sk))
		sock_prot_dec_use(sk->sk_prot);

	/* Step 3: Hash TW into TIMEWAIT half of established hash table. */
	inet_twsk_add_node(tw, &(ehead + hashinfo->ehash_size)->chain);
	atomic_inc(&tw->tw_refcnt);

	write_unlock(&ehead->lock);
}

EXPORT_SYMBOL_GPL(__inet_twsk_hashdance);

struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, const int state)
{
	struct inet_timewait_sock *tw =
		kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
				 SLAB_ATOMIC);
	if (tw != NULL) {
		const struct inet_sock *inet = inet_sk(sk);

		/* Give us an identity. */
		tw->tw_daddr	    = inet->daddr;
		tw->tw_rcv_saddr    = inet->rcv_saddr;
		tw->tw_bound_dev_if = sk->sk_bound_dev_if;
		tw->tw_num	    = inet->num;
		tw->tw_state	    = TCP_TIME_WAIT;
		tw->tw_substate	    = state;
		tw->tw_sport	    = inet->sport;
		tw->tw_dport	    = inet->dport;
		tw->tw_family	    = sk->sk_family;
		tw->tw_reuse	    = sk->sk_reuse;
		tw->tw_hash	    = sk->sk_hash;
		tw->tw_ipv6only	    = 0;
		tw->tw_prot	    = sk->sk_prot_creator;
		atomic_set(&tw->tw_refcnt, 1);
		inet_twsk_dead_node_init(tw);
		__module_get(tw->tw_prot->owner);
	}

	return tw;
}

EXPORT_SYMBOL_GPL(inet_twsk_alloc);

/* Returns non-zero if quota exceeded.  */
static int inet_twdr_do_twkill_work(struct inet_timewait_death_row *twdr,
				    const int slot)
{
	struct inet_timewait_sock *tw;
	struct hlist_node *node;
	unsigned int killed;
	int ret;

	/* NOTE: compare this to previous version where lock
	 * was released after detaching chain. It was racy,
	 * because tw buckets are scheduled in not serialized context
	 * in 2.3 (with netfilter), and with softnet it is common, because
	 * soft irqs are not sequenced.
	 */
	killed = 0;
	ret = 0;
rescan:
	inet_twsk_for_each_inmate(tw, node, &twdr->cells[slot]) {
		__inet_twsk_del_dead_node(tw);
		spin_unlock(&twdr->death_lock);
		__inet_twsk_kill(tw, twdr->hashinfo);
		inet_twsk_put(tw);
		killed++;
		spin_lock(&twdr->death_lock);
		if (killed > INET_TWDR_TWKILL_QUOTA) {
			ret = 1;
			break;
		}

		/* While we dropped twdr->death_lock, another cpu may have
		 * killed off the next TW bucket in the list, therefore
		 * do a fresh re-read of the hlist head node with the
		 * lock reacquired.  We still use the hlist traversal
		 * macro in order to get the prefetches.
		 */
		goto rescan;
	}

	twdr->tw_count -= killed;
	NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITED, killed);

	return ret;
}

void inet_twdr_hangman(unsigned long data)
{
	struct inet_timewait_death_row *twdr;
	int unsigned need_timer;

	twdr = (struct inet_timewait_death_row *)data;
	spin_lock(&twdr->death_lock);

	if (twdr->tw_count == 0)
		goto out;

	need_timer = 0;
	if (inet_twdr_do_twkill_work(twdr, twdr->slot)) {
		twdr->thread_slots |= (1 << twdr->slot);
		mb();
		schedule_work(&twdr->twkill_work);
		need_timer = 1;
	} else {
		/* We purged the entire slot, anything left?  */
		if (twdr->tw_count)
			need_timer = 1;
	}
	twdr->slot = ((twdr->slot + 1) & (INET_TWDR_TWKILL_SLOTS - 1));
	if (need_timer)
		mod_timer(&twdr->tw_timer, jiffies + twdr->period);
out:
	spin_unlock(&twdr->death_lock);
}

