[deliverable/linux.git] / include / net / inet_timewait_sock.h

/*
 * 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.
 *
 *		Definitions for a generic INET TIMEWAIT sock
 *
 *		From code originally in net/tcp.h
 *
 *		This program is free software; you can redistribute it and/or
 *		modify it under the terms of the GNU General Public License
 *		as published by the Free Software Foundation; either version
 *		2 of the License, or (at your option) any later version.
 */
#ifndef _INET_TIMEWAIT_SOCK_
#define _INET_TIMEWAIT_SOCK_


#include <linux/kmemcheck.h>
#include <linux/list.h>
#include <linux/timer.h>
#include <linux/types.h>
#include <linux/workqueue.h>

#include <net/inet_sock.h>
#include <net/sock.h>
#include <net/tcp_states.h>
#include <net/timewait_sock.h>

#include <linux/atomic.h>

struct inet_hashinfo;

#define INET_TWDR_RECYCLE_SLOTS_LOG	5
#define INET_TWDR_RECYCLE_SLOTS		(1 << INET_TWDR_RECYCLE_SLOTS_LOG)

/*
 * If time > 4sec, it is "slow" path, no recycling is required,
 * so that we select tick to get range about 4 seconds.
 */
#if HZ <= 16 || HZ > 4096
# error Unsupported: HZ <= 16 or HZ > 4096
#elif HZ <= 32
# define INET_TWDR_RECYCLE_TICK (5 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
#elif HZ <= 64
# define INET_TWDR_RECYCLE_TICK (6 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
#elif HZ <= 128
# define INET_TWDR_RECYCLE_TICK (7 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
#elif HZ <= 256
# define INET_TWDR_RECYCLE_TICK (8 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
#elif HZ <= 512
# define INET_TWDR_RECYCLE_TICK (9 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
#elif HZ <= 1024
# define INET_TWDR_RECYCLE_TICK (10 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
#elif HZ <= 2048
# define INET_TWDR_RECYCLE_TICK (11 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
#else
# define INET_TWDR_RECYCLE_TICK (12 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
#endif

static inline u32 inet_tw_time_stamp(void)
{
	return jiffies;
}

/* TIME_WAIT reaping mechanism. */
#define INET_TWDR_TWKILL_SLOTS	8 /* Please keep this a power of 2. */

#define INET_TWDR_TWKILL_QUOTA 100

struct inet_timewait_death_row {
	/* Short-time timewait calendar */
	int			twcal_hand;
	unsigned long		twcal_jiffie;
	struct timer_list	twcal_timer;
	struct hlist_head	twcal_row[INET_TWDR_RECYCLE_SLOTS];

	spinlock_t		death_lock;
	int			tw_count;
	int			period;
	u32			thread_slots;
	struct work_struct	twkill_work;
	struct timer_list	tw_timer;
	int			slot;
	struct hlist_head	cells[INET_TWDR_TWKILL_SLOTS];
	struct inet_hashinfo 	*hashinfo;
	int			sysctl_tw_recycle;
	int			sysctl_max_tw_buckets;
};

void inet_twdr_hangman(unsigned long data);
void inet_twdr_twkill_work(struct work_struct *work);
void inet_twdr_twcal_tick(unsigned long data);

struct inet_bind_bucket;

/*
 * This is a TIME_WAIT sock. It works around the memory consumption
 * problems of sockets in such a state on heavily loaded servers, but
 * without violating the protocol specification.
 */
struct inet_timewait_sock {
	/*
	 * Now struct sock also uses sock_common, so please just
	 * don't add nothing before this first member (__tw_common) --acme
	 */
	struct sock_common	__tw_common;
#define tw_family		__tw_common.skc_family
#define tw_state		__tw_common.skc_state
#define tw_reuse		__tw_common.skc_reuse
#define tw_bound_dev_if		__tw_common.skc_bound_dev_if
#define tw_node			__tw_common.skc_nulls_node
#define tw_bind_node		__tw_common.skc_bind_node
#define tw_refcnt		__tw_common.skc_refcnt
#define tw_hash			__tw_common.skc_hash
#define tw_prot			__tw_common.skc_prot
#define tw_net			__tw_common.skc_net
#define tw_daddr        	__tw_common.skc_daddr
#define tw_rcv_saddr    	__tw_common.skc_rcv_saddr
#define tw_dport		__tw_common.skc_dport
#define tw_num			__tw_common.skc_num

	int			tw_timeout;
	volatile unsigned char	tw_substate;
	unsigned char		tw_rcv_wscale;

