inet_timewait_sock.h revision 32aaeffbd4a7457bf2f7448b33b5946ff2a960eb
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
18
19#include <linux/kmemcheck.h>
20#include <linux/list.h>
21#include <linux/timer.h>
22#include <linux/types.h>
23#include <linux/workqueue.h>
24
25#include <net/inet_sock.h>
26#include <net/sock.h>
27#include <net/tcp_states.h>
28#include <net/timewait_sock.h>
29
30#include <linux/atomic.h>
31
32struct 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
61/* TIME_WAIT reaping mechanism. */
62#define INET_TWDR_TWKILL_SLOTS	8 /* Please keep this a power of 2. */
63
64#define INET_TWDR_TWKILL_QUOTA 100
65
66struct inet_timewait_death_row {
67	/* Short-time timewait calendar */
68	int			twcal_hand;
69	unsigned long		twcal_jiffie;
70	struct timer_list	twcal_timer;
71	struct hlist_head	twcal_row[INET_TWDR_RECYCLE_SLOTS];
72
73	spinlock_t		death_lock;
74	int			tw_count;
75	int			period;
76	u32			thread_slots;
77	struct work_struct	twkill_work;
78	struct timer_list	tw_timer;
79	int			slot;
80	struct hlist_head	cells[INET_TWDR_TWKILL_SLOTS];
81	struct inet_hashinfo 	*hashinfo;
82	int			sysctl_tw_recycle;
83	int			sysctl_max_tw_buckets;
84};
85
86extern void inet_twdr_hangman(unsigned long data);
87extern void inet_twdr_twkill_work(struct work_struct *work);
88extern void inet_twdr_twcal_tick(unsigned long data);
89
90struct inet_bind_bucket;
91
92/*
93 * This is a TIME_WAIT sock. It works around the memory consumption
94 * problems of sockets in such a state on heavily loaded servers, but
95 * without violating the protocol specification.
96 */
97struct inet_timewait_sock {
98	/*
99	 * Now struct sock also uses sock_common, so please just
100	 * don't add nothing before this first member (__tw_common) --acme
101	 */
102	struct sock_common	__tw_common;
103#define tw_family		__tw_common.skc_family
104#define tw_state		__tw_common.skc_state
105#define tw_reuse		__tw_common.skc_reuse
106#define tw_bound_dev_if		__tw_common.skc_bound_dev_if
107#define tw_node			__tw_common.skc_nulls_node
108#define tw_bind_node		__tw_common.skc_bind_node
109#define tw_refcnt		__tw_common.skc_refcnt
110#define tw_hash			__tw_common.skc_hash
111#define tw_prot			__tw_common.skc_prot
112#define tw_net			__tw_common.skc_net
113#define tw_daddr        	__tw_common.skc_daddr
114#define tw_rcv_saddr    	__tw_common.skc_rcv_saddr
115	int			tw_timeout;
116	volatile unsigned char	tw_substate;
117	unsigned char		tw_rcv_wscale;
118
119	/* Socket demultiplex comparisons on incoming packets. */
120	/* these three are in inet_sock */
121	__be16			tw_sport;
122	__be16			tw_dport;
123	__u16			tw_num;
124	kmemcheck_bitfield_begin(flags);
125	/* And these are ours. */
126	unsigned int		tw_ipv6only     : 1,
127				tw_transparent  : 1,
128				tw_pad		: 6,	/* 6 bits hole */
129				tw_tos		: 8,
130				tw_ipv6_offset  : 16;
131	kmemcheck_bitfield_end(flags);
132	unsigned long		tw_ttd;
133	struct inet_bind_bucket	*tw_tb;
134	struct hlist_node	tw_death_node;
135};
136#define tw_tclass tw_tos
137
138static inline void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
139				      struct hlist_nulls_head *list)
140{
141	hlist_nulls_add_head_rcu(&tw->tw_node, list);
142}
143
144static inline void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
145					   struct hlist_head *list)
146{
147	hlist_add_head(&tw->tw_bind_node, list);
148}
149
150static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
151{
152	return !hlist_unhashed(&tw->tw_death_node);
153}
154
155static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
156{
157	tw->tw_death_node.pprev = NULL;
158}
159
160static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
161{
162	__hlist_del(&tw->tw_death_node);
163	inet_twsk_dead_node_init(tw);
164}
165
166static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
167{
168	if (inet_twsk_dead_hashed(tw)) {
169		__inet_twsk_del_dead_node(tw);
170		return 1;
171	}
172	return 0;
173}
174
175#define inet_twsk_for_each(tw, node, head) \
176	hlist_nulls_for_each_entry(tw, node, head, tw_node)
177
178#define inet_twsk_for_each_inmate(tw, node, jail) \
179	hlist_for_each_entry(tw, node, jail, tw_death_node)
180
181#define inet_twsk_for_each_inmate_safe(tw, node, safe, jail) \
182	hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
183
184static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
185{
186	return (struct inet_timewait_sock *)sk;
187}
188
189static inline __be32 sk_rcv_saddr(const struct sock *sk)
190{
191/* both inet_sk() and inet_twsk() store rcv_saddr in skc_rcv_saddr */
192	return sk->__sk_common.skc_rcv_saddr;
193}
194
195extern void inet_twsk_put(struct inet_timewait_sock *tw);
196
197extern int inet_twsk_unhash(struct inet_timewait_sock *tw);
198
199extern int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
200				 struct inet_hashinfo *hashinfo);
201
202extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
203						  const int state);
204
205extern void __inet_twsk_hashdance(struct inet_timewait_sock *tw,
206				  struct sock *sk,
207				  struct inet_hashinfo *hashinfo);
208
209extern void inet_twsk_schedule(struct inet_timewait_sock *tw,
210			       struct inet_timewait_death_row *twdr,
211			       const int timeo, const int timewait_len);
212extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
213				 struct inet_timewait_death_row *twdr);
214
215extern void inet_twsk_purge(struct inet_hashinfo *hashinfo,
216			    struct inet_timewait_death_row *twdr, int family);
217
218static inline
219struct net *twsk_net(const struct inet_timewait_sock *twsk)
220{
221#ifdef CONFIG_NET_NS
222	return rcu_dereference_raw(twsk->tw_net); /* protected by locking, */
223						  /* reference counting, */
224						  /* initialization, or RCU. */
225#else
226	return &init_net;
227#endif
228}
229
230static inline
231void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
232{
233#ifdef CONFIG_NET_NS
234	rcu_assign_pointer(twsk->tw_net, net);
235#endif
236}
237#endif	/* _INET_TIMEWAIT_SOCK_ */
238