EXPORT_SYMBOL_GPL(inet_twdr_hangman);

extern void twkill_slots_invalid(void);

void inet_twdr_twkill_work(struct work_struct *work)
{
	struct inet_timewait_death_row *twdr =
		container_of(work, struct inet_timewait_death_row, twkill_work);
	int i;

	if ((INET_TWDR_TWKILL_SLOTS - 1) > (sizeof(twdr->thread_slots) * 8))
		twkill_slots_invalid();

	while (twdr->thread_slots) {
		spin_lock_bh(&twdr->death_lock);
		for (i = 0; i < INET_TWDR_TWKILL_SLOTS; i++) {
			if (!(twdr->thread_slots & (1 << i)))
				continue;

			while (inet_twdr_do_twkill_work(twdr, i) != 0) {
				if (need_resched()) {
					spin_unlock_bh(&twdr->death_lock);
					schedule();
					spin_lock_bh(&twdr->death_lock);
				}
			}

			twdr->thread_slots &= ~(1 << i);
		}
		spin_unlock_bh(&twdr->death_lock);
	}
}

EXPORT_SYMBOL_GPL(inet_twdr_twkill_work);

/* These are always called from BH context.  See callers in
 * tcp_input.c to verify this.
 */

/* This is for handling early-kills of TIME_WAIT sockets. */
void inet_twsk_deschedule(struct inet_timewait_sock *tw,
			  struct inet_timewait_death_row *twdr)
{
	spin_lock(&twdr->death_lock);
	if (inet_twsk_del_dead_node(tw)) {
		inet_twsk_put(tw);
		if (--twdr->tw_count == 0)
			del_timer(&twdr->tw_timer);
	}
	spin_unlock(&twdr->death_lock);
	__inet_twsk_kill(tw, twdr->hashinfo);
}

EXPORT_SYMBOL(inet_twsk_deschedule);

void inet_twsk_schedule(struct inet_timewait_sock *tw,
		       struct inet_timewait_death_row *twdr,
		       const int timeo, const int timewait_len)
{
	struct hlist_head *list;
	int slot;

	/* timeout := RTO * 3.5
	 *
	 * 3.5 = 1+2+0.5 to wait for two retransmits.
	 *
	 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
	 * our ACK acking that FIN can be lost. If N subsequent retransmitted
	 * FINs (or previous seqments) are lost (probability of such event
	 * is p^(N+1), where p is probability to lose single packet and
	 * time to detect the loss is about RTO*(2^N - 1) with exponential
	 * backoff). Normal timewait length is calculated so, that we
	 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
	 * [ BTW Linux. following BSD, violates this requirement waiting
	 *   only for 60sec, we should wait at least for 240 secs.
	 *   Well, 240 consumes too much of resources 8)
	 * ]
	 * This interval is not reduced to catch old duplicate and
	 * responces to our wandering segments living for two MSLs.
	 * However, if we use PAWS to detect
	 * old duplicates, we can reduce the interval to bounds required
	 * by RTO, rather than MSL. So, if peer understands PAWS, we
	 * kill tw bucket after 3.5*RTO (it is important that this number
	 * is greater than TS tick!) and detect old duplicates with help
	 * of PAWS.
	 */
	slot = (timeo + (1 << INET_TWDR_RECYCLE_TICK) - 1) >> INET_TWDR_RECYCLE_TICK;

	spin_lock(&twdr->death_lock);

