blob: 4b8ece22b8e94417879a6504e87afe5bd0ecc371 [file] [log] [blame]
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -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 * 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
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070018
19#include <linux/list.h>
Arnaldo Carvalho de Meloeeb2b852005-10-10 21:25:23 -070020#include <linux/module.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070021#include <linux/timer.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070022#include <linux/types.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070023#include <linux/workqueue.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070024
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020025#include <net/inet_sock.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070026#include <net/sock.h>
27#include <net/tcp_states.h>
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -080028#include <net/timewait_sock.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070029
30#include <asm/atomic.h>
31
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070032struct 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;
Eric Dumazet187f5f842007-03-05 13:32:48 -080069 unsigned long twcal_jiffie;
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070070 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
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -070086extern void inet_twdr_hangman(unsigned long data);
David Howells65f27f32006-11-22 14:55:48 +000087extern void inet_twdr_twkill_work(struct work_struct *work);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -070088extern void inet_twdr_twcal_tick(unsigned long data);
89
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070090#if (BITS_PER_LONG == 64)
91#define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 8
92#else
93#define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 4
94#endif
95
96struct inet_bind_bucket;
97
98/*
99 * This is a TIME_WAIT sock. It works around the memory consumption
100 * problems of sockets in such a state on heavily loaded servers, but
101 * without violating the protocol specification.
102 */
103struct inet_timewait_sock {
104 /*
105 * Now struct sock also uses sock_common, so please just
106 * don't add nothing before this first member (__tw_common) --acme
107 */
108 struct sock_common __tw_common;
109#define tw_family __tw_common.skc_family
110#define tw_state __tw_common.skc_state
111#define tw_reuse __tw_common.skc_reuse
112#define tw_bound_dev_if __tw_common.skc_bound_dev_if
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800113#define tw_node __tw_common.skc_nulls_node
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700114#define tw_bind_node __tw_common.skc_bind_node
115#define tw_refcnt __tw_common.skc_refcnt
Eric Dumazet81c3d542005-10-03 14:13:38 -0700116#define tw_hash __tw_common.skc_hash
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700117#define tw_prot __tw_common.skc_prot
Eric W. Biederman07feaeb2007-09-12 11:58:02 +0200118#define tw_net __tw_common.skc_net
Arnaldo Carvalho de Meload8bb782008-02-03 04:08:26 -0800119 int tw_timeout;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700120 volatile unsigned char tw_substate;
121 /* 3 bits hole, try to pack */
122 unsigned char tw_rcv_wscale;
123 /* Socket demultiplex comparisons on incoming packets. */
124 /* these five are in inet_sock */
Al Viro23f33c22006-09-27 18:43:50 -0700125 __be16 tw_sport;
126 __be32 tw_daddr __attribute__((aligned(INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES)));
127 __be32 tw_rcv_saddr;
128 __be16 tw_dport;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700129 __u16 tw_num;
130 /* And these are ours. */
KOVACS Krisztianf5715ae2008-10-01 07:30:02 -0700131 __u8 tw_ipv6only:1,
132 tw_transparent:1;
Arnaldo Carvalho de Melo0fa1a532005-12-13 23:23:09 -0800133 /* 15 bits hole, try to pack */
134 __u16 tw_ipv6_offset;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700135 unsigned long tw_ttd;
136 struct inet_bind_bucket *tw_tb;
137 struct hlist_node tw_death_node;
138};
139
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800140static inline void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
141 struct hlist_nulls_head *list)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700142{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800143 hlist_nulls_add_head_rcu(&tw->tw_node, list);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700144}
145
146static inline void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
147 struct hlist_head *list)
148{
149 hlist_add_head(&tw->tw_bind_node, list);
150}
151
152static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
153{
Akinobu Mitada753be2006-04-28 15:21:23 -0700154 return !hlist_unhashed(&tw->tw_death_node);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700155}
156
157static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
158{
159 tw->tw_death_node.pprev = NULL;
160}
161
162static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
163{
164 __hlist_del(&tw->tw_death_node);
165 inet_twsk_dead_node_init(tw);
166}
167
168static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
169{
170 if (inet_twsk_dead_hashed(tw)) {
171 __inet_twsk_del_dead_node(tw);
172 return 1;
173 }
174 return 0;
175}
176
177#define inet_twsk_for_each(tw, node, head) \
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800178 hlist_nulls_for_each_entry(tw, node, head, tw_node)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700179
180#define inet_twsk_for_each_inmate(tw, node, jail) \
181 hlist_for_each_entry(tw, node, jail, tw_death_node)
182
183#define inet_twsk_for_each_inmate_safe(tw, node, safe, jail) \
184 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
185
186static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
187{
188 return (struct inet_timewait_sock *)sk;
189}
190
Al Viro82103232006-09-27 18:44:10 -0700191static inline __be32 inet_rcv_saddr(const struct sock *sk)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700192{
193 return likely(sk->sk_state != TCP_TIME_WAIT) ?
194 inet_sk(sk)->rcv_saddr : inet_twsk(sk)->tw_rcv_saddr;
195}
196
Pavel Emelyanov7054fb92007-12-20 15:32:54 -0800197extern void inet_twsk_put(struct inet_timewait_sock *tw);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700198
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700199extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
200 const int state);
201
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700202extern void __inet_twsk_hashdance(struct inet_timewait_sock *tw,
203 struct sock *sk,
204 struct inet_hashinfo *hashinfo);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700205
206extern void inet_twsk_schedule(struct inet_timewait_sock *tw,
207 struct inet_timewait_death_row *twdr,
208 const int timeo, const int timewait_len);
209extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
210 struct inet_timewait_death_row *twdr);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900211
Daniel Lezcanod3154922008-09-08 13:17:27 -0700212extern void inet_twsk_purge(struct net *net, struct inet_hashinfo *hashinfo,
213 struct inet_timewait_death_row *twdr, int family);
214
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900215static inline
216struct net *twsk_net(const struct inet_timewait_sock *twsk)
217{
218#ifdef CONFIG_NET_NS
219 return twsk->tw_net;
220#else
221 return &init_net;
222#endif
223}
224
225static inline
Denis V. Lunevf5aa23f2008-03-26 00:48:17 -0700226void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900227{
228#ifdef CONFIG_NET_NS
229 twsk->tw_net = net;
230#endif
231}
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700232#endif /* _INET_TIMEWAIT_SOCK_ */