	/* Socket demultiplex comparisons on incoming packets. */
	/* these three are in inet_sock */
	__be16			tw_sport;
	kmemcheck_bitfield_begin(flags);
	/* And these are ours. */
	unsigned int		tw_ipv6only     : 1,
				tw_transparent  : 1,
				tw_pad		: 6,	/* 6 bits hole */
				tw_tos		: 8,
				tw_ipv6_offset  : 16;
	kmemcheck_bitfield_end(flags);
	u32			tw_ttd;
	struct inet_bind_bucket	*tw_tb;
	struct hlist_node	tw_death_node;
};
#define tw_tclass tw_tos

static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
{
	return !hlist_unhashed(&tw->tw_death_node);
}

static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
{
	tw->tw_death_node.pprev = NULL;
}

static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
{
	__hlist_del(&tw->tw_death_node);
	inet_twsk_dead_node_init(tw);
}

static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
{
	if (inet_twsk_dead_hashed(tw)) {
		__inet_twsk_del_dead_node(tw);
		return 1;
	}
	return 0;
}

#define inet_twsk_for_each(tw, node, head) \
	hlist_nulls_for_each_entry(tw, node, head, tw_node)

#define inet_twsk_for_each_inmate(tw, jail) \
	hlist_for_each_entry(tw, jail, tw_death_node)

#define inet_twsk_for_each_inmate_safe(tw, safe, jail) \
	hlist_for_each_entry_safe(tw, safe, jail, tw_death_node)

static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
{
	return (struct inet_timewait_sock *)sk;
}

void inet_twsk_free(struct inet_timewait_sock *tw);
void inet_twsk_put(struct inet_timewait_sock *tw);

int inet_twsk_unhash(struct inet_timewait_sock *tw);

int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
			  struct inet_hashinfo *hashinfo);

struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
					   const int state);

void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
			   struct inet_hashinfo *hashinfo);

void inet_twsk_schedule(struct inet_timewait_sock *tw,
			struct inet_timewait_death_row *twdr,
			const int timeo, const int timewait_len);
void inet_twsk_deschedule(struct inet_timewait_sock *tw,
			  struct inet_timewait_death_row *twdr);

void inet_twsk_purge(struct inet_hashinfo *hashinfo,
		     struct inet_timewait_death_row *twdr, int family);

static inline
struct net *twsk_net(const struct inet_timewait_sock *twsk)
{
	return read_pnet(&twsk->tw_net);
}