	/* Unlink it, if it was scheduled */
	if (inet_twsk_del_dead_node(tw))
		twdr->tw_count--;
	else
		atomic_inc(&tw->tw_refcnt);

	if (slot >= INET_TWDR_RECYCLE_SLOTS) {
		/* Schedule to slow timer */
		if (timeo >= timewait_len) {
			slot = INET_TWDR_TWKILL_SLOTS - 1;
		} else {
			slot = (timeo + twdr->period - 1) / twdr->period;
			if (slot >= INET_TWDR_TWKILL_SLOTS)
				slot = INET_TWDR_TWKILL_SLOTS - 1;
		}
		tw->tw_ttd = jiffies + timeo;
		slot = (twdr->slot + slot) & (INET_TWDR_TWKILL_SLOTS - 1);
		list = &twdr->cells[slot];
	} else {
		tw->tw_ttd = jiffies + (slot << INET_TWDR_RECYCLE_TICK);

		if (twdr->twcal_hand < 0) {
			twdr->twcal_hand = 0;
			twdr->twcal_jiffie = jiffies;
			twdr->twcal_timer.expires = twdr->twcal_jiffie +
					      (slot << INET_TWDR_RECYCLE_TICK);
			add_timer(&twdr->twcal_timer);
		} else {
			if (time_after(twdr->twcal_timer.expires,
				       jiffies + (slot << INET_TWDR_RECYCLE_TICK)))
				mod_timer(&twdr->twcal_timer,
					  jiffies + (slot << INET_TWDR_RECYCLE_TICK));
			slot = (twdr->twcal_hand + slot) & (INET_TWDR_RECYCLE_SLOTS - 1);
		}
		list = &twdr->twcal_row[slot];
	}

	hlist_add_head(&tw->tw_death_node, list);

	if (twdr->tw_count++ == 0)
		mod_timer(&twdr->tw_timer, jiffies + twdr->period);
	spin_unlock(&twdr->death_lock);
}

EXPORT_SYMBOL_GPL(inet_twsk_schedule);

void inet_twdr_twcal_tick(unsigned long data)
{
	struct inet_timewait_death_row *twdr;
	int n, slot;
	unsigned long j;
	unsigned long now = jiffies;
	int killed = 0;
	int adv = 0;

	twdr = (struct inet_timewait_death_row *)data;

	spin_lock(&twdr->death_lock);
	if (twdr->twcal_hand < 0)
		goto out;

	slot = twdr->twcal_hand;
	j = twdr->twcal_jiffie;

	for (n = 0; n < INET_TWDR_RECYCLE_SLOTS; n++) {
		if (time_before_eq(j, now)) {
			struct hlist_node *node, *safe;
			struct inet_timewait_sock *tw;

			inet_twsk_for_each_inmate_safe(tw, node, safe,
						       &twdr->twcal_row[slot]) {
				__inet_twsk_del_dead_node(tw);
				__inet_twsk_kill(tw, twdr->hashinfo);
				inet_twsk_put(tw);
				killed++;
			}
		} else {
			if (!adv) {
				adv = 1;
				twdr->twcal_jiffie = j;
				twdr->twcal_hand = slot;
			}

			if (!hlist_empty(&twdr->twcal_row[slot])) {
				mod_timer(&twdr->twcal_timer, j);
				goto out;
			}
		}
		j += 1 << INET_TWDR_RECYCLE_TICK;
		slot = (slot + 1) & (INET_TWDR_RECYCLE_SLOTS - 1);
	}
	twdr->twcal_hand = -1;

out:
	if ((twdr->tw_count -= killed) == 0)
		del_timer(&twdr->tw_timer);
	NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITKILLED, killed);
	spin_unlock(&twdr->death_lock);
}

EXPORT_SYMBOL_GPL(inet_twdr_twcal_tick);
Commit	Line	Data
e48c414e ACM	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
e48c414e ACM	11
	12	#include <net/inet_hashtables.h>
	13	#include <net/inet_timewait_sock.h>
696ab2d3	14	#include <net/ip.h>
e48c414e ACM	15
	16	/* Must be called with locally disabled BHs. */
	17	void __inet_twsk_kill(struct inet_timewait_sock tw, struct inet_hashinfo hashinfo)
	18	{
	19	struct inet_bind_hashbucket *bhead;
	20	struct inet_bind_bucket *tb;
	21	/* Unlink from established hashes. */
81c3d547	22	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, tw->tw_hash);
e48c414e ACM	23
	24	write_lock(&ehead->lock);
	25	if (hlist_unhashed(&tw->tw_node)) {
	26	write_unlock(&ehead->lock);
	27	return;
	28	}
	29	__hlist_del(&tw->tw_node);
	30	sk_node_init(&tw->tw_node);
	31	write_unlock(&ehead->lock);
	32
	33	/* Disassociate with bind bucket. */
	34	bhead = &hashinfo->bhash[inet_bhashfn(tw->tw_num, hashinfo->bhash_size)];
	35	spin_lock(&bhead->lock);
	36	tb = tw->tw_tb;
	37	__hlist_del(&tw->tw_bind_node);
	38	tw->tw_tb = NULL;
	39	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
	40	spin_unlock(&bhead->lock);
	41	#ifdef SOCK_REFCNT_DEBUG
	42	if (atomic_read(&tw->tw_refcnt) != 1) {
	43	printk(KERN_DEBUG "%s timewait_sock %p refcnt=%d\n",
	44	tw->tw_prot->name, tw, atomic_read(&tw->tw_refcnt));
	45	}
	46	#endif
	47	inet_twsk_put(tw);
	48	}
	49
d8c97a94 ACM	50	EXPORT_SYMBOL_GPL(__inet_twsk_kill);
d8c97a94 ACM	51
e48c414e ACM	52	/*
	53	* Enter the time wait state. This is called with locally disabled BH.
	54	* Essentially we whip up a timewait bucket, copy the relevant info into it
	55	* from the SK, and mess with hash chains and list linkage.
	56	*/
	57	void __inet_twsk_hashdance(struct inet_timewait_sock tw, struct sock sk,
	58	struct inet_hashinfo *hashinfo)
	59	{
	60	const struct inet_sock *inet = inet_sk(sk);
463c84b9	61	const struct inet_connection_sock *icsk = inet_csk(sk);
81c3d547	62	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
e48c414e ACM	63	struct inet_bind_hashbucket *bhead;
	64	/* Step 1: Put TW into bind hash. Original socket stays there too.
	65	Note, that any socket with inet->num != 0 MUST be bound in
	66	binding cache, even if it is closed.
	67	*/
	68	bhead = &hashinfo->bhash[inet_bhashfn(inet->num, hashinfo->bhash_size)];
	69	spin_lock(&bhead->lock);
463c84b9 ACM	70	tw->tw_tb = icsk->icsk_bind_hash;
463c84b9 ACM	71	BUG_TRAP(icsk->icsk_bind_hash);
e48c414e ACM	72	inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
	73	spin_unlock(&bhead->lock);
	74
	75	write_lock(&ehead->lock);
	76
	77	/* Step 2: Remove SK from established hash. */
	78	if (__sk_del_node_init(sk))
	79	sock_prot_dec_use(sk->sk_prot);
	80
	81	/* Step 3: Hash TW into TIMEWAIT half of established hash table. */
	82	inet_twsk_add_node(tw, &(ehead + hashinfo->ehash_size)->chain);
	83	atomic_inc(&tw->tw_refcnt);
	84
	85	write_unlock(&ehead->lock);
	86	}
c676270b	87
696ab2d3 ACM	88	EXPORT_SYMBOL_GPL(__inet_twsk_hashdance);
696ab2d3 ACM	89