static inline
void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
{
	write_pnet(&twsk->tw_net, net);
}
#endif	/* _INET_TIMEWAIT_SOCK_ */
Commit	Line	Data
8feaf0c0 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	* Definitions for a generic INET TIMEWAIT sock
	7	*
	8	* From code originally in net/tcp.h
	9	*
	10	* This program is free software; you can redistribute it and/or
	11	* modify it under the terms of the GNU General Public License
	12	* as published by the Free Software Foundation; either version
	13	* 2 of the License, or (at your option) any later version.
	14	*/
	15	#ifndef _INET_TIMEWAIT_SOCK_
	16	#define _INET_TIMEWAIT_SOCK_
	17
8feaf0c0	18
9e337b0f	19	#include <linux/kmemcheck.h>
8feaf0c0	20	#include <linux/list.h>
295ff7ed	21	#include <linux/timer.h>
8feaf0c0	22	#include <linux/types.h>
295ff7ed	23	#include <linux/workqueue.h>
8feaf0c0	24
14c85021	25	#include <net/inet_sock.h>
8feaf0c0 ACM	26	#include <net/sock.h>
8feaf0c0 ACM	27	#include <net/tcp_states.h>
6d6ee43e	28	#include <net/timewait_sock.h>
8feaf0c0	29
60063497	30	#include <linux/atomic.h>
8feaf0c0	31
295ff7ed ACM	32	struct inet_hashinfo;
	33
	34	#define INET_TWDR_RECYCLE_SLOTS_LOG 5
	35	#define INET_TWDR_RECYCLE_SLOTS (1 << INET_TWDR_RECYCLE_SLOTS_LOG)
	36
	37	/*
	38	* If time > 4sec, it is "slow" path, no recycling is required,
	39	* so that we select tick to get range about 4 seconds.
	40	*/
	41	#if HZ <= 16 \|\| HZ > 4096
	42	# error Unsupported: HZ <= 16 or HZ > 4096
	43	#elif HZ <= 32
	44	# define INET_TWDR_RECYCLE_TICK (5 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
	45	#elif HZ <= 64
	46	# define INET_TWDR_RECYCLE_TICK (6 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
	47	#elif HZ <= 128
	48	# define INET_TWDR_RECYCLE_TICK (7 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
	49	#elif HZ <= 256
	50	# define INET_TWDR_RECYCLE_TICK (8 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
	51	#elif HZ <= 512
	52	# define INET_TWDR_RECYCLE_TICK (9 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
	53	#elif HZ <= 1024
	54	# define INET_TWDR_RECYCLE_TICK (10 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
	55	#elif HZ <= 2048
	56	# define INET_TWDR_RECYCLE_TICK (11 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
	57	#else
	58	# define INET_TWDR_RECYCLE_TICK (12 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
	59	#endif
	60
96f817fe ED	61	static inline u32 inet_tw_time_stamp(void)
	62	{
	63	return jiffies;
	64	}
	65
295ff7ed ACM	66	/* TIME_WAIT reaping mechanism. */
	67	#define INET_TWDR_TWKILL_SLOTS 8 /* Please keep this a power of 2. */
	68
	69	#define INET_TWDR_TWKILL_QUOTA 100
	70
	71	struct inet_timewait_death_row {
	72	/* Short-time timewait calendar */
	73	int twcal_hand;
187f5f84	74	unsigned long twcal_jiffie;
295ff7ed ACM	75	struct timer_list twcal_timer;
	76	struct hlist_head twcal_row[INET_TWDR_RECYCLE_SLOTS];
	77
	78	spinlock_t death_lock;
	79	int tw_count;
	80	int period;
	81	u32 thread_slots;
	82	struct work_struct twkill_work;
	83	struct timer_list tw_timer;
	84	int slot;
	85	struct hlist_head cells[INET_TWDR_TWKILL_SLOTS];
	86	struct inet_hashinfo *hashinfo;
	87	int sysctl_tw_recycle;
	88	int sysctl_max_tw_buckets;
	89	};
	90
1fd51155 JP	91	void inet_twdr_hangman(unsigned long data);
	92	void inet_twdr_twkill_work(struct work_struct *work);
	93	void inet_twdr_twcal_tick(unsigned long data);
696ab2d3	94
8feaf0c0 ACM	95	struct inet_bind_bucket;
	96
	97	/*
	98	* This is a TIME_WAIT sock. It works around the memory consumption
	99	* problems of sockets in such a state on heavily loaded servers, but
	100	* without violating the protocol specification.
	101	*/
	102	struct inet_timewait_sock {
	103	/*
	104	* Now struct sock also uses sock_common, so please just
	105	* don't add nothing before this first member (__tw_common) --acme
	106	*/
	107	struct sock_common __tw_common;
	108	#define tw_family __tw_common.skc_family
	109	#define tw_state __tw_common.skc_state
	110	#define tw_reuse __tw_common.skc_reuse