c676270b ACM	90	struct inet_timewait_sock inet_twsk_alloc(const struct sock sk, const int state)
c676270b ACM	91	{
6d6ee43e ACM	92	struct inet_timewait_sock *tw =
	93	kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
	94	SLAB_ATOMIC);
c676270b ACM	95	if (tw != NULL) {
	96	const struct inet_sock *inet = inet_sk(sk);
	97
	98	/* Give us an identity. */
	99	tw->tw_daddr = inet->daddr;
	100	tw->tw_rcv_saddr = inet->rcv_saddr;
	101	tw->tw_bound_dev_if = sk->sk_bound_dev_if;
	102	tw->tw_num = inet->num;
	103	tw->tw_state = TCP_TIME_WAIT;
	104	tw->tw_substate = state;
	105	tw->tw_sport = inet->sport;
	106	tw->tw_dport = inet->dport;
	107	tw->tw_family = sk->sk_family;
	108	tw->tw_reuse = sk->sk_reuse;
81c3d547	109	tw->tw_hash = sk->sk_hash;
c676270b ACM	110	tw->tw_ipv6only = 0;
	111	tw->tw_prot = sk->sk_prot_creator;
	112	atomic_set(&tw->tw_refcnt, 1);
	113	inet_twsk_dead_node_init(tw);
eeb2b856	114	__module_get(tw->tw_prot->owner);
c676270b ACM	115	}
	116
	117	return tw;
	118	}
696ab2d3 ACM	119
	120	EXPORT_SYMBOL_GPL(inet_twsk_alloc);
	121
	122	/* Returns non-zero if quota exceeded. */
	123	static int inet_twdr_do_twkill_work(struct inet_timewait_death_row *twdr,
	124	const int slot)
	125	{
	126	struct inet_timewait_sock *tw;
	127	struct hlist_node *node;
	128	unsigned int killed;
	129	int ret;
	130
	131	/* NOTE: compare this to previous version where lock
	132	* was released after detaching chain. It was racy,
	133	* because tw buckets are scheduled in not serialized context
	134	* in 2.3 (with netfilter), and with softnet it is common, because
	135	* soft irqs are not sequenced.
	136	*/
	137	killed = 0;
	138	ret = 0;
	139	rescan:
	140	inet_twsk_for_each_inmate(tw, node, &twdr->cells[slot]) {
	141	__inet_twsk_del_dead_node(tw);
	142	spin_unlock(&twdr->death_lock);
	143	__inet_twsk_kill(tw, twdr->hashinfo);
	144	inet_twsk_put(tw);
	145	killed++;
	146	spin_lock(&twdr->death_lock);
	147	if (killed > INET_TWDR_TWKILL_QUOTA) {
	148	ret = 1;
	149	break;
	150	}
	151
	152	/* While we dropped twdr->death_lock, another cpu may have
	153	* killed off the next TW bucket in the list, therefore
	154	* do a fresh re-read of the hlist head node with the
	155	* lock reacquired. We still use the hlist traversal
	156	* macro in order to get the prefetches.
	157	*/
	158	goto rescan;
	159	}
	160
	161	twdr->tw_count -= killed;
	162	NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITED, killed);
	163
	164	return ret;
	165	}
	166
	167	void inet_twdr_hangman(unsigned long data)
	168	{
	169	struct inet_timewait_death_row *twdr;
	170	int unsigned need_timer;
	171
	172	twdr = (struct inet_timewait_death_row *)data;
	173	spin_lock(&twdr->death_lock);
	174
	175	if (twdr->tw_count == 0)
	176	goto out;
	177
	178	need_timer = 0;
	179	if (inet_twdr_do_twkill_work(twdr, twdr->slot)) {
	180	twdr->thread_slots \|= (1 << twdr->slot);
	181	mb();
	182	schedule_work(&twdr->twkill_work);
183	need_timer = 1;
184	} else {
185	/* We purged the entire slot, anything left? */
186	if (twdr->tw_count)
187	need_timer = 1;
188	}
189	twdr->slot = ((twdr->slot + 1) & (INET_TWDR_TWKILL_SLOTS - 1));