	111	#define tw_bound_dev_if __tw_common.skc_bound_dev_if
3ab5aee7	112	#define tw_node __tw_common.skc_nulls_node
8feaf0c0 ACM	113	#define tw_bind_node __tw_common.skc_bind_node
8feaf0c0 ACM	114	#define tw_refcnt __tw_common.skc_refcnt
81c3d547	115	#define tw_hash __tw_common.skc_hash
8feaf0c0	116	#define tw_prot __tw_common.skc_prot
07feaebf	117	#define tw_net __tw_common.skc_net
68835aba ED	118	#define tw_daddr __tw_common.skc_daddr
68835aba ED	119	#define tw_rcv_saddr __tw_common.skc_rcv_saddr
ce43b03e ED	120	#define tw_dport __tw_common.skc_dport
ce43b03e ED	121	#define tw_num __tw_common.skc_num
ce43b03e	122
ad8bb780	123	int tw_timeout;
8feaf0c0	124	volatile unsigned char tw_substate;
8feaf0c0	125	unsigned char tw_rcv_wscale;
68835aba	126
8feaf0c0	127	/* Socket demultiplex comparisons on incoming packets. */
68835aba	128	/* these three are in inet_sock */
23f33c2d	129	__be16 tw_sport;
9e337b0f	130	kmemcheck_bitfield_begin(flags);
8feaf0c0	131	/* And these are ours. */
abf90cca ED	132	unsigned int tw_ipv6only : 1,
abf90cca ED	133	tw_transparent : 1,
66b13d99 ED	134	tw_pad : 6, /* 6 bits hole */
66b13d99 ED	135	tw_tos : 8,
abf90cca	136	tw_ipv6_offset : 16;
9e337b0f	137	kmemcheck_bitfield_end(flags);
96f817fe	138	u32 tw_ttd;
8feaf0c0 ACM	139	struct inet_bind_bucket *tw_tb;
	140	struct hlist_node tw_death_node;
	141	};
b903d324	142	#define tw_tclass tw_tos
8feaf0c0	143
8feaf0c0 ACM	144	static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
8feaf0c0 ACM	145	{
da753bea	146	return !hlist_unhashed(&tw->tw_death_node);
8feaf0c0 ACM	147	}
	148
	149	static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
	150	{
	151	tw->tw_death_node.pprev = NULL;
	152	}
	153
	154	static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
	155	{
	156	__hlist_del(&tw->tw_death_node);
	157	inet_twsk_dead_node_init(tw);
	158	}
	159
	160	static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
	161	{
	162	if (inet_twsk_dead_hashed(tw)) {
	163	__inet_twsk_del_dead_node(tw);
	164	return 1;
	165	}
	166	return 0;
	167	}
	168
	169	#define inet_twsk_for_each(tw, node, head) \
3ab5aee7	170	hlist_nulls_for_each_entry(tw, node, head, tw_node)
8feaf0c0	171
b67bfe0d SL	172	#define inet_twsk_for_each_inmate(tw, jail) \
b67bfe0d SL	173	hlist_for_each_entry(tw, jail, tw_death_node)
8feaf0c0	174
b67bfe0d SL	175	#define inet_twsk_for_each_inmate_safe(tw, safe, jail) \
b67bfe0d SL	176	hlist_for_each_entry_safe(tw, safe, jail, tw_death_node)
8feaf0c0 ACM	177
	178	static inline struct inet_timewait_sock inet_twsk(const struct sock sk)
	179	{
	180	return (struct inet_timewait_sock *)sk;
	181	}
	182
05dbc7b5	183	void inet_twsk_free(struct inet_timewait_sock *tw);
1fd51155	184	void inet_twsk_put(struct inet_timewait_sock *tw);
e48c414e	185
1fd51155	186	int inet_twsk_unhash(struct inet_timewait_sock *tw);
13475a30	187
1fd51155 JP	188	int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
1fd51155 JP	189	struct inet_hashinfo *hashinfo);
3cdaedae	190
1fd51155 JP	191	struct inet_timewait_sock inet_twsk_alloc(const struct sock sk,
1fd51155 JP	192	const int state);
c676270b	193
1fd51155 JP	194	void __inet_twsk_hashdance(struct inet_timewait_sock tw, struct sock sk,
1fd51155 JP	195	struct inet_hashinfo *hashinfo);
696ab2d3	196
1fd51155 JP	197	void inet_twsk_schedule(struct inet_timewait_sock *tw,
	198	struct inet_timewait_death_row *twdr,
	199	const int timeo, const int timewait_len);
	200	void inet_twsk_deschedule(struct inet_timewait_sock *tw,
	201	struct inet_timewait_death_row *twdr);
3b1e0a65	202
1fd51155 JP	203	void inet_twsk_purge(struct inet_hashinfo *hashinfo,
1fd51155 JP	204	struct inet_timewait_death_row *twdr, int family);
d315492b	205
3b1e0a65 YH	206	static inline
	207	struct net twsk_net(const struct inet_timewait_sock twsk)
	208	{
f943cbe6	209	return read_pnet(&twsk->tw_net);
3b1e0a65 YH	210	}
	211
	212	static inline
f5aa23fd	213	void twsk_net_set(struct inet_timewait_sock twsk, struct net net)
3b1e0a65	214	{
f943cbe6	215	write_pnet(&twsk->tw_net, net);
3b1e0a65	216	}
8feaf0c0	217	#endif /* _INET_TIMEWAIT_SOCK_ */