190	if (need_timer)
191	mod_timer(&twdr->tw_timer, jiffies + twdr->period);
192	out:
193	spin_unlock(&twdr->death_lock);
194	}
195
196	EXPORT_SYMBOL_GPL(inet_twdr_hangman);
197
198	extern void twkill_slots_invalid(void);
199
65f27f38	200	void inet_twdr_twkill_work(struct work_struct *work)
696ab2d3	201	{
65f27f38 DH	202	struct inet_timewait_death_row *twdr =
65f27f38 DH	203	container_of(work, struct inet_timewait_death_row, twkill_work);
696ab2d3 ACM	204	int i;
	205
	206	if ((INET_TWDR_TWKILL_SLOTS - 1) > (sizeof(twdr->thread_slots) * 8))
	207	twkill_slots_invalid();
	208
	209	while (twdr->thread_slots) {
	210	spin_lock_bh(&twdr->death_lock);
	211	for (i = 0; i < INET_TWDR_TWKILL_SLOTS; i++) {
	212	if (!(twdr->thread_slots & (1 << i)))
	213	continue;
	214
	215	while (inet_twdr_do_twkill_work(twdr, i) != 0) {
	216	if (need_resched()) {
	217	spin_unlock_bh(&twdr->death_lock);
	218	schedule();
	219	spin_lock_bh(&twdr->death_lock);
	220	}
	221	}
	222
	223	twdr->thread_slots &= ~(1 << i);
	224	}
	225	spin_unlock_bh(&twdr->death_lock);
	226	}
	227	}
	228
	229	EXPORT_SYMBOL_GPL(inet_twdr_twkill_work);
	230
	231	/* These are always called from BH context. See callers in
	232	* tcp_input.c to verify this.
	233	*/
	234
	235	/* This is for handling early-kills of TIME_WAIT sockets. */
	236	void inet_twsk_deschedule(struct inet_timewait_sock *tw,
	237	struct inet_timewait_death_row *twdr)
	238	{
	239	spin_lock(&twdr->death_lock);
	240	if (inet_twsk_del_dead_node(tw)) {
	241	inet_twsk_put(tw);
	242	if (--twdr->tw_count == 0)
	243	del_timer(&twdr->tw_timer);
	244	}
	245	spin_unlock(&twdr->death_lock);
	246	__inet_twsk_kill(tw, twdr->hashinfo);
	247	}
	248
	249	EXPORT_SYMBOL(inet_twsk_deschedule);
	250
	251	void inet_twsk_schedule(struct inet_timewait_sock *tw,
	252	struct inet_timewait_death_row *twdr,
	253	const int timeo, const int timewait_len)
	254	{
	255	struct hlist_head *list;
	256	int slot;
	257
	258	/* timeout := RTO * 3.5
	259	*
	260	* 3.5 = 1+2+0.5 to wait for two retransmits.
	261	*
	262	* RATIONALE: if FIN arrived and we entered TIME-WAIT state,
	263	* our ACK acking that FIN can be lost. If N subsequent retransmitted
	264	* FINs (or previous seqments) are lost (probability of such event
	265	* is p^(N+1), where p is probability to lose single packet and
	266	* time to detect the loss is about RTO*(2^N - 1) with exponential
	267	* backoff). Normal timewait length is calculated so, that we
268	* waited at least for one retransmitted FIN (maximal RTO is 120sec).
269	* [ BTW Linux. following BSD, violates this requirement waiting
270	* only for 60sec, we should wait at least for 240 secs.
271	* Well, 240 consumes too much of resources 8)
272	* ]
273	* This interval is not reduced to catch old duplicate and
274	* responces to our wandering segments living for two MSLs.
275	* However, if we use PAWS to detect
276	* old duplicates, we can reduce the interval to bounds required
277	* by RTO, rather than MSL. So, if peer understands PAWS, we
278	* kill tw bucket after 3.5*RTO (it is important that this number
279	* is greater than TS tick!) and detect old duplicates with help
280	* of PAWS.
281	*/
282	slot = (timeo + (1 << INET_TWDR_RECYCLE_TICK) - 1) >> INET_TWDR_RECYCLE_TICK;
283
284	spin_lock(&twdr->death_lock);
285
286	/* Unlink it, if it was scheduled */
287	if (inet_twsk_del_dead_node(tw))
288	twdr->tw_count--;
289	else
290	atomic_inc(&tw->tw_refcnt);
291
292	if (slot >= INET_TWDR_RECYCLE_SLOTS) {
293	/* Schedule to slow timer */
294	if (timeo >= timewait_len) {
295	slot = INET_TWDR_TWKILL_SLOTS - 1;
296	} else {
297	slot = (timeo + twdr->period - 1) / twdr->period;
298	if (slot >= INET_TWDR_TWKILL_SLOTS)
299	slot = INET_TWDR_TWKILL_SLOTS - 1;
300	}
301	tw->tw_ttd = jiffies + timeo;
302	slot = (twdr->slot + slot) & (INET_TWDR_TWKILL_SLOTS - 1);
303	list = &twdr->cells[slot];
304	} else {
305	tw->tw_ttd = jiffies + (slot << INET_TWDR_RECYCLE_TICK);
306
307	if (twdr->twcal_hand < 0) {
308	twdr->twcal_hand = 0;
309	twdr->twcal_jiffie = jiffies;
310	twdr->twcal_timer.expires = twdr->twcal_jiffie +
311	(slot << INET_TWDR_RECYCLE_TICK);
312	add_timer(&twdr->twcal_timer);
313	} else {
314	if (time_after(twdr->twcal_timer.expires,
315	jiffies + (slot << INET_TWDR_RECYCLE_TICK)))
316	mod_timer(&twdr->twcal_timer,
317	jiffies + (slot << INET_TWDR_RECYCLE_TICK));
318	slot = (twdr->twcal_hand + slot) & (INET_TWDR_RECYCLE_SLOTS - 1);
319	}
320	list = &twdr->twcal_row[slot];
321	}
322
323	hlist_add_head(&tw->tw_death_node, list);
324
325	if (twdr->tw_count++ == 0)
326	mod_timer(&twdr->tw_timer, jiffies + twdr->period);
327	spin_unlock(&twdr->death_lock);
328	}
329
330	EXPORT_SYMBOL_GPL(inet_twsk_schedule);
331
332	void inet_twdr_twcal_tick(unsigned long data)
333	{
334	struct inet_timewait_death_row *twdr;
335	int n, slot;
336	unsigned long j;
337	unsigned long now = jiffies;
338	int killed = 0;
339	int adv = 0;
340
341	twdr = (struct inet_timewait_death_row *)data;
342
343	spin_lock(&twdr->death_lock);
344	if (twdr->twcal_hand < 0)
345	goto out;
346
347	slot = twdr->twcal_hand;
348	j = twdr->twcal_jiffie;
349
350	for (n = 0; n < INET_TWDR_RECYCLE_SLOTS; n++) {
351	if (time_before_eq(j, now)) {
352	struct hlist_node node, safe;
353	struct inet_timewait_sock *tw;
354
355	inet_twsk_for_each_inmate_safe(tw, node, safe,
356	&twdr->twcal_row[slot]) {
357	__inet_twsk_del_dead_node(tw);
358	__inet_twsk_kill(tw, twdr->hashinfo);
359	inet_twsk_put(tw);
360	killed++;
361	}
362	} else {
363	if (!adv) {
364	adv = 1;
365	twdr->twcal_jiffie = j;
366	twdr->twcal_hand = slot;
367	}
368
369	if (!hlist_empty(&twdr->twcal_row[slot])) {
370	mod_timer(&twdr->twcal_timer, j);
371	goto out;
372	}
373	}
374	j += 1 << INET_TWDR_RECYCLE_TICK;
375	slot = (slot + 1) & (INET_TWDR_RECYCLE_SLOTS - 1);
376	}
377	twdr->twcal_hand = -1;
378
379	out:
380	if ((twdr->tw_count -= killed) == 0)
381	del_timer(&twdr->tw_timer);
382	NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITKILLED, killed);
383	spin_unlock(&twdr->death_lock);
384	}
385
386	EXPORT_SYMBOL_GPL(inet_twdr_twcal_